<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Adam Gospodarczyk on Medium]]></title>
        <description><![CDATA[Stories by Adam Gospodarczyk on Medium]]></description>
        <link>https://medium.com/@overment?source=rss-30f3c12d2009------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*8taeH0HukYZEjs4d0RFYZA.jpeg</url>
            <title>Stories by Adam Gospodarczyk on Medium</title>
            <link>https://medium.com/@overment?source=rss-30f3c12d2009------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Tue, 17 Nov 2026 03:02:15 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@overment/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com" rel="hub"/>
        <item>
            <title><![CDATA[My Day on Autopilot]]></title>
            <link>https://medium.com/@overment/my-day-on-autopilot-23f528a1c195?source=rss-30f3c12d2009------2</link>
            <guid isPermaLink="false">https://medium.com/p/23f528a1c195</guid>
            <category><![CDATA[apple]]></category>
            <category><![CDATA[openai]]></category>
            <category><![CDATA[productivity]]></category>
            <category><![CDATA[automation]]></category>
            <dc:creator><![CDATA[Adam Gospodarczyk]]></dc:creator>
            <pubDate>Sat, 05 Aug 2023 12:10:17 GMT</pubDate>
            <atom:updated>2023-08-05T12:10:17.001Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*4laRl5fxteFe7yAvC8NXUg.jpeg" /></figure><blockquote>The following article is a free chapter from my e-book “Everyday on Autopilot available at <a href="https://everyday.techsistence.com/">https://everyday.techsistence.com</a></blockquote><p>You already know the tools and work techniques I use in my everyday life. However, I thought it would be a good idea to show you how it all works for me in practice. So, I invite you to the chapter where you will see what my “Everyday on Autopilot” looks like.</p><p>Over the past week, I have been collecting various notes that allowed me to observe my behaviors, habits, and certain details that accompany me in my work. Some of them are directly related to automation, while others only to some extent. I have chosen the ones that I think you might find the most interesting.</p><p>Before we begin, however, I would like to emphasize that <strong>this is my everyday life</strong> and not all the threads presented here may be useful to you. On the other hand, even in them, you may find inspiration that will give you ideas for designing solutions that suit you. So, prepare yourself a coffee or a good tea and sit comfortably. In the meantime, I will make sure that this chapter of the e-book is interesting for you.</p><h3>My own context</h3><p>I would like to remind you that information about the equipment and applications I use can always be found on the <a href="http://brain.overment.com">brain.overment.com</a> website, so I won’t list them all here.</p><p>Besides that, you need to know that my work requires switching between many areas of expertise. That’s why I use different tools, change my working style and mindset, and adjust my environment to support what I’m currently doing. Please take this into account because when you do, it will become more understandable to you why I pay so much attention to small details. Everything you are about to see is fully customized to me, my personality, the type of tasks I perform, and even my current energy level. At the same time, it’s not something particularly complex that requires my attention all the time. On the other hand, nothing here is random, and I often <strong>find myself modifying various elements to effectively balance on the edge of my current competence.</strong>### My Principles</p><p>Here are some selected principles (there are more) that will help you better understand my lifestyle.</p><ol><li>Sleep is a Priority</li><li>I manage energy, not time</li><li>I have ONE task</li><li>I follow curiosity</li><li>I have (almost) an empty calendar</li><li>I don’t have a to-do list longer than 1 day</li><li>I check email 1–2 times a day</li><li>I operate in DND mode for most of the day</li><li>When I don’t have energy, I don’t work</li><li>“Discipline brings freedom”</li></ol><p>Some of the points above are obvious, some are not. So let me briefly explain each of them.</p><ol><li>I sleep for 7 hours, the time I go to bed and wake up doesn’t matter much. The key here is “7 hours”. I don’t bend this rule because everything else depends on it.</li><li>I don’t pay attention to the time except for alerts that remind me to eat or finish work. Instead, I adjust my daily schedule based on my current energy level. That’s why I don’t use time blocks or a specific order for tasks. I adapt dynamically to the current situation.</li><li>Every day, I choose only ONE task that I must accomplish. Naturally, I do much more, but that one task has to be done. It’s my “daily victory” that makes a significant difference.</li><li>If I have to choose between what is important and what interests me, I always go for the latter. This is crucial because it allows me to accomplish tasks at an unimaginable level of focus. What is important will ultimately get done anyway, but curiosity and excitement always push me to break boundaries.</li><li>And if a meeting is necessary, I come prepared, immediately get to the point, and wrap up the topic in 20–30 minutes. During the meeting, I create a summary note that I send after it ends.</li><li>I don’t see a reason.</li><li>Same here. I check emails once a day or less frequently. The exception is days when I address important threads or run sales campaigns.</li><li>DND mode (do not disturb) turns off all my notifications. Only two people can contact me during that time, but they know to do it only when necessary.</li><li>Sometimes there are such days. In some cases, you can find energy in various ways, and in others, you just need to take it easy.</li><li>My loved ones claim that I am the most disciplined person they know. I don’t consider myself as such, but I unwaveringly follow through with what I say. I constantly push my own limits and “get back” on the right path of achieving my “process”.</li></ol><p>As I mentioned, there are definitely more principles, but these play the biggest role in my daily life, so I decided to mention them. As you can see, some of them are quite unique (curiosity &gt; importance), and others are hard to believe (no plans, goals, personal to-do list longer than a week [there are product roadmaps, and some tasks are assigned to me]).</p><p>Similarly, implementing some of these points was quite problematic, or even initially negatively received by some people. However, when results appear, it suddenly becomes clear that it’s not the same. Nevertheless, some people attribute them to “innate genius” or other external sources. Well, it’s hard to admit that by breaking certain patterns, you can achieve different results.</p><h3>Tools I Work With</h3><p>Although I have already mentioned the list of apps and tools I use, I still wanted to highlight a few of them.</p><p>The thing is, even if I showed you how to use Raycast, there are still some details that are specific to me. Sometimes you might get the impression that some apps I use “start dancing” only when we connect them with other apps or apply techniques that are not always obvious.</p><p>Here again, I have often heard that I have a talent for using commonly available tools in an unconventional way, and even observing how I use the iPad can be enlightening and inspiring (these are not my words).</p><p>The tools I want to focus on are:</p><ul><li>Arc (browser)</li><li>Raycast</li><li>Keyboard Maestro</li><li>Shortcuts</li><li>iA Writer</li><li>Better Touch Tool</li><li>Notion</li><li>Airtable</li><li>Make</li><li>Webflow</li></ul><p>Everything on this list significantly affects how I function. For example, the Arc browser has changed the way I use the Internet. On the other hand, iA Writer has greatly contributed to the fact that in the past 6 months, I have written about 3 million characters of text on my own.</p><p>However, I think it would be best if I take you through what my day looks like.</p><h3>The beginning of my day</h3><p>My day usually starts around 6:00 with a ~35-minute walk with Alexa, my dog. As I pass by the room where I work, you can hear quiet music coming from it (Homepod 2) and see the lights on (Phillips Hue). And even though the atmosphere in there invites me to get to work, it will have to wait because the first 2 hours of the day are for me.</p><p>Three times a week, I follow the <a href="https://www.reddit.com/r/bodyweightfitness/wiki/kb/recommended_routine/">recommended routine</a> and exercise with the Pliability app, which I display on the TV screen using Chromecast.</p><p>During walks, I usually listen to audiobooks (Airpods Pro 2 &amp; Audible). Sometimes I accompany it with music, and other times I prefer silence. The choice depends on my current energy level and what awaits me that day. If you remember what my phone screen looks like, you’ll understand that the limited number of icons there is not accidental. I just really want to do as little as possible on my phone, other than what Alice, Audible, Spotify, and Kindle offer. There are other apps on my phone, but almost all of them remain hidden, reducing the chance that I will spend time with them.