Loading video...

Video Failed to Load

Go Home

People of pi. Cloudflare release with a minor breaking change. - JSON validation no longer relies on AJV, but uses Typebox. Update your existing extensions as per changelog (@sinclair/typebox -> @typebox) - OSC 9;4 indicator in supporting terminals for Lucas Meijer thanks to MVP Kayla Cinnamon ☕ - Tools...

badlogicgames's profile picture

Mario Zechner

52,437 subscribers

20,680 views • 4 months ago •via X (Twitter)

LIVE

Anya Rossi• Live Now

Private livecam show

0 Comments

No comments available

Comments from the original post will appear here

Related Videos

I plan on building up People First as the party of choice for the minor parties. No other party have the policies, the know how or the grassroots membership to do this. People First plans to set up branches, something One Nation hasn't done in 30 years. This clip is from a live I did with Scott Challen yesterday. You can watch the whole thing here: Join People First Party now and be a part of building a movement to take on the major parties and be a clear choice for change:
2:41

I plan on building up People First as the party of choice for the minor parties. No other party have the policies, the know how or the grassroots membership to do this. People First plans to set up branches, something One Nation hasn't done in 30 years. This clip is from a live I did with Scott Challen yesterday. You can watch the whole thing here: Join People First Party now and be a part of building a movement to take on the major parties and be a clear choice for change:

Gerard Rennick

53,548 views • 1 year ago

Released SuperCmd v2-1.0.4 🚀 Big fat update this time!! Here's what's new 💫 • Completely local translation supporting 30+ languages, plus a Translate Selection command for translating large texts from any app • Support for custom AI providers via OpenAI and Anthropic-compatible APIs • 10+ search engines for browser searches (Bing, Baidu, Yandex...) and the option to create your own custom search engine • Support for custom Quick AI prompts, each with its own hotkey • Support for Apple Shortcuts with hotkeys • Read Aloud Edge TTS now has a voice catalogue covering 40+ languages and 100+ dialects • Support for selecting the reasoning effort for ChatGPT in AI Chat • Support for backing up SuperCmd data • Added gitignore-style exclusion patterns for the file indexing. • New Commands: Window Fill, Reset Launcher Position, Changelog • More cosmetic changes to the UI and a bunch of bug fixes Other key improvements 🧐 • Better UI for Quick Chat, Next / Prev chat option • Improved file search results, now shows complete dir, complete indexing. • Improved UI for widgets, now takes up the rest of the space in the launcher • Settings now appears in Cmd-Tab while open. This was only about 20%. Read the entire changelog: Many more improvements are coming, with a lot in the queue. Download:
0:45

Released SuperCmd v2-1.0.4 🚀 Big fat update this time!! Here's what's new 💫 • Completely local translation supporting 30+ languages, plus a Translate Selection command for translating large texts from any app • Support for custom AI providers via OpenAI and Anthropic-compatible APIs • 10+ search engines for browser searches (Bing, Baidu, Yandex...) and the option to create your own custom search engine • Support for custom Quick AI prompts, each with its own hotkey • Support for Apple Shortcuts with hotkeys • Read Aloud Edge TTS now has a voice catalogue covering 40+ languages and 100+ dialects • Support for selecting the reasoning effort for ChatGPT in AI Chat • Support for backing up SuperCmd data • Added gitignore-style exclusion patterns for the file indexing. • New Commands: Window Fill, Reset Launcher Position, Changelog • More cosmetic changes to the UI and a bunch of bug fixes Other key improvements 🧐 • Better UI for Quick Chat, Next / Prev chat option • Improved file search results, now shows complete dir, complete indexing. • Improved UI for widgets, now takes up the rest of the space in the launcher • Settings now appears in Cmd-Tab while open. This was only about 20%. Read the entire changelog: Many more improvements are coming, with a lot in the queue. Download:

Shobhit - Building SuperCmd

12,022 views • 18 days ago

pixel pumper got two updates 🩷 1.0.1 → the features drop • multi-image censors in beta safe, rotate through a pool of images either per-detection (stable per tracked object) or on a timer • caption rotation speed slider • "visual only" toggle for mind mush presets so you can keep your audio settings while switching visuals • custom caption text that actually works now (auto-switches to custom mode when you type) 1.0.2 → the polish drop • your custom censor images now persist in a library, add them once, they stay • goon cave mute actually holds through mix and remix now • smoother mode switching, no more UI hitch on the gpu cleanup pause • url loader no longer crashes when you switch modes mid-load • shader fixes for older intel gpus • versioned downloads on itch so you can roll back if needed two full audit passes under the hood. 10+ crash vectors killed.
0:27

pixel pumper got two updates 🩷 1.0.1 → the features drop • multi-image censors in beta safe, rotate through a pool of images either per-detection (stable per tracked object) or on a timer • caption rotation speed slider • "visual only" toggle for mind mush presets so you can keep your audio settings while switching visuals • custom caption text that actually works now (auto-switches to custom mode when you type) 1.0.2 → the polish drop • your custom censor images now persist in a library, add them once, they stay • goon cave mute actually holds through mix and remix now • smoother mode switching, no more UI hitch on the gpu cleanup pause • url loader no longer crashes when you switch modes mid-load • shader fixes for older intel gpus • versioned downloads on itch so you can roll back if needed two full audit passes under the hood. 10+ crash vectors killed.