</p><p>After returning from the walk, I usually sit down with coffee, a book, and music from Spotify or <a href="http://endel.io">endel.io</a>. If you’re wondering how I manage to <a href="https://www.goodreads.com/user/show/91387486-adam-gospodarczyk">read quite a lot</a>, it’s largely thanks to the morning routine. ~90 minutes quickly add up, and a few/few dozen pages suddenly turn into whole chapters and more books.</p><h3>My office and immediate surroundings</h3><p>When I finish my morning routine and start working, I’m always greeted by the same view:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*S9sQFw0KKgVAWipIHYjIJQ.png" /><figcaption>Hardware I use — <a href="https://brain.overment.com/Tools/Hardware.html">https://brain.overment.com/Tools/Hardware.html</a></figcaption></figure><p>Pay attention to some details in the above photo:</p><ul><li>There is only a keyboard and trackpad on the desk. A coffee mug appears there twice a day.</li><li>I work on one large 32-inch screen.</li><li>I have the option to work standing up.</li><li>My microphone and camera are set up for immediate use. When I’m not using them, their presence doesn’t bother me at all.</li></ul><p>In addition, there are a few small things, such as the fact that when I unlock my computer, all applications <strong>are closed</strong>. This is thanks to a simple Keyboard Maestro macro:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*hUuMB5aml-uoHl1bA_Niag.png" /><figcaption>Automatically closing and launching apps every day</figcaption></figure><p>This is a psychological trick that serves two purposes: everything I did the previous day must be completed, and a new day starts with a “clean slate.”</p><p>Combining this with the fact that I have no icons on my desktop brings additional peace and space to make decisions about “what should I focus on now?”</p><h3>Email overview and inbox zero</h3><p>The first few minutes of work are usually spent reviewing if anything particularly deserves my attention that happened overnight. For example, this morning my inbox looked like this:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*C2YLLOSJzreINuFQbOpJrg.png" /><figcaption>Superhuman.com &amp; inbox zero</figcaption></figure><p>It’s very rare for any emails to land in my inbox. However, that doesn’t mean I don’t receive any emails; it means they are automatically sorted. The most important ones go to the <strong>VIP</strong> tab, and those that require action from me are placed in “Action Required.” The best newsletters I read are redirected to “Brain Food.”</p><p>I should also mention that I have several mailboxes associated with different projects I’m working on. The most important messages automatically go to my Notion dashboard.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*d8yEF1dOJt6WKF6tLcaHrw.png" /><figcaption>Second Brain in Notion powered with Automations &amp; AI</figcaption></figure><p>Currently, these are messages from easy_, but you can only imagine how much time I would need to check eduweb, easy_, ahoy, Alice, and overment mailboxes. Instead, I have time to write this e-book and don’t have to worry about anything.</p><p>If you’re wondering how the above messages are marked, it’s done with the help of automatic filters that we discussed when talking about email automation.</p><h3>Meeting organization and calendar</h3><p>In addition to email, there’s also a calendar that I rarely use, but even so, the process is highly optimized for me. As I mentioned, scheduling a meeting with me is possible with the help of the <a href="http://zencal.io">zencal.io</a> link, which I can share by typing ;cal30.</p><p>I manage events in my calendar using the Cron app:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*YEGCwZO9VX0YQCwREo5JtQ.png" /><figcaption>Cron.com</figcaption></figure><p>As you can see, there aren’t many events here, and that’s pretty much how it looks for most of the year. At the same time, Cron is a brilliant app that takes inspiration from Superhuman in many ways. If you’re not convinced by its aesthetic appearance, try installing it and using the command palette (⌘K), and create a new event with the Meet With command. I can’t imagine anyone wanting to create events in any other way afterwards.</p><p>Notice that despite the fact that it’s easy to schedule a meeting with me, my calendar is empty. It’s not that I have to avoid meetings, but rather that my work style simply doesn’t require them. Even in situations where someone needs something more substantial from me, they will simply receive an email that concisely describes the most important points. The principle of “I respect your time, and I ask for the same in return” works well here. Of course, its practical application didn’t happen overnight.</p><h3>The role of music in deep work</h3><p>Meanwhile, I forgot about the most important thing! I work most of the day with headphones that have Active Noise Cancelling (ANC) technology, which blocks out most of the distracting sounds for me. When I particularly need to focus, I use the <a href="http://endel.io">endel.io</a> app in Deeper Focus mode.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*DZ_uYu9R6jpt5yqc1wWYBA.png" /><figcaption>Endel.io</figcaption></figure><p>In other situations, I listen to Spotify for most of the day. Alice starts playing my favorite music for me with the following prompt:</p><pre>Ready to embark on a musical journey? I can help you discover a specific track or artist to match your vibe. I&#39;ll present my recommendation in a structured format, like this:</pre><pre>{&quot;query&quot;: &quot;The perfect Spotify track or playlist based on your needs&quot;} Just remember to be as precise as you can about your current mood or activity, and even the genre or era of music you&#39;d prefer. For instance:</pre><pre>Q: &quot;I&#39;m in the mood for some gritty rock music.&quot;<br>A: {&quot;query&quot;: &quot;Led Zeppelin - Whole Lotta Love&quot;}</pre><pre>Q: &quot;I need some beats to power through my workout.&quot;<br>A: {&quot;query&quot;: &quot;Eminem - Till I Collapse&quot;}</pre><pre>Q: &quot;Can you play something from the Beatles?&quot;<br>A: {&quot;query&quot;: &quot;The Beatles - Here Comes The Sun&quot;}</pre><pre>So, Adam, what are you in the mood for today?</pre><p>The query generated by Alice is sent to the following scenario. The Spotify module searches the database and only considers the first result, which it then sends back to Alice.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*KjBGviWsK_4AMEVYL5R8qA.png" /><figcaption>make.com &amp; spotify integration</figcaption></figure><p>But something else happens in this scenario. I want to actually play the found music. For this purpose, I use the x-scheme-url that we’ve already talked about. Its value is passed to the Spotify macro, which uses the “Remote Web Trigger” option. It’s a special trigger in the form of a URL (similar to a webhook) that I can refer to from <a href="http://make.com">make.com</a> using the HTTP module.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*VTPcsJnQtyzyL4drM5xCRg.png" /><figcaption>Play Music with Keyboard Maestro Remote Web Trigger</figcaption></figure><p>🔗 <a href="https://cloud.overment.com/Play-Music-1690533608.kmmacros">Download Keyboard Maestro macro</a></p><p>🔗 <a href="https://cloud.overment.com/play-music-1690533664.json">Download make.com blueprint</a></p><p>Fun fact: I’m also developing a very advanced version of Alice that is currently not available to others, and it’s hard for me to predict if it will ever be. However, I suspect that the technology offered by OpenAI will continue to evolve. It’s possible that in the future, the integration with GPT-4 will include the ability to remember information about you. This fact can be used in various ways, even for simple things like asking to play your <strong>favorite music</strong>!</p><p>As you can see below, Alice knows exactly what I mean.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*DTBPxDmjN_SoLB1rfevJ2A.png" /></figure><p>The last macro related to music is the ability to easily pause and resume what I’m currently listening to in my headphones. This is done by an incredibly simple macro that runs an AppleScript script to control Spotify’s behavior.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*xnBimHs3kraZechlDat-tQ.png" /></figure><p>🔗 <a href="https://cloud.overment.com/Playpause-1690534204.kmmacros">Download the macro</a></p><p>The above macro really makes my life easier when working at the computer. The fact that I’ve run it over 11,000 times should be proof enough. However, I can’t agree with the information about the time saved. Maybe the execution of this action does indeed translate into 33 hours saved. But considering how many times I’ve avoided distractions by running this macro, I think the result should be in the triple digits.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*hV54a9UqdJVBD4OHShBUmg.png" /><figcaption>Simple Keyboard Maestro macro for playpause Spotify</figcaption></figure><h3>Writing with the help of automation and AI</h3><p>Writing is one of the most important activities for me, so I make sure that the process is optimized for focus and simply enjoyable.