Pixel Pumper

24,719 views • 4 months ago

Introducing... the NoCodeFusion MVP builder! 🥳 Type in your app idea, get a full-stack Bubble app in ~10 minutes with authentication, database, UI, and basic create/update/delete workflows. The perfectionist in me is dying, but f**k it - we're launching. See a demo below. It's not perfect, but an excellent way to kickstart a project. It uses my component library. You can easily change the main font and primary colors from the Bubble style variables. You can then download the Chrome extension which lets you continue to build on your app in the Bubble editor with a copilot chat interface OR hire a developer through us to help you finish the project. This project was built 100% by myself entirely on Bubble (and of course using NoCodeFusion) and is very much in beta. If there's a bug or something, go easy on me. Send me a message and I'll take a look. AI + Visual editor is now an incredible option to create full stack web apps. It allows you the power of creating something in minutes, then having an intuitive interface to make edits and understand what's actually going on in the logic. Let me know what you think ✌️
5:22

Introducing... the NoCodeFusion MVP builder! 🥳 Type in your app idea, get a full-stack Bubble app in ~10 minutes with authentication, database, UI, and basic create/update/delete workflows. The perfectionist in me is dying, but f**k it - we're launching. See a demo below. It's not perfect, but an excellent way to kickstart a project. It uses my component library. You can easily change the main font and primary colors from the Bubble style variables. You can then download the Chrome extension which lets you continue to build on your app in the Bubble editor with a copilot chat interface OR hire a developer through us to help you finish the project. This project was built 100% by myself entirely on Bubble (and of course using NoCodeFusion) and is very much in beta. If there's a bug or something, go easy on me. Send me a message and I'll take a look. AI + Visual editor is now an incredible option to create full stack web apps. It allows you the power of creating something in minutes, then having an intuitive interface to make edits and understand what's actually going on in the logic. Let me know what you think ✌️

Peter Kwitny

20,070 views • 1 year ago

I asked Garry Tan how to use meta prompting to get better at AI: "My partners at YC Jared Friedman and Pete Koomen showed me how to do this. You can take almost anything that you do all the time and just drop it into a context window. And then say, “Here’s a bunch of inputs and outputs." And maybe you also add a bunch of notes. And then you tell it, “Write me a prompt that can act as an agent that takes this input and makes this output over here.” You can do this for almost any type of knowledge work. And you can even introspect. "What are things you notice that I did to convert this from the input to the output?”. And then you can just start using the prompt. Initially, it’s going to suck. Because it’s just not that smart yet. But what’s funny is now, I also use it to Iterate my writing. You can be very direct, "I would never say that", "Don’t say it like this", or "Oh, you used the long word there, use the short word". Just speak to it conversationally. And then when you're happy with the output, you can use that new output to make a new prompt. "Based on this conversation, give me a better initial prompt that incorporates all the things we talked about." And you can do this with literally everything. And in theory, there’s so much it applies to that people do day-to-day. You could use it for tweets. You could use it for editing podcasts. You can use it for pretty much everything. I have a folder of prompts that I use all the time. My YouTube prompt is on v27 or something. I'll go through this process with all the different max models. I'll use GPT 5.2 Pro. I’ll use Grok. I'll use Claude. Then, I’ll take all the outputs from all the models and put them into Claude and say "Here’s my prompt, here’s the output from four LLMs, including yourself. Rate each response and tell me what the pros and cons of each approach are." And I usually say "give it to me in numbered form". And then you can agree with one, disagree with two, tell it three is this or that. And then after that, you say given all of this, synthesize it."
4:16

I asked Garry Tan how to use meta prompting to get better at AI: "My partners at YC Jared Friedman and Pete Koomen showed me how to do this. You can take almost anything that you do all the time and just drop it into a context window. And then say, “Here’s a bunch of inputs and outputs." And maybe you also add a bunch of notes. And then you tell it, “Write me a prompt that can act as an agent that takes this input and makes this output over here.” You can do this for almost any type of knowledge work. And you can even introspect. "What are things you notice that I did to convert this from the input to the output?”. And then you can just start using the prompt. Initially, it’s going to suck. Because it’s just not that smart yet. But what’s funny is now, I also use it to Iterate my writing. You can be very direct, "I would never say that", "Don’t say it like this", or "Oh, you used the long word there, use the short word". Just speak to it conversationally. And then when you're happy with the output, you can use that new output to make a new prompt. "Based on this conversation, give me a better initial prompt that incorporates all the things we talked about." And you can do this with literally everything. And in theory, there’s so much it applies to that people do day-to-day. You could use it for tweets. You could use it for editing podcasts. You can use it for pretty much everything. I have a folder of prompts that I use all the time. My YouTube prompt is on v27 or something. I'll go through this process with all the different max models. I'll use GPT 5.2 Pro. I’ll use Grok. I'll use Claude. Then, I’ll take all the outputs from all the models and put them into Claude and say "Here’s my prompt, here’s the output from four LLMs, including yourself. Rate each response and tell me what the pros and cons of each approach are." And I usually say "give it to me in numbered form". And then you can agree with one, disagree with two, tell it three is this or that. And then after that, you say given all of this, synthesize it."