</p><p>I’ve already mentioned that my main text editor is the iA Writer app. Its main advantage is its super minimalist interface, consisting solely of a blinking cursor on the screen. Additionally, iA Writer works with plain .txt files and Markdown syntax, which has become popular in recent years and can now even be found in Google Docs.</p><p>If you’re not familiar with Markdown, here’s an example article describing how to write text using it: <a href="https://itnext.io/github-markdown-cheatsheet-50642835effa">Markdown Cheatsheet</a></p><p>Currently, I always have Alice by my side when writing texts. Snippets responsible for translation or proofreading are available to me with a single keyboard shortcut. Alice <strong>doesn’t generate text for me</strong>, but she can help me in certain situations, increasing the comfort of writing (e.g., when I need a synonym or a better definition of a term).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*clJzbCJVCOaoVvwh52mDHA.png" /><figcaption>Alice AI as a companion for writing</figcaption></figure><p>In addition, I use various macros that make my work with iA Writer easier. One of them is for quickly opening a new document or returning to the one I last edited.</p><p>🔗 <a href="https://cloud.overment.com/iA-Writer-1690537718.kmmacros">Download the macro</a></p><p>Configuring the above macros requires changing the username in the script responsible for downloading the latest file. Simply replace the phrase “overment” with your profile name.</p><p>There is also another, very advanced macro that uses the <a href="https://pandoc.org/">pandoc</a> tool (which you can install using <a href="http://brew.sh">brew.sh</a>). Its task is to <strong>convert Markdown syntax to rich text</strong>. In other words, thanks to it, I can easily adapt the text written in iA Writer to text ready to be pasted, for example, into Superhuman.</p><p>🔗 <a href="https://cloud.overment.com/Styled-Text-1690538093.kmmacros">Download the macro</a></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*0hQIoRVtt2JUNrRZ_2tDFw.png" /><figcaption>Markdown to Styled Text</figcaption></figure><p>Thanks to the transformation from Markdown to styled text (or HTML), I can freely write in my favorite editor and then transfer the text to the intended destination.</p><h3>Graphic design with automation support</h3><p>I have already shown that Alice is able to generate graphics for me based on a pre-defined template. Creating designs is based on Webflow and my own solution that converts a website into an image.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*A_AqlXDp0F_HyHXGTIK6Dg.png" /><figcaption>Webflow as … graphic generator</figcaption></figure><p>The above template is used by the following scenario. The automation designed in this way is triggered by various events, either directly by me or during a conversation with Alice. In the latter case, it involves a prompt that generates a specific JSON object and sends it to the appropriate scenario. Generating graphics in this way requires detailed planning of the template and preparing it in a way that allows it to adapt flexibly to the content. This has its advantage because creating designs does not require opening Figma or Webflow. However, the disadvantage is that introducing any changes is very complex and it is better to avoid it.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*05zyoGnmQmEB9NbsEZBZ6w.png" /><figcaption>Make.com and Image Generation with Webflow and my own service which may be replaced with <a href="https://htmlcsstoimage.com/">https://htmlcsstoimage.com</a></figcaption></figure><p>However, there is no need to use such advanced tools to make design easier. Currently, Figma and its features that allow creating components based on auto-layout enable efficient creation of subsequent designs, even considering the diversity of formats.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*HIwZ3-us26BTep-ziDb1oA.png" /><figcaption>Figma Components &amp; Auto-Layout rocks!</figcaption></figure><p>Undoubtedly, Generative AI and tools like Midjourney or Adobe Firefly play an important role in designing graphics. Together with Michał, who has been creating almost all the designs for our projects for years, we have developed a relatively simple prompt for generating graphics that match our brand’s style.</p><p>Generating graphics in combination with the mentioned components and other features of Webflow or Figma allows us to relatively easily create subsequent designs for our initiatives.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*eAQLxKQFyxrUCiuXSbVHbg.png" /><figcaption>Midjourney and a consistent style</figcaption></figure><p>In summary, the latest tools (Figma) and automation and AI support us in design. Combining our previous experience with new technology yields quite interesting results.</p><h3>Pitstop! The role of rest and regeneration supported by robots</h3><p>Around 12:00, I take a break. It’s a good time for eating, reading, and possibly going for a walk. Usually, I experience a significant drop in energy at that time, and it’s a good time for less demanding tasks.</p><p>Despite the fact that the projects I work on are already operating on a fairly large scale, I still handle many activities on my own, including practically all processes: from marketing and sales to implementation. For example, when chatting with easy_, there’s a good chance you’ll meet me there, and what’s more, I’m often able to help immediately, even with complex technical queries.</p><h3>Low energy mode and its recovery</h3><p>I usually don’t keep a typical to-do list, or at least I don’t write down everything I do. I also don’t spend too much time on regular planning of my activities. However, when my energy starts to drop, it’s the perfect time to do less demanding activities, such as:</p><ul><li>checking emails and comments</li><li>browsing social media and private messages</li><li>reviewing notes I’ve gathered in Notion</li></ul><p>Usually, such a session lasts from 15 minutes to 3 hours, depending on the tasks I have to do. It’s the perfect time for me to relax, listen to music, or browse through lighter materials and publications.</p><p>If I don’t have specific tasks to do, I don’t force myself to find something to do, but I reach for Feedly and browse through suggested articles. I also share some of them on social media (e.g., on Instagram).</p><p>During the workday, regular breaks are important to me, and I try not to rest only when my energy level is low. For example, when going for an afternoon walk with Alexa, I want to have enough energy to enjoy it together with her.</p><p>Alice reminds me about regular breaks. The following macro runs every 45 minutes and makes Alice inform me in her own words that she has stepped away from the computer. For the voice notification, Alice’s voice is generated using the <a href="https://elevenlabs.io/">elevenlabs.io</a> service.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*erbvPchM9TxNszAcaL-5Lw.png" /><figcaption>Voice Notifications with ElevenLabs</figcaption></figure><p>Similar notifications are about upcoming meetings and important events that I should react to.</p><h3>Context Switching &amp; Automation</h3><p>Working in focus doesn’t necessarily exclude the possibility of changing context and switching between different applications. For example, while writing this e-book, I constantly switch to Arc or other programs to take screenshots. Despite this, Rize indicates that 57% of my work time is spent in focus. Of course, in my case, context switching carries the risk of distraction, but macros help me practically “teleport” to where I need to be. There’s nothing groundbreaking about it, as I’m talking about very simple macros that open websites, launch specific programs, or provide useful information.</p><p>The way I organize the information I work with also plays a big role here. I’m referring to Airtable and Notion databases, where <strong>everything I publish</strong> goes. Now, with the help of Alice, the recorded knowledge is properly organized, and I have easy access to it. Unfortunately, the following feature is not available in the public version, and I don’t have simple instructions to help you translate it into your everyday life. However, it’s possible that soon there will be tools that allow you to connect, for example, Airtable to GPT-4.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ZqK29L2kNL-7C5H6AFI2-w.png" /><figcaption>Long-term memory in a private version of Alice</figcaption></figure><p>The fact that Alice is able to easily and very accurately retrieve practically any information from my knowledge base combines the capabilities of GPT-4, so-called vector databases, search mechanisms, and a clear database structure.</p><p>For over 4 years, I’ve stored almost all information in Airtable and Notion. The funny thing is that I rarely visit these databases because robots manage them. Whenever I need something, Alice is able to easily provide it to me. In addition, there are a number of automations within these databases that, for example, allow me to change the thumbnails of YouTube videos without accessing the settings of my channel! One click is enough to trigger an automation that replaces the covers of all videos, generating them based on templates I’ve created.