The Peel

51,632 views • 6 months ago

The S3XY Dash from Enhance - S3XY Buttons is one of the coolest Tesla accessories I've seen yet. I know many people want Android Auto/Apple Car Play, and this gives the OPTION to use those with your Tesla, along with tons of other info. They have 20% off for preorder, and you can get an additional 10% off with my link: I love using Waze on road trips, and this looks like the way to do it going forward. On top of having a customizable sized android auto experience, the Dash can display charging, vehicle, motor, battery and tons of other info. Enhance was also smart enough to not force it directly in front of the air vents. You can place it where you'd like with their included kit. All the widgets are also customizable, so you can add any you like and resize them. You can also make multiple pages to quickly flip through different info. There's SO much you can do with this *if you want it*. Great job on this one.
0:14

The S3XY Dash from Enhance - S3XY Buttons is one of the coolest Tesla accessories I've seen yet. I know many people want Android Auto/Apple Car Play, and this gives the OPTION to use those with your Tesla, along with tons of other info. They have 20% off for preorder, and you can get an additional 10% off with my link: I love using Waze on road trips, and this looks like the way to do it going forward. On top of having a customizable sized android auto experience, the Dash can display charging, vehicle, motor, battery and tons of other info. Enhance was also smart enough to not force it directly in front of the air vents. You can place it where you'd like with their included kit. All the widgets are also customizable, so you can add any you like and resize them. You can also make multiple pages to quickly flip through different info. There's SO much you can do with this *if you want it*. Great job on this one.

Dirty Tesla

59,866 views • 9 months ago

TLDR; GH actions, but for agents. ~0ms cache, retry-on-failure, insanely fast. Agents need validation. CI is the last defense. They shouldn't bother you unless everything is green! GH Actions is usually in the top-5 expenses for dev-teams. Add agents to that mix? It'll easily double. It's the wrong tool for the right job: Slow boot, slow cache, retrieving logs is token expensive for agents, the list goes on... So I built a tool with one amazing feature: live-reload for failures. Agent-CI is a local CI runner. I tweaked the control pane and mounts to provide 0ms caching, insanely fast boots. When a step fails it pauses, provides the agent with the failure, and waits for the agent to fix and retry just that step. It uses the standard GH Actions image (via Docker), but emulates the control pane via a local HTTP server. You don't have to change any of your existing GH workflows. Tighter loops. Greener builds. Less babysitting. (Demo below.)
0:41

TLDR; GH actions, but for agents. ~0ms cache, retry-on-failure, insanely fast. Agents need validation. CI is the last defense. They shouldn't bother you unless everything is green! GH Actions is usually in the top-5 expenses for dev-teams. Add agents to that mix? It'll easily double. It's the wrong tool for the right job: Slow boot, slow cache, retrieving logs is token expensive for agents, the list goes on... So I built a tool with one amazing feature: live-reload for failures. Agent-CI is a local CI runner. I tweaked the control pane and mounts to provide 0ms caching, insanely fast boots. When a step fails it pauses, provides the agent with the failure, and waits for the agent to fix and retry just that step. It uses the standard GH Actions image (via Docker), but emulates the control pane via a local HTTP server. You don't have to change any of your existing GH workflows. Tighter loops. Greener builds. Less babysitting. (Demo below.)

Peter Pistorius

15,248 views • 5 months ago

my new song “BREAKDOWN.” is out on the 21st of June!!! 🖤🖤🖤 I wrote this poem because it’s been the hardest year for my mental health. In my life I’ve always never felt good enough, it’s just the thing that’s eaten me up. For as long as I can remember i have felt constantly afraid of how quickly my head can turn dark. It’s always been so hard to fight the darkness that i inevitably have. A lot of people will say it’s a phase and it will go away. But it doesn’t and the reality of the situation is I have to find strategies to deal with it. To put it plainly the things I don’t like about myself will probably never change, people tell me one day I’ll come-to terms with them one day but I want that day to be FUCKING NOW. This song is a message to myself to try and exist alongside my insecurities and my darkness by grounding myself and remembering what is real in life and the world is so much bigger than me. Try and get out of your head and notice the world around you, notice the things and people around you. Connect with them, the chances are they probably feel the same. Don’t let the bullshit inside your head consume you. It just wastes precious time. Remember what is real. Help people, be kind, help the world, help yourself. If you think you can’t do it, you can. You can get through this, trust me. Use this poem in a mornin to get u out of bed, use it when youre about to back out of something last minute, use it when you’re at your darkest. It’s got a little bit of light in it. Don’t forget to put your feet in the grass … Mind
0:55

my new song “BREAKDOWN.” is out on the 21st of June!!! 🖤🖤🖤 I wrote this poem because it’s been the hardest year for my mental health. In my life I’ve always never felt good enough, it’s just the thing that’s eaten me up. For as long as I can remember i have felt constantly afraid of how quickly my head can turn dark. It’s always been so hard to fight the darkness that i inevitably have. A lot of people will say it’s a phase and it will go away. But it doesn’t and the reality of the situation is I have to find strategies to deal with it. To put it plainly the things I don’t like about myself will probably never change, people tell me one day I’ll come-to terms with them one day but I want that day to be FUCKING NOW. This song is a message to myself to try and exist alongside my insecurities and my darkness by grounding myself and remembering what is real in life and the world is so much bigger than me. Try and get out of your head and notice the world around you, notice the things and people around you. Connect with them, the chances are they probably feel the same. Don’t let the bullshit inside your head consume you. It just wastes precious time. Remember what is real. Help people, be kind, help the world, help yourself. If you think you can’t do it, you can. You can get through this, trust me. Use this poem in a mornin to get u out of bed, use it when youre about to back out of something last minute, use it when you’re at your darkest. It’s got a little bit of light in it. Don’t forget to put your feet in the grass … Mind

YUNGBLUD

66,420 views • 2 years ago

Grok Build CLI just got an update with v0.2.113! 🚀 Changelog v0.2.113 Features: • MCP servers can now be enabled or disabled directly from the CLI with `grok mcp enable ` and `grok mcp disable `. • Full plan markdown can now be copied to the clipboard with `y` during plan approval or preview. • Added support for the new SuperGrok Plus subscription tier in authentication and feature gating. • Enabled automatic recovery from repetitive loops in model output by default. Performance: • Cold start shows the UI instantly while models and settings load in the background. • Large session forks and resumes now use far less memory and avoid spikes. • Prevented thread exhaustion on high-core shared machines by limiting the workspace daemon's worker threads. Bug Fixes: • Terminal command output is no longer lost or duplicated when the gateway is unreachable. • Invalid MCP server entries in config.toml no longer prevent Grok from starting; problems are shown in `grok inspect`. • SessionEnd hooks now run on exit in non-leader TUI and headless sessions. • Paste chips now display with the correct background in inline prompts and question inputs. • Pasted content chips now behave consistently when editing answers in the question view. • Background task status now shows only elapsed duration instead of absolute timestamps. • Session lists no longer drop real sessions when the remote registry reports an outdated turn count of zero. • /loop now stores prompts that include stop conditions so recurring tasks can terminate themselves when done. • Reduced spurious warning messages for common auth and config scenarios. • Fixed conda activation (and other sourced scripts that read $@) when using persistent or login-capture shells. • Fixed stuck background-task tray rows after long foreground shell commands complete. • Agent subprocesses and idle inhibitors are now cleaned up when the parent CLI process dies unexpectedly. • Fixed truncated plans in minimal mode and improved visual separation between reasoning and output (including NO_COLOR). • Fixed credential loss across multiple grok processes sharing the same auth file. • Fixed doubled Enter and other keys on older Alacritty terminals. • Fixed false paywall messages for free-tier and unmatched users.
0:49

Grok Build CLI just got an update with v0.2.113! 🚀 Changelog v0.2.113 Features: • MCP servers can now be enabled or disabled directly from the CLI with `grok mcp enable ` and `grok mcp disable `. • Full plan markdown can now be copied to the clipboard with `y` during plan approval or preview. • Added support for the new SuperGrok Plus subscription tier in authentication and feature gating. • Enabled automatic recovery from repetitive loops in model output by default. Performance: • Cold start shows the UI instantly while models and settings load in the background. • Large session forks and resumes now use far less memory and avoid spikes. • Prevented thread exhaustion on high-core shared machines by limiting the workspace daemon's worker threads. Bug Fixes: • Terminal command output is no longer lost or duplicated when the gateway is unreachable. • Invalid MCP server entries in config.toml no longer prevent Grok from starting; problems are shown in `grok inspect`. • SessionEnd hooks now run on exit in non-leader TUI and headless sessions. • Paste chips now display with the correct background in inline prompts and question inputs. • Pasted content chips now behave consistently when editing answers in the question view. • Background task status now shows only elapsed duration instead of absolute timestamps. • Session lists no longer drop real sessions when the remote registry reports an outdated turn count of zero. • /loop now stores prompts that include stop conditions so recurring tasks can terminate themselves when done. • Reduced spurious warning messages for common auth and config scenarios. • Fixed conda activation (and other sourced scripts that read $@) when using persistent or login-capture shells. • Fixed stuck background-task tray rows after long foreground shell commands complete. • Agent subprocesses and idle inhibitors are now cleaned up when the parent CLI process dies unexpectedly. • Fixed truncated plans in minimal mode and improved visual separation between reasoning and output (including NO_COLOR). • Fixed credential loss across multiple grok processes sharing the same auth file. • Fixed doubled Enter and other keys on older Alacritty terminals. • Fixed false paywall messages for free-tier and unmatched users.