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*m9LTSIXpyzv5FKSO8i2AAg.png" /><figcaption>Airtable as a single-source of truth for all of my projects</figcaption></figure><p>If you think that all of this required a lot of work, I don’t perceive it that way in practice. For me, it’s purely the result of consistency and a fair amount of discipline that have translated into these results over the years. Everything I’ve shown you so far is not the result of my actions in the past few weeks, but of many years. Despite this, my list of ideas for adapting my environment is still very long, and new entries are constantly being added to it.</p><p>My point here is that even if you’re not naturally an organized person, by using sensible tools and developing a few techniques, you can design systems that will work for you for years. The benefits that result from this can also extend to other people you collaborate with. However, I must emphasize the fact that for people who don’t understand the topic of automation, using all these tools can be very frustrating. Having at least some knowledge of how Airtable, Notion, Make, or GPT-4 work is absolutely essential.</p><h3>Effective Communication with Robots</h3><p>I think one of the most important factors influencing how I manage my time is how I communicate with others and why I do it.</p><p>By default, I avoid meetings in favor of asynchronous communication. However, it doesn’t mean that I never chat on Slack or Zoom. The calendar I showed you earlier suggests that such situations are not frequent. Moreover, when I do have a meeting, I am always prepared and get straight to the point, skipping unnecessary small talk.</p><p>By organizing my time in this way, I always have space for physical meetings, which usually have a purely social character. I pay a lot of attention to building long-term relationships with others. I feel that people from my closest circle value the time spent with me much more than “small talk” at the beginning of a business conversation. In my opinion, stretching meetings during working hours has a severely negative impact on our efficiency and, consequently, our well-being. After all, no one likes the situation when, after a whole day of meetings, the to-do list remains untouched and approaching deadlines make it difficult to have a peaceful sleep.</p><p>The same goes for emails. Instead of exchanging dozens of lengthy messages, I only write when necessary. Equally important, I don’t expect an immediate response from the other person. It is important for me to organize my work in a way that doesn’t require getting an answer “here and now”. Even if I have to wait 1–3 days to get the information I’m interested in, it’s okay for me. Of course, there are critical situations where I can’t handle something on my own and need immediate help. However, I can’t remember the last time something like that happened.</p><p>And now, let’s move on from the topic of conversations with others. I think it’s worth mentioning public discussions, in which I personally almost never engage. The goal of most of these discussions is to prove oneself right, but I’m only interested in getting to the truth. In other words, I value realizing that I’m wrong more than “winning” the discussion.</p><p>Moreover, engaging in public discussions, for example in comments, takes up a lot of energy and time. This can be clearly seen in cafes, where people trying to work on their laptops spend time chatting and constantly reaching for their phones. Incorporating such activities into daily duties is usually a bad idea, which is why I limit them to a minimum.</p><h3>Acquiring Knowledge “on Autopilot”</h3><p>In addition to reading, I often watch videos on YouTube, video courses, and read articles that are valuable to me. I usually do this directly on my computer or iPad. However, during the learning process, I practically don’t take any notes or write anything down. Honestly, <strong>I don’t even try to remember anything, but I constantly strive to understand what I’m dealing with at the moment</strong>. However, I realize that the process of effectively acquiring knowledge may differ for each of us. I think the best strategy in this area is to develop our own system of acquiring and memorizing knowledge that is simply tailored to us.</p><p>Despite my approach to learning, I still have a few automations and macros that help me. Specifically, I’m talking about:</p><ul><li>the ability to save the currently viewed page (or notes from it) and save it in my database (which Alice also reads) and in Feedly</li><li>the ability to add a quick note from anywhere on the computer or even from my phone (voice note)</li><li>the ability to easily save a quote that will go to my Goodreads profile and my knowledge base</li></ul><p>Saving information is done through macros and integration with OpenAI. In addition to the content itself, Alice also takes care of enriching the saved information with additional facts. For example, this is what a note created based on a selected text fragment on a website looks like.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*mZ2gPjNeuADbVvzYAlrNBA.png" /><figcaption>Quick notes from everywhere are automatically saved in Notion</figcaption></figure><p>One keyboard shortcut is enough to run a macro that:</p><ul><li>Copies the selected text</li><li>Retrieves the URL from the browser window</li><li>Sends the note to Alice</li><li>Runs a scenario to add an entry to Notion and Airtable</li><li>Sends me a notification confirming the completed task</li></ul><p>In the above list, the most important thing is to copy the note content along with the URL. This is done by a very simple macro that uses the keyboard shortcut ⇧⌘C, which in the Arc browser is responsible for copying the address. The remaining automation is a matter of the <a href="http://make.com">make.com</a> scenario, which saves the submitted note in the specified databases.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*eyqmZXn-vaUyI0qnlSTtQw.png" /></figure><h3>My free time</h3><p>I usually work until around 6:00 PM. The average working time during the day is about 8.5 to 9.5 hours. Of course, there are times when I work longer or don’t work at all. However, I thought you might be interested in what activities fill my free time and how the latest technology helps me with that.</p><h3>Video games and streaming</h3><p>Evenings are often spent with my PlayStation or Steam Deck. Sometimes I also take advantage of the opportunity to play “in the cloud” through services like Nvidia GeForce Now, PS Plus, or Xbox Cloud Gaming. Spending time with games is important to me because it allows me to relax deeply.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ABvOfA3y3ByxPfiDtOXlgg.png" /><figcaption>Steam Deck</figcaption></figure><p>In addition, staying up to date with the latest technologies (streaming, VR, AR) is a great way for me to broaden my interests. It often turns out that the gaming industry has a direct impact on other areas of technology. This includes hardware, interface design, artistic aspects, as well as shaping behaviors and habits.</p><h3>Reading books and drawing</h3><p>I also spend my free time with my iPad and apps like Kindle or ProCreate. Combining them with music filling my home or a view of the mountains is the best form of relaxation for me. Unlike previous graphic tablets I had, with the iPad, I can take it anywhere, and the device itself can last for many hours without needing to be charged.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*9WlEMFcy8Kj5vvCzjlWdDQ.png" /></figure><p>The iPad also works very well for reading books. In my case, just like on my phone, I use the Kindle app. I also have a <a href="https://paperlike.com/">PaperLike</a> cover, which, in my opinion, has a positive impact on the displayed image and the screen surface.</p><p>An interesting tip is to use a black and white screen and Shortcuts macros, which are also available on the iPad, as long as they don’t contain actions specific to iOS or macOS.</p><h3>Spending time with loved ones</h3><p>The last aspect I would like to address in the context of “optimizing” free time is <strong>not optimizing it</strong>, or at least not in a literal sense. However, I have a few simple rules that help me make the time spent with my loved ones more enjoyable.</p><ul><li><strong>DND mode</strong>: after 7:30 PM, Siri Shortcuts activates the Do Not Disturb mode, blocking all sounds and notifications.</li><li><strong>Phone out of reach</strong>: During work, I usually leave my phone in the living room. When I start my free time, the phone goes to my office.</li><li><strong>Critical notifications</strong>: Alice has permission to send me messages on a separate messenger, which is added to the exceptions of the Do Not Disturb mode. If something critically important happens, Alice has the right to inform me. The same applies to Grzegorz and Katarzyna.</li></ul><p>Whether I spend time with someone alone or by myself, I always make sure that various distractions don’t get in my way. I also avoid browsing social media and instead of chatting, I prefer face-to-face meetings (if they are physically possible).</p><p>Sometimes I get asked if I ever rest at all. The question is valid because it’s not something you can see on my social media profiles. The reason for this is that when I spend time with loved ones, I want those moments to remain between me and the people I’m with at that time. I think that’s okay.