Puck

94,213 views • 27 days ago

Cheryl Reeve on knowing the level of leadership from Courtney Williams and Kayla McBride had before the year (This is a long one, but is my favorite Reeve quote to date) “No, I mean you got to do your homework, you can’t just bring anybody in you know? It’s been the Lynx way, the culture, I think people have long said that about us. That you can see it in the way we play, and that it’s real and it’s sustainable, it helps you through hard times… it’s fun for them, and it’s real, you can’t fake that kind of stuff… I have made mistakes along the way, for sure, but it only reinforces… the single most important thing they’re doing is the selection of people, of course the talent is probably what puts you on the radar, but the selection of who the individual is and who they’ll be for your organization because it’s really hard to undue the damage that it causes when you don’t put the time or when you think you can change someone. Coach’s often think we can change someone, and so, I hope I’ve made my last mistake, and we don’t go down that path again. I got some safeguards, guardrails in place to holler at me if I consider such things. It leads to greater quality of life”
1:57

Cheryl Reeve on knowing the level of leadership from Courtney Williams and Kayla McBride had before the year (This is a long one, but is my favorite Reeve quote to date) “No, I mean you got to do your homework, you can’t just bring anybody in you know? It’s been the Lynx way, the culture, I think people have long said that about us. That you can see it in the way we play, and that it’s real and it’s sustainable, it helps you through hard times… it’s fun for them, and it’s real, you can’t fake that kind of stuff… I have made mistakes along the way, for sure, but it only reinforces… the single most important thing they’re doing is the selection of people, of course the talent is probably what puts you on the radar, but the selection of who the individual is and who they’ll be for your organization because it’s really hard to undue the damage that it causes when you don’t put the time or when you think you can change someone. Coach’s often think we can change someone, and so, I hope I’ve made my last mistake, and we don’t go down that path again. I got some safeguards, guardrails in place to holler at me if I consider such things. It leads to greater quality of life”

Andrew Dukowitz

32,503 views • 2 months ago

Interviewer: congrats. "Still We Rise"is one of your main quotes. It's not a coincidence, right? Lewis Hamilton: No. No. Interviewer: And what message would you like to share with your fans? Lewis Hamilton: "Um, to my fans, I have to continue to express my gratitude to them. So, who've been with me since day one, through thick and thin. I mean, last year was the hardest year, but they're supporting me, but them to continue to hold on, to continue to believe in me, remind me of who I am and remind me of what I can do" "I literally wouldn't be standing here today without the fans. Um, the letters, the messages, seeing them on race weekends, the support, them... they've showered me with so much love, and I will never forget that" "You know, I hope that they're happy with these results, that we finally got back. I think perseverance is really the thing. The truth is, you can never give up. You can never give up, and you can never stop believing in yourself, even when everyone around you is not believing in you, or... it's trying to pull you down. You've just got to keep fighting, keep going, and keep giving it everything you've got" 🥹❤️
1:24

Interviewer: congrats. "Still We Rise"is one of your main quotes. It's not a coincidence, right? Lewis Hamilton: No. No. Interviewer: And what message would you like to share with your fans? Lewis Hamilton: "Um, to my fans, I have to continue to express my gratitude to them. So, who've been with me since day one, through thick and thin. I mean, last year was the hardest year, but they're supporting me, but them to continue to hold on, to continue to believe in me, remind me of who I am and remind me of what I can do" "I literally wouldn't be standing here today without the fans. Um, the letters, the messages, seeing them on race weekends, the support, them... they've showered me with so much love, and I will never forget that" "You know, I hope that they're happy with these results, that we finally got back. I think perseverance is really the thing. The truth is, you can never give up. You can never give up, and you can never stop believing in yourself, even when everyone around you is not believing in you, or... it's trying to pull you down. You've just got to keep fighting, keep going, and keep giving it everything you've got" 🥹❤️

LH44(A)

23,882 views • 2 months ago

Google just turned every TikTok, Instagram Reel, and YouTube Short into part of your SEO strategy. Google Search Console can now show you which searches lead people to your social posts, along with clicks, impressions, and average search position. So your video doesn’t have to disappear after two days in the feed. It can keep getting discovered when someone searches for that topic on Google. But there’s a catch. More searchable content also means more average content. So don’t start by asking AI for 50 random ideas. Start with real demand. Go to AnswerThePublic, type in the topic you want to be known for, and look at the questions people are already searching. Let’s say you search “losing weight.” You might see: “Why am I not losing weight?” “Can one cheat day ruin my progress?” “Why do I gain weight on the weekends?” The average video says, “Eat less and exercise more.” But now you can use AI as a thought partner to connect the dots and find a sharper angle: “You’re not failing during the week. You’re undoing it every weekend.” That’s the difference. Find the demand. Add the angle. Make the best answer on the internet.
0:56

Google just turned every TikTok, Instagram Reel, and YouTube Short into part of your SEO strategy. Google Search Console can now show you which searches lead people to your social posts, along with clicks, impressions, and average search position. So your video doesn’t have to disappear after two days in the feed. It can keep getting discovered when someone searches for that topic on Google. But there’s a catch. More searchable content also means more average content. So don’t start by asking AI for 50 random ideas. Start with real demand. Go to AnswerThePublic, type in the topic you want to be known for, and look at the questions people are already searching. Let’s say you search “losing weight.” You might see: “Why am I not losing weight?” “Can one cheat day ruin my progress?” “Why do I gain weight on the weekends?” The average video says, “Eat less and exercise more.” But now you can use AI as a thought partner to connect the dots and find a sharper angle: “You’re not failing during the week. You’re undoing it every weekend.” That’s the difference. Find the demand. Add the angle. Make the best answer on the internet.