</p><h3>Ending the day</h3><p>My day definitely ends around 11:30 PM, although it’s not a fixed time. What matters more to me is getting 7 hours of sleep rather than going to bed at the same time every night.</p><p>There are a few automations that help me prepare for sleep. Some of them send me notifications that remind me of various activities.</p><p>First and foremost, I have scheduled screen transitions (using the Flux app or the system function) and lights to switch to evening mode. This means applying blue light filters to my screens and changing the color temperature of the lights.</p><p>Another thing is that my phone sends me a notification when the end-of-day scenarios are triggered. The fact that I receive this notification (despite being in Do Not Disturb mode) serves as a clear signal that it’s time to sleep. It may sound absurd, but if you recall interesting discussions with friends or an engaging movie, you’ll understand that it’s not obvious.</p><p>And finally, there are a few things happening in the background that I’m not even notified about. I’m referring to:</p><ul><li>“cleaning up” my dashboard</li><li>preparing daily summaries</li><li>closing open applications</li><li>turning off lights</li><li>switching the air purifier to night mode</li><li>turning on the air conditioning 30–40 minutes before the end of the night</li></ul><p><strong>If you’d like to read more and get early access to the Alice</strong> app, check this link: <a href="https://everyday.techsistence.com/">https://everyday.techsistence.com</a></p><p>That’s all.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=23f528a1c195" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[I asked ChatGPT for Actions not Answers]]></title>
            <link>https://medium.com/@overment/i-asked-chatgpt-for-actions-not-answers-9b039b76685?source=rss-30f3c12d2009------2</link>
            <guid isPermaLink="false">https://medium.com/p/9b039b76685</guid>
            <category><![CDATA[no-code]]></category>
            <category><![CDATA[automation]]></category>
            <category><![CDATA[chatgpt]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Adam Gospodarczyk]]></dc:creator>
            <pubDate>Sun, 12 Mar 2023 15:23:06 GMT</pubDate>
            <atom:updated>2023-03-12T18:18:37.827Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*O3EvmkeqhYhxOEaRRVSWNg.png" /></figure><p>If you consider your daily life in terms of a really broad perspective, you’ll notice that all of this may be summarized with a single word: <strong>Communication</strong>.</p><p>As you already know, ChatGPT is pretty good at this. It can answer your question, and most of the time, a response will be accurate. What’s more, it looks like <a href="https://github.com/microsoft/visual-chatgpt">ChatGPT will be able to understand images and even videos</a>. But still, there is one piece missing here, and that is: <strong>an action</strong>.</p><p>It may be first hard to imagine, that text generation may lead to performing some task. But if you’re a programmer, you already know, it constantly happens. Based on the user’s input, an application may generate a structure (usually a JSON object) which is sent as an HTTP request to an API.</p><p>So, if ChatGPT may generate some text, and what’s more important, a response may be formatted in a way we’ll describe, like JSON, we can use it to trigger any action we may need. It sounds simple, and in fact, it is, but there are some things we have to keep in mind to overcome some barriers, like Authentication for example.</p><h3>Talk JSON to me</h3><p>Asking ChatGPT for JSON is fairly simple. We can even ask to generate JSON based on provided text. In the example below, I’ve asked to convert my {message} to an object with properties: title, description, and deadline (in the given format).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*nKTQ7dgb0M2p2CTWCIGTlA.png" /></figure><p>Directly in ChatGPT, there is little more we can do with this object. However, if we use the ChatGPT API, we can create our own interface. Based on the user’s description, action, or other kind of event, an object may be generated and sent to our own API, where the provided information will be sent to our Google Calendar or any other app.</p><p>I’ve created such an interface by myself. As you see, the only thing I have to do is to provide a description of an event and press “Add to the Calendar”. As a response, I’ve received an object with the same structure I had in ChatGPT.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*pQi0tSar_PTRHMe5RkCaUA.png" /><figcaption>GPT-3.5 Turbo may generate a JSON object based on a provided message.</figcaption></figure><p>In this case, however, receiving a response isn’t the only thing I may have. I can easily detect if there is some kind of object (using regex, for example), parse it and send it to the specified endpoint. In fact, I did this as well. Because of that other action — adding Quick Notes to the Notion, looks a bit different.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*G57Inex-E8mxnkUbfs-AXg.png" /><figcaption>Generated JSON may be used for HTTP request</figcaption></figure><p>As you can see, the response I’ve received is “I’ve added this to Notion”. These are not just words, and you can see the proof in the screenshot below.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*kAOhEwJaubBbj-9b8bDSdQ.png" /><figcaption>GPT-3.5 Turbo to Notion</figcaption></figure><p>There are at least a few questions here. How did GPT-3.5 Turbo know how to respond to my message? Where is the JSON object I’ve mentioned, and how did it happen that this note appeared in my Notion? Let’s break it down.</p><p>First, let’s take a look at how this action was defined. As you can see, there is a prompt, which is similar to the one we’ve created previously. But here we have an option to provide a Webhook for Remote Action. So the JSON object returned from GPT-3.5 is simply being sent to the provided URL.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*u81fQ7RN9V-wncVnfujFdA.png" /><figcaption>Custom action for GPT-3.5 Turbo in my Assistant app</figcaption></figure><p>Remote Action was responsible for the actual connection with my Notion and appending the provided text to the specified block. Then it returned a response, “I’ve added this note to Notion.”</p><p>That’s it! This simple concept enables ChatGPT to not only generate text, but also to perform an action based on a description. All we have to do is ask ChatGPT for a JSON object, grab it, and send it over HTTP to any service, app, or device we need. Such a request may return a response which can be displayed to the user or passed back to ChatGPT to make it less generic.</p><h3>AI &amp; Automation</h3><p>Programming even simple actions like “sending a message on Slack” or “creating a draft message” isn’t simple. It requires you to delve into documentation, go through an authentication process, handle errors, domain configuration, etc. Such things make this whole concept very difficult or even unachievable. And this is the moment where Automation comes in. With platforms like make.com, creating a scenario that will connect to the various services and return a response takes just a few clicks.</p><p>Instead of writing all integrations from scratch, I have created a scenario that can be triggered with a webhook. Then, I just connect to my Notion and specify which information goes where. The last step is to return a webhook response with any details I need to display in my app.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*j-R0GVhJgiEhEMztBd6eMA.png" /><figcaption>Make.com scenario for sending a Quick Note to the Notion</figcaption></figure><p>Actually, here we have three things that come into play, and those are: AI, automation (no-code), and programming. Luckily, in my case, I’m familiar with all of them.</p><p>I won’t try to convince anyone, but I hope that the examples above clearly show what may be possible, and what’s more, how easy it is. In my case, defining an action is just about writing 2–3 sentences of a prompt, and a couple of clicks on make.com to create an automation. I hope that this will encourage you to dive deeper into both ChatGPT and Automation. If you can write code already, it looks like you’re in a good place these days.</p><h3>It’s not really about prompts, code or even actions</h3><p>Everything you just saw here is just an idea. It works and is really useful, I have to say, but still, it’s simply an idea. What is essential here is solving a real problem instead of trying to create one and then provide a solution.</p><p>There are many features, apps, services, and products that try to use AI, but many times it’s just not a good idea. Today, it’s fairly simple to come up with a solution that uses simple API calls to OpenAI and generates some fancy responses. In practice though, in most cases, you don’t need AI to achieve even better results.</p><p>But if you discover a real problem for which a solution is outside the reach of some programming, but it is for AI, then go for it. What’s more, with some no-code tools and automation, you may save a lot of energy and time, which you can use for something else.