Neil Patel

16,427 views • 23 days ago

Introducing Clips - 100% free, open source, agent-native alternative to Loom Unlike Loom, agent's can fully understand Clips just from a URL. Every Clip comes with APIs and metadata for agents to explore their contents. Agents can "see and hear" anything in a Clip - not just transcripts, but everything visually in the video at any timestamp. Easily share bug reports, feedback, analyses, or anything else in a way that you can easily pass to agents to use to improve products, reports, or more. Also unlike Loom, you own the software, so no one can jack up prices on you suddenly like Loom did to us. Clips is made to be customized. The built-in agent can customize its own code, so you can personalize the app to your needs and workflows. This, in my opinion, is the future of software. Open-source, forkable, customizable with agents, to make your own personal version of anything. You can also import Looms just from a URL and upload videos as well. I got so sick of telling people "don't send me feedback as looms, I can't pass those to agents, I need text and images" that I had to just solve this once and for all. There's a free hosted version you can use too, or fork and self host yourself. Will link to both in the replies.
0:35

Introducing Clips - 100% free, open source, agent-native alternative to Loom Unlike Loom, agent's can fully understand Clips just from a URL. Every Clip comes with APIs and metadata for agents to explore their contents. Agents can "see and hear" anything in a Clip - not just transcripts, but everything visually in the video at any timestamp. Easily share bug reports, feedback, analyses, or anything else in a way that you can easily pass to agents to use to improve products, reports, or more. Also unlike Loom, you own the software, so no one can jack up prices on you suddenly like Loom did to us. Clips is made to be customized. The built-in agent can customize its own code, so you can personalize the app to your needs and workflows. This, in my opinion, is the future of software. Open-source, forkable, customizable with agents, to make your own personal version of anything. You can also import Looms just from a URL and upload videos as well. I got so sick of telling people "don't send me feedback as looms, I can't pass those to agents, I need text and images" that I had to just solve this once and for all. There's a free hosted version you can use too, or fork and self host yourself. Will link to both in the replies.

Steve (Builder.io)

232,092 views • 2 months ago

A Scouse Muslim who’s been here all of his life has been in touch with me. He feels the page is bias against Muslims and I have overlooked the good charity work he and his group do. After I explained the concept of the page to him that anyone can post what they want on this page. They come to me, I don’t go to them. It is open to anyone for anything. Now he gets it, and he has sent this in. Total respect to these people. This is the kind of thing I would much rather be promoting from the Muslim community but unfortunately this type of content isn’t sent in often enough. If you want to change the narrative. Only you the reader of this can do that. Insulting people in the comment section changes nothing. Send it in like he did. Give them a follow on Instagram and get involved. The Liverpool Muslim Outreach.
0:41

A Scouse Muslim who’s been here all of his life has been in touch with me. He feels the page is bias against Muslims and I have overlooked the good charity work he and his group do. After I explained the concept of the page to him that anyone can post what they want on this page. They come to me, I don’t go to them. It is open to anyone for anything. Now he gets it, and he has sent this in. Total respect to these people. This is the kind of thing I would much rather be promoting from the Muslim community but unfortunately this type of content isn’t sent in often enough. If you want to change the narrative. Only you the reader of this can do that. Insulting people in the comment section changes nothing. Send it in like he did. Give them a follow on Instagram and get involved. The Liverpool Muslim Outreach.

Around Liverpool

57,521 views • 2 days ago

🔌Today I created a virtual web-based null modem so now you can finally play Quake 1 (from 1996) in multiplayer in the browser with other people online! This week I was able to create a virtual printer that listens on the COM2 port to make a web-based printer work in Windows 3.11, like on you can actually print to, it was one of my ideas when I started but it was always too difficult to make, I had no idea where to start, but this week AI was able to do it! Today I woke up and thought "if we can listen to COM2 via JS, how about listen to COM1 from Quake 1?" then we could forward the COM1 data via Websockets and to another user to create a virtual null modem (A null modem was a cable you'd use to connect two computers to each other directly in the 90s, if you couldn't afford buying a network card you'd use this to network, but you could only play with one other computer not multiple) So I built it, it took about an hour and works! I wanted to get Quake 1 in DOS to work with multiplayer for 2 years but I could never get it to work, but now it did it This project is a good benchmark of how capable AI is becoming, figuring stuff out that humans would take months within an hour You can try it here, I don't know how it will match lots of people but let's see, you have to exit to DOS first (ALT-F4 or FN-Option-F4 on Mac), then type CD games, then CD quake, then quake, then ESC, then Multiplayer, then either Create New Game or Join Game, and use Direct Connect, keep settings same and you're in!
1:35

🔌Today I created a virtual web-based null modem so now you can finally play Quake 1 (from 1996) in multiplayer in the browser with other people online! This week I was able to create a virtual printer that listens on the COM2 port to make a web-based printer work in Windows 3.11, like on you can actually print to, it was one of my ideas when I started but it was always too difficult to make, I had no idea where to start, but this week AI was able to do it! Today I woke up and thought "if we can listen to COM2 via JS, how about listen to COM1 from Quake 1?" then we could forward the COM1 data via Websockets and to another user to create a virtual null modem (A null modem was a cable you'd use to connect two computers to each other directly in the 90s, if you couldn't afford buying a network card you'd use this to network, but you could only play with one other computer not multiple) So I built it, it took about an hour and works! I wanted to get Quake 1 in DOS to work with multiplayer for 2 years but I could never get it to work, but now it did it This project is a good benchmark of how capable AI is becoming, figuring stuff out that humans would take months within an hour You can try it here, I don't know how it will match lots of people but let's see, you have to exit to DOS first (ALT-F4 or FN-Option-F4 on Mac), then type CD games, then CD quake, then quake, then ESC, then Multiplayer, then either Create New Game or Join Game, and use Direct Connect, keep settings same and you're in!