</p><p>There’s a quote I love from “Getting Real”:</p><blockquote>The best designers and the best programmers aren’t the ones with the best skills, or the nimblest fingers, or the ones who can rock and roll with Photoshop or their environment of choice, they are the ones that can determine what just doesn’t matter. That’s where the real gains are made.</blockquote><p>And the same applies to the AI.</p><h3>Okay, so what can I do?</h3><p>There are a few skills I’ve used to create my app, like:</p><ul><li>programming (JavaScript / Node.js / Svelte / Rust)</li><li>prompting (start with OpenAI docs and learnprompting.org)</li><li>automation (especially make.com)</li></ul><p>As you can see, the road is simple but not easy. If you already know how to code, then go and deep dive into the OpenAI documentation and play with no-code tools like make.com or n8n.io. Then just connect these skills together to tackle some real-world problems and try to solve them because now you have an idea of how to use AI not just for generating texts, but also for performing any action you need.</p><p>P.S.: 💎 If you want to be among the first people using my app I’ve mentioned in this article, please, sign up <a href="https://getmonday.app">for the waitlist</a>.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9b039b76685" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[overment labs — M.O.N.D.A.Y an Intelligent Assistant]]></title>
            <link>https://medium.com/@overment/overment-labs-m-o-n-d-a-y-an-intelligent-assistant-34ae716cc4f9?source=rss-30f3c12d2009------2</link>
            <guid isPermaLink="false">https://medium.com/p/34ae716cc4f9</guid>
            <category><![CDATA[automation]]></category>
            <category><![CDATA[no-code]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[gpt-3]]></category>
            <category><![CDATA[openai]]></category>
            <dc:creator><![CDATA[Adam Gospodarczyk]]></dc:creator>
            <pubDate>Sun, 15 Jan 2023 14:35:56 GMT</pubDate>
            <atom:updated>2023-01-15T14:42:29.006Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*OQBZwg_HiHtULfqeiFFnHA.jpeg" /></figure><h3>M.O.N.D.A.Y an Intelligent Assistant built with Make.com, Airtable and GPT-3</h3><p>Recently I created overmentAI, which is able to answer questions related to me and my publications. Despite the impressive quality of the answers, I didn’t see much value for end-user in this tool.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/953/1*7_osuBA84G2n-uJy0YGvuA.jpeg" /></figure><p><a href="https://medium.com/@overment/sarcastic-gpt-3-chatbot-who-can-remember-anything-built-in-node-js-openai-api-a84d64d5febc">I have described a process of creating an GPT-3 chatbot</a> that is able to remember everything and answer questions. Preparation and development of the bot was based on an application created in Node.js, and adding integrations e.g. with Airtable required time which I didn’t want to spend on this project. In the end, OvermentAI in this form was nothing more than a GPT-3 search engine, returning results in a rather non-standard search engine form.</p><p>Furthermore, a well-configured Elasticsearch or Algolia would provide much more value to the user, by redirecting them to a specific source of information. Theoretically, overmentAI can also return URLs, but in the end for me it’s still “not enough”.</p><p>This does not change the fact that the concept of being able to “remember” any information and access it when needed is great. From this came the idea of preparing an intelligent assistant. So instead of releasing the project externally, such an overmentAI would work for me, intelligently helping me with everyday tasks.</p><h3>Vision</h3><p>The idea of a smart assistant mainly consists in being able to collect various information about my activities, resources and everything that will be needed to carry out tasks in the right way. It is already happening with tools like Keyboard Maestro and Shortcuts. However, it is not only about simple data collection but also about categorizing, tagging and formatting them appropriately.</p><p>OvermentAI showed me that despite the large capabilities of GPT-3, going beyond something more than “a nice search engine” requires a solid preparation of information sets. It was not a particular surprise but I could see in practice what it exactly means.</p><p>Anyway, at this point automation and scripts and macros that I already use come into play. I’ll just expand them with an additional step, recording my actions in Airtable, which will later be used by an intelligent assistant called <strong>M.O.N.D.A.Y</strong> (Master of Nous Design and Automation Y’all-in-one).</p><blockquote>&gt; The Greek word “Nous” translates to “reason”, “intelligence” or “understanding”. It is a philosophical term that refers to the capacity for knowledge and comprehension of reality. In some contexts, it is also used as an alternative for the idea of “soul” or “mind”.</blockquote><p>Such an assistant could perform several roles for me and taking into account the fact that he would have access to a large database of knowledge about my activities, he would do it very effectively. M.O.N.D.A.Y in short will be able to:</p><ul><li>have a perfect memory</li><li>organize informations for me</li><li>performing voice / text commands powered with automation scenario</li><li>provide me anything I need based on my personal database</li><li>manage some processes</li><li>help me in decision making process</li><li>generate text based on “everything I know”</li></ul><p>Interestingly, I already have almost all of these processes addressed in some way with automation, but it’s up to me to “think” about how to use this data. Now I could connect it with AI. The effects could be incredibly interesting and for example I could:</p><ul><li>Ask about “everything I know about the topic”</li><li>“Remember” anything and recall it easily</li><li>Automatically transfer notes/information to the right place only based on their content</li><li>Ask for any information that could be updated in real time (because the database is in Airtable)</li></ul><p>The whole idea boils down to strengthening my potential, memory and delegating even those tasks that automation was not able to achieve on its own.</p><h3>Challenge</h3><p>The vision may be attractive but it’s still just a vision composed of individual ideas and a few tests that may not work at even a medium scale. Additionally, the system would have to be thought out very well from the start, as making changes to a large amount of information could be costly (either in time or money). For example, indexing 1.5 million characters of text I wrote cost me ~$25. If I wanted to extract keywords etc., the cost would be correspondingly higher.</p><p>The next challenge is the speed of such AI. Each query is executed in ~5–10 seconds, which means that the precision must be high enough to justify the waiting time for the response. Otherwise, it will be faster to come up with a solution on your own.</p><p>The last challenge is the fact that some of the information I’m working with can’t go beyond my local database. These are not particularly confidential information, but I just prefer to keep them with me. I already have this problem partially solved by the fact that each set of information has its own unique identifier, which I can use to also retrieve the data that openAI doesn’t know about.</p><p>All of the above seems possible to solve and I think that, just like with automation, the process of building such an assistant won’t happen in one evening. Moreover, it can’t be like that because M.O.N.D.A.Y has to directly learn and adapt to my everyday life, developing skills and changing role. Both the learning mechanism and the work with new and older sets of information should be flexible and easy to develop, modify and remove elements that are no longer needed. In other words, it has to be a “living organism”.</p><h3>Solution</h3><p>Experiments with overmentAI turned out to be a hit, because I understood what I need to build M.O.N.D.A.Y. What’s more, I was able to successfully transfer overmentAI from code written in JavaScript &amp; node.js to a <a href="https://make.com">https://make.com</a> scenario. Currently, the “brain” looks like below but it will certainly change soon and this is not the final version:.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*unlZSZUMmG1ILhfMNfomng.png" /><figcaption>make.com automation scenario</figcaption></figure><p>I would ideally like to get to the point where the above scenario is the place where absolutely all requests from other scenarios come to. It is here that the main decisions will be made and my data will be accessed directly. All other scenarios will communicate with this one.</p><p>I think that the whole project is possible to realize and that I will see the first effects already in the coming days. Of course I will share with you the effects and progress. At the moment when I finish the scenario, it is not unlikely that I will make it available to others. At this moment it is not ready yet.</p><p>What do you think about all this? 🙃</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=34ae716cc4f9" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Sarcastic GPT-3 chatbot who can remember anything built in Node.js & OpenAI API]]></title>