@levelsio

103,163 views • 2 months ago

Ghostty 1.3 is going to have a preview of AppleScript support. All windows, tabs, splits, terminals are exposed via AppleScript. For macOS users, this satisfies some of the most common requests: broadcast commands, automatic layouts, jump to working directory, etc. PR here: I also think this is an incredibly important feature to ship early/now so agentic tools like Claude and Codex can take advantage of a scriptable Ghostty on macOS. I still plan on a generic cross-platform scripting/plugin API (two separate things), but integrating with native features like AppleScript is entirely in scope of the Ghostty mission and like our Shortcuts integration it importantly lets you connect multiple sources since AppleScript can control multiple applications from one script. Note I normally don't ship features so down to the wire with a release, but this one is very isolated in its impact and I'm going to explicit document it in the release notes and website as a preview since I fully expect we're going to iterate a lot on the objects and commands exposed. From a security perspective, Apple already prompts for permission to control different applications so we're covered, but there is a master kill switch you can put in the config, too.
2:42

Ghostty 1.3 is going to have a preview of AppleScript support. All windows, tabs, splits, terminals are exposed via AppleScript. For macOS users, this satisfies some of the most common requests: broadcast commands, automatic layouts, jump to working directory, etc. PR here: I also think this is an incredibly important feature to ship early/now so agentic tools like Claude and Codex can take advantage of a scriptable Ghostty on macOS. I still plan on a generic cross-platform scripting/plugin API (two separate things), but integrating with native features like AppleScript is entirely in scope of the Ghostty mission and like our Shortcuts integration it importantly lets you connect multiple sources since AppleScript can control multiple applications from one script. Note I normally don't ship features so down to the wire with a release, but this one is very isolated in its impact and I'm going to explicit document it in the release notes and website as a preview since I fully expect we're going to iterate a lot on the objects and commands exposed. From a security perspective, Apple already prompts for permission to control different applications so we're covered, but there is a master kill switch you can put in the config, too.