            <link>https://medium.com/@overment/sarcastic-gpt-3-chatbot-who-can-remember-anything-built-in-node-js-openai-api-a84d64d5febc?source=rss-30f3c12d2009------2</link>
            <guid isPermaLink="false">https://medium.com/p/a84d64d5febc</guid>
            <category><![CDATA[javascript]]></category>
            <category><![CDATA[nodejs]]></category>
            <category><![CDATA[api]]></category>
            <category><![CDATA[openai]]></category>
            <category><![CDATA[gpt-3]]></category>
            <dc:creator><![CDATA[Adam Gospodarczyk]]></dc:creator>
            <pubDate>Wed, 04 Jan 2023 18:10:20 GMT</pubDate>
            <atom:updated>2023-01-05T16:24:33.478Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*qtRgPvSo74OkNfALzMKjXA.png" /><figcaption>Sarcastic GPT-3 chatbot built with Node.js &amp; OpenAI API</figcaption></figure><h3>Creating a Sarcastic Chatbot with Node.js &amp; OpenAI API</h3><p><strong>Goal:</strong> Chatbot who can remember everything and answer questions. <br><strong>Tools:</strong> Node.js (NestJS), PostgreSQL (Prisma), OpenAI API &amp; Pinecone (as vector database). <br><strong>Concepts:</strong> API, Embedding, AI, Prompt Design, SQL and Vector Database<br><strong>Source: </strong><a href="https://github.com/iceener/tuesday-chatbot-gpt-3">https://github.com/iceener/tuesday-chatbot-gpt-3</a></p><p>Using the GPT-3 API is a very interesting tool we can use to build something on top of it. GPT-3 is, by default, quite intelligent and can answer various questions, but obviously has no knowledge about something specific to us. Therefore, we will create a bot that will answer questions using our own information.</p><h3><strong>Completions and prompts</strong></h3><p>OpenAI API can create a <a href="https://beta.openai.com/docs/api-reference/completions">completion</a> for provided prompt and parameters. Prompt is limited to 2048–4096 tokens (depending on a model) which means that both prompt and returned completion have to be shorter than something like 1500–2500 words. Keep that in mind.</p><p>Basic conversation looks like this:</p><blockquote>- (me) How are you?<br>- (gpt-3) I’m doing well, thank you. How about you?</blockquote><p>But I can modify prompt to specify what exactly I need, just look:</p><blockquote>- (me) How are you? Answer in a sarcastic way.<br>- (gpt-3) Oh, just peachy.</blockquote><p>I have a problem with this. All of those requests are independent, which means we have no context here. Of course we can add one.</p><blockquote>Context: “Adam is a great guy!”<br>Answer a question below using context.</blockquote><blockquote>Question (me): Who is Adam?<br>Answer (gpt-3): Adam is a great guy.</blockquote><p>I think you already know what we can do with this: build a chatbot that can remember everything. We can just save any user input in our database and add it as a context to figure out how to keep it below the prompt limit.</p><h3>Setup a project</h3><p>A first step will be to create a new NestJS project with Nest CLI.</p><pre>$ nest new tuesday &amp;&amp; cd tuesday</pre><p>Next we have to generate a module and a couple of services:</p><pre>$ nest g mo prisma<br>$ nest g mo messages<br>$ nest g s messages<br>$ nest g co messages<br>$ nest g s messages/openai<br>$ nest g s messages/pinecone</pre><p>And now we need to setup Prisma, just by following [a recipe](<a href="https://docs.nestjs.com/recipes/prisma">https://docs.nestjs.com/recipes/prisma</a>) from NestJS docs:</p><pre>$ pnpm i prisma @prisma/client -D<br>$ snpx prisma init</pre><p>Go to .env file and update DATABASE_URL. I use PostgreSQL here:</p><pre>DATABASE_URL=&quot;postgresql://overment:password@localhost:5432/tuesday?schema=public&quot;</pre><p>There is a need to set up a connection in PrismaService.</p><pre>import { INestApplication, Injectable, OnModuleInit } from &#39;@nestjs/common&#39;;<br>import { PrismaClient } from &#39;@prisma/client&#39;;<br><br>@Injectable()<br>export class PrismaService extends PrismaClient implements OnModuleInit {<br>  async onModuleInit() {<br>    await this.$connect();<br>  }<br><br>  async enableShutdownHooks(app: INestApplication) {<br>    this.$on(&#39;beforeExit&#39;, async () =&gt; {<br>      await app.close();<br>    });<br>  }<br>}</pre><p>Make PrismaModule global by adding <a href="http://twitter.com/Global">@Global</a>() decorator and exports array in <a href="http://twitter.com/Module">@Module</a>() decorator:</p><pre>import { Global, Module } from &#39;@nestjs/common&#39;;<br>import { PrismaService } from &#39;./prisma.service&#39;;<br><br>@Global()<br>@Module({<br>  providers: [PrismaService],<br>  exports: [PrismaService],<br>})<br>export class PrismaModule {}</pre><p>And add listeners in main.ts file.</p><pre>// (...)<br>import { PrismaService } from &#39;./prisma/prisma.service&#39;;<br><br>async function bootstrap() {<br>  const app = await NestFactory.create(AppModule);<br>  // ---- add this<br>  const prismaService = app.get(PrismaService);<br>  await prismaService.enableShutdownHooks(app);<br>  // ----<br>  await app.listen(3000);<br>}<br>bootstrap();</pre><p>The last step to set up Prisma is to specify a model. In our case, it will be a simple one, so just open the `prisma/schema.prisma` file and update it like so:</p><pre>generator client {<br>  provider = &quot;prisma-client-js&quot;<br>}<br><br>datasource db {<br>  provider = &quot;postgresql&quot;<br>  url      = env(&quot;DATABASE_URL&quot;)<br>}<br><br>model Message {<br>  id    Int     @default(autoincrement()) @id<br>  message String<br>}</pre><p>You can extend this model in the future, but at this point, this is everything we need. The only thing we need to do is to migrate our database with a command:</p><pre>$ npx prisma migrate dev --name init</pre><h3>First endpoint</h3><p>Our SQL database is ready. We will store all of the messages our bot will receive and use them for creating context of a current prompt. So, go to the <strong>messages.service.ts</strong> and update it by adding <strong>createMessage</strong> method and <strong>checkIfMessageExists</strong> method, to make sure we won’t store duplicated messages.</p><pre>constructor(private readonly prisma: PrismaService) {}<br>  <br>  async createMessage(message: string) {<br>    const exists = await this.checkIfMessageExists(message);<br>    if (exists) {<br>      return {<br>        ...exists,<br>        created: false,<br>      };<br>    }<br>    <br>    const newMessage = await this.prisma.message.create({<br>      data: {<br>        message,<br>      },<br>    });<br>    <br>    return {<br>      ...newMessage,<br>      created: true,<br>    };<br>  }<br>  <br>  private async checkIfMessageExists(message) {<br>    return this.prisma.message.findFirst({<br>      where: {<br>        message,<br>      },<br>    });<br>  }</pre><p>We can use the Service methods in <strong>messages.controller.ts</strong> to create our first endpoint which will respond to <strong>POST localhost:3000/messages</strong> requests, save the message in an SQL database, and return a new message ID.</p><pre>constructor(<br>    private readonly messagesService: MessagesService,<br>    private readonly openaiService: OpenaiService,<br>    private readonly pineconeService: PineconeService,<br>  ) {}<br><br>  @Post()<br>  async createMessage(@Body() data: { message: string }) {<br>    const { id, created } = await this.messagesService.createMessage(<br>      data.message,<br>    );<br>    <br>    return id;<br>  }</pre><h3>Connecting to OpenAI API</h3><p>We need to start talking to the OpenAI API. Please keep in mind that this is a paid API; after creating an account, make sure you have proper usage limits set. The good news is that simple API calls shouldn’t cost much. After creating an account, just grab an API key and come back to me.</p><p>Install openai and update .env file</p><pre>$ pnpm i openai</pre><pre>// .env file<br>OPENAI_API_KEY=sk-...................</pre><p>Now we can go to the <strong>openai.service.ts</strong> and add a method which will get a completion for a given prompt. In <strong>createCompletion</strong> there is something I need to explain:</p><ul><li>The ‘text-davinci-003’ model is the most advanced and expensive model currently available. If you want or need to, you can try other models.</li><li>Remember that the ‘max_tokens’ value is the total length of a prompt and completion, not a character count; however, for now, 250 should be enough for a simple chat.</li><li>A temperature indicates how “creative” a response should be. Values closer to 0 indicate a more strict answer, while values closer to 1 indicate a more creative one. But keep in mind this is also a simplification and not exact description of this value, but you can think about it in this way.</li></ul><pre>import { Injectable } from &#39;@nestjs/common&#39;;<br>import { Configuration, OpenAIApi } from &#39;openai&#39;;<br><br>@Injectable()<br>export class OpenaiService {<br>  constructor() {<br>    const configuration = new Configuration({<br>      apiKey: process.env.OPENAI_API_KEY,<br>    });<br>    this.openai = new OpenAIApi(configuration);<br>  }<br><br>  async createCompletion(prompt, context) {<br>    const completion = await this.openai.createCompletion({<br>      model: &#39;text-davinci-003&#39;,<br>      prompt: prompt,<br>      max_tokens: 250,<br>      temperature: 0.2,<br>    });<br><br>    return completion?.data.choices?.[0]?.text;<br>  }<br>}</pre><p>We can go back to the <strong>messages.controller.ts</strong> and replace <strong>createCompletion</strong> method like this:</p><pre>  @Post()<br>  async createMessage(@Body() data: { message: string }) {<br>    const { id, created } = await this.messagesService.createMessage(<br>      data.message,<br>    );<br>   <br>    return this.openaiService.createCompletion(data.message, &#39;&#39;);<br>  }</pre><h3>Designing a Prompt</h3><p>And this point our endpoint should return a response from Open AI API, which is great! It’s time to update our prompt, so bot will answer only based on provided context or reply “I don’t know” in a sarcastic way. We can do this in <strong>openai.service.ts</strong> and again there is a couple things I need to explain.</p><p>First of all, our prompt will now look something like this:</p><blockquote>Based on this context: \n\n\n<br>[generated context]<br>\n\n\n<br>Answer the question below as truthfully as you can, if you don’t know the answer, say you don’t know in a sarcastic way otherwise, just answer.<br>\n\n\n<br>[current user input / question]</blockquote><p>As you can see, we’re clearly specifying a context and instruction for generating an answer. All of those sections are separated with <strong>\n\n\n</strong> and at this point our context is empty, so bot will always reply “I don’t know” in a sarcastic way.</p><pre>@Injectable()<br>export class OpenaiService {<br>  private CONTEXT_INSTRUCTION = &#39;Based on this context:&#39;;<br>  private INSTRUCTION = `Answer the question below as truthfully as you can, if you don&#39;t know the answer, say you don&#39;t know in a sarcastic way otherwise, just answer.`;<br>  private openai;<br>  constructor() {<br>    const configuration = new Configuration({<br>      apiKey: process.env.OPENAI_API_KEY,<br>    });<br>    this.openai = new OpenAIApi(configuration);<br>  }<br><br>  async createCompletion(prompt, context) {<br>    const completion = await this.openai.createCompletion({<br>      model: &#39;text-davinci-003&#39;,<br>      prompt: `${this.CONTEXT_INSTRUCTION}\n\n\nContext: &quot;${context}&quot; \n\n\n${this.INSTRUCTION} \n\n\n ${prompt}`,<br>      max_tokens: 250,<br>      temperature: 0.2,<br>    });<br><br>    return completion?.data.choices?.[0]?.text;<br>  }<br>}</pre><h3>Creating Context and Embeddings</h3><p>We’re getting closer. Our prompt needs a proper context, but we can’t just grab everything we have in our PostgreSQL database. Instead, we will get the current user input and fetch only meaningful information from it and use it as a “context”. We can do this by creating embeddings based a data we have (an array of something called “vectors”). These embeddings will be saved in the <strong>Pinecone </strong>database. Pinecone is also able to query these embeddings and find those which are close to each other. In a simple words, we will be able to search for everything we have which is in some way connected with a current user input. Thanks to that our context won’t be too long and in the same time it will be useful for creating a good completion.</p><p>Let’s go to the <strong>pinecone.service.ts</strong>. We need there a three things:</p><ul><li>setup PineconeClient (which needs to be installed with <strong>pnpm i pinecone-client</strong>`) You have to add API KEY, Base URL and Pinecone Namespace in .env file. You can get them in <a href="https://app.pinecone.io/organizations/-NKr0C04jLfQI1aQuNYM/projects/us-west1-gcp:b0b21e1/indexes/monday">Pinecone Console</a>. New index should have a name (in my case it’s `tuesday`), dimension set to 1536 and Pod Type set to P2. Please keep in mind that Base URL should start with `<a href="https://`">https://`</a>, like: `<a href="https://tuesday-XXXXXXX.svc.us-west1-gcp.pinecone.io`">https://tuesday-XXXXXXX.svc.us-west1-gcp.pinecone.io`</a></li><li>a method for upserting (create or update) a vectors. As you can see, this is something really simple.</li><li>a query method which will take a given vector and return top 10 results, which then we’re filtering and mapping to an array of numbers (identifiers). Those identifiers will be used for querying informations from our SQL database.</li></ul><p>If at this point something is not clear, please go ahead. It will be in a moment.</p><pre>import { Injectable } from &#39;@nestjs/common&#39;;<br>import { PineconeClient } from &#39;pinecone-client&#39;;<br><br>@Injectable()<br>export class PineconeService {<br>  private pinecone;<br>  constructor() {<br>    this.pinecone = new PineconeClient({<br>      apiKey: process.env.PINECONE_API_KEY,<br>      baseUrl: process.env.PINECONE_BASE_URL,<br>      namespace: process.env.PINECONE_NAMESPACE,<br>    });<br>  }<br><br>  upsert(vectors) {<br>    return this.pinecone.upsert({<br>      vectors: [vectors],<br>    });<br>  }<br><br>  async query(vector) {<br>    const { matches } = await this.pinecone.query({<br>      vector,<br>      topK: 10,<br>      includeMetadata: true,<br>      includeValues: false,<br>    });<br><br>    return matches<br>      .filter((match) =&gt; match.score &gt; 0.8)<br>      .map((match) =&gt; parseInt(match.id));<br>  }<br>}</pre><p>Now we need to generate embeddings and we can do this in <strong>openai.service.ts</strong> by adding a method like so:</p><pre>async createEmbedding(prompt) {<br>    const { data: embed } = await this.openai.createEmbedding({<br>      input: prompt,<br>      model: &#39;text-embedding-ada-002&#39;,<br>    });<br><br>    return embed;<br>}</pre><p>Method above will take a prompt (user input), generate a list of vectors and return them as an embed. This embed will have to go to the Pinecone database.</p><p>Now we can put this all together, but there is one more thing we need and I am speaking of two methods in <strong>messages.service.ts</strong>. First one will find messages based on array of identifiers. A second one will create a context by joining <strong>messages</strong> together.</p><pre>async messages(ids: number[]) {<br>    return this.prisma.message.findMany({<br>      where: {<br>        id: {<br>          in: ids,<br>        },<br>      },<br>    });<br>  }<br><br>  async getContext(ids: number[]) {<br>    return (await this.messages(ids))<br>      .filter(<br>        (message, index, self) =&gt;<br>          index === self.findIndex((t) =&gt; t.message === message.message),<br>      )<br>      .reduce((acc, message) =&gt; {<br>        return acc + message.message + &#39;\n&#39;;<br>      }, &#39;&#39;);<br>  }</pre><h3>Finishing</h3><p>The last step will be to update `messages.controller.ts` and use all of those methods.</p><ul><li>endpoint takes a message from a user</li><li>we’re storing this message in our database or returning an existing one</li><li>user’s message is converted to the embedding</li><li>this embedding is used to query Pinecone database which will return us an array of `ids` which we can use in `messages.service.ts` to fetch an actual content.</li><li>based on returned `ids` we’re fetching messages and joining them, so we have a context</li><li>if a message is new, we store its embedding in a Pinecone</li><li>we’re asking GPT-3 for a completion based on a given prompt and generated context</li></ul><p>In other words, our app takes user input and saves it both in an SQL database and in Pinecone (as an embedding). Next, it uses the generated embedding to search for other vectors that are close. Because the IDs in Pinecone and PostgreSQL are the same values, we can use a result from Pinecone to fetch a message from PostgreSQL and create context. The last step is to use a prompt that uses this context, instruction, and user input. The bot will respond with “I don’t know” or answer accordingly if it is able to do so based on the provided input.</p><p>A result looks like this:</p><blockquote>- (me): Who is overment?<br>- (bot): I don’t know, who is overment?<br>- (me): overment is a smart guy from Krakow.<br>- (bot): I don’t know.</blockquote><blockquote>- (me): Do you know where overment lives?<br>- (bot): I don’t know, but I’m sure it’s somewhere in Krakow.<br>- (me): Who is overment?<br>- (bot): overment is Adam’s nickname.</blockquote><p>A last response shows that the bot has remembered something I told him earlier about me and was able to correctly answer my latest question. It means that we have a bot who is able to remember everything we say and use that information in a further discussion.</p><h3>What’s next?</h3><p>At this point, our bot is very simple. It’s easy to trick him, and the quality of his answers depends mostly on what we tell him, but that’s okay. We can think of preparing a set of information which he will remember and then use to answer our questions or search for them.</p><p>Definitely, it would be a good idea to learn more about prompt design, embeddings, and creating datasets, which may lead us to making a bot that can search for information, answer questions based on data we provide, or even take some actions based on given instructions.</p><p>If you have any feedback or an ideas you’d like to share with me, just leave a comment.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a84d64d5febc" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>