Mitchell Hashimoto

182,058 views • 5 months ago

I finally finished my Rust version of Mario Zechner's (Mario Zechner) excellent Pi Agent, which I made with his blessing and which is called pi_agent_rust. You can get it here: If you're not familiar with Pi, it's a minimalist and extensible agent harness (similar to Claude Code and Codex) and, among other uses, serves as the core agent harness inside the OpenClaw project. I say my Rust "version" instead of "port" because it's really quite different in how it's implemented for it to be called a port. Arguably, the incremental functionality in the implementation was more complex than the rest of the project combined. Still, it provides the same features and functionality as the original, and is proven to be compatible with hundreds of popular extensions to Pi (the conformance harness shows 224 out of 224 extensions working perfectly). But the way it's architected has some major changes. Pi Agent relies on node or bun to provide access to the filesystem and for various other tasks, and that is also how Pi's extension system works. I decided early on that I didn't want to do things that way. Instead, I wanted to integrate that functionality directly into the binary itself; that is, to provide equivalent functionality for everything that would normally be provided by node/bun in the original. I did this for several reasons: one, it's a lot more performant in terms of footprint and latency. On realistic end-to-end large-session workloads (not toy microbenchmarks), pi_agent_rust is now: - 4.95x faster than legacy Node and 2.80x faster than legacy Bun at 1mm-token session scale - 4.32x faster than legacy Node and 2.14x faster than legacy Bun at 5mm-token session scale - ~8x to ~13x lower RSS memory footprint in those same scenarios But the other reason is security and control: by handling everything internally in an end-to-end way, we can do all sorts of clever things to harden the system against insecure or malicious extensions. Those extensions no longer have direct access to the ambient filesystem: they now need to go through pi_agent_rust, and we can analyze extensions carefully before ever running them and also block things that look suspicious at runtime. In practice that means explicit capability-gated hostcalls, with policy/risk/quota enforcement and runtime telemetry/auditability. In order to do all this, I had to effectively build the missing runtime substrate from scratch in Rust, not just translate TypeScript syntax: - define and implement a typed hostcall ABI for extension->host interactions - build native Rust connectors for tool/exec/http/session/ui/events instead of ambient Node/Bun access - implement a compatibility/shim layer so real-world Pi extensions still behave correctly - add capability policy evaluation, runtime risk scoring, per-extension quotas, and audit telemetry on the execution path - wire the whole thing through structured concurrency (asupersync) so cancellation/lifetimes are deterministic and failure handling is explicit - build a conformance + benchmark harness large enough to validate behavior/perf across hundreds of extensions and realistic long-session workloads This was a full re-architecture of the execution model while preserving the Pi workflow and extension ecosystem. And indeed, this aspect of it dwarfs the entire rest of the project in size and complexity. To put hard numbers on that: the extension/runtime/security subsystem alone is now about 86.5k lines of Rust across src/extensions.rs (~48.1k), src/extensions_js.rs (~23.4k), src/extension_dispatcher.rs (~13.4k), and src/extension_index.rs (~1.7k), with roughly 2.5k callable units in just those files. For context, the original Pi coding-agent production code is about 27.4k lines total. So this one subsystem by itself is roughly 3.2x the size of the original harness, which is why calling this a “port” would seriously undersell what had to be built. And on top of that, pi_agent_rust introduces a bunch of genuinely new capabilities beyond the legacy harness, not just a faster core: - Security and enforcement are materially stronger at runtime: capability-gated hostcalls with explicit policy profiles (safe/balanced/permissive), per-extension trust lifecycle (pending -> acknowledged -> trusted -> killed), explicit kill-switch operations, and audited state transitions. - Shell execution mediation is deterministic and argument-aware: rule/feature-based risk scoring plus heredoc AST inspection (dcg_rule_hit, dcg_heredoc_hit) before spawn, instead of relying on coarse deny patterns. - Containment and forensics are first-class: tamper-evident runtime risk ledger tooling (verify/replay/calibrate), unified incident evidence bundles, and forced-compat controls that let you contain issues without disabling the whole extension system. - The extension runtime architecture is native: JS extensions run in embedded QuickJS with typed hostcall boundaries and Rust-native connectors for tool/exec/http/session/ui/events, plus compatibility shims for real-world legacy extensions. - Runtime behavior under load is explicitly engineered: deterministic hostcall reactor mesh, fast-lane vs compat-lane routing, and warm-isolate prewarm handoff for more predictable throughput and latency under contention. - Long-session reliability is upgraded: JSONL v3 sessions with indexed sidecar acceleration and optional SQLite-backed sessions, plus operational controls via --session-durability, --no-migrations, and migrate. - Provider and auth coverage are broader and more operationally explicit: native Anthropic/OpenAI (Chat + Responses)/Gemini/Cohere/Azure/Bedrock/Vertex/Copilot/GitLab plus large OpenAI-compatible routing; pi --list-providers currently shows 90 providers with aliases and required auth env keys. - Auth is not just API keys: OAuth (Anthropic/OpenAI Codex/Gemini CLI/Antigravity/Kimi/Copilot/GitLab plus extension-defined OAuth), AWS credential chains (Bedrock), service-key exchange (SAP AI Core), and bearer-token flows. - Operator tooling is stronger: pi doctor supports scoped checks (config, dirs, auth, shell, sessions, extensions), machine-readable output (--format json|markdown), and safe auto-remediation (--fix). - Extension/package lifecycle workflows are built in: install, remove, update, update-index, search, info, and list. I want to thank Mario for making a great harness and for not telling me to get lost when I asked him if he was OK with me porting it to Rust. I may give him a hard time in jest about not going "full clanker," but that doesn't mean that I don't respect his work a huge amount. PS: There could still be bugs. If you find some, please let me know in GitHub Issues and I'll fix them same day. There's always a tradeoff between perfect and getting stuff out the door and I felt like it was time to release this.
0:38

I finally finished my Rust version of Mario Zechner's (Mario Zechner) excellent Pi Agent, which I made with his blessing and which is called pi_agent_rust. You can get it here: If you're not familiar with Pi, it's a minimalist and extensible agent harness (similar to Claude Code and Codex) and, among other uses, serves as the core agent harness inside the OpenClaw project. I say my Rust "version" instead of "port" because it's really quite different in how it's implemented for it to be called a port. Arguably, the incremental functionality in the implementation was more complex than the rest of the project combined. Still, it provides the same features and functionality as the original, and is proven to be compatible with hundreds of popular extensions to Pi (the conformance harness shows 224 out of 224 extensions working perfectly). But the way it's architected has some major changes. Pi Agent relies on node or bun to provide access to the filesystem and for various other tasks, and that is also how Pi's extension system works. I decided early on that I didn't want to do things that way. Instead, I wanted to integrate that functionality directly into the binary itself; that is, to provide equivalent functionality for everything that would normally be provided by node/bun in the original. I did this for several reasons: one, it's a lot more performant in terms of footprint and latency. On realistic end-to-end large-session workloads (not toy microbenchmarks), pi_agent_rust is now: - 4.95x faster than legacy Node and 2.80x faster than legacy Bun at 1mm-token session scale - 4.32x faster than legacy Node and 2.14x faster than legacy Bun at 5mm-token session scale - ~8x to ~13x lower RSS memory footprint in those same scenarios But the other reason is security and control: by handling everything internally in an end-to-end way, we can do all sorts of clever things to harden the system against insecure or malicious extensions. Those extensions no longer have direct access to the ambient filesystem: they now need to go through pi_agent_rust, and we can analyze extensions carefully before ever running them and also block things that look suspicious at runtime. In practice that means explicit capability-gated hostcalls, with policy/risk/quota enforcement and runtime telemetry/auditability. In order to do all this, I had to effectively build the missing runtime substrate from scratch in Rust, not just translate TypeScript syntax: - define and implement a typed hostcall ABI for extension->host interactions - build native Rust connectors for tool/exec/http/session/ui/events instead of ambient Node/Bun access - implement a compatibility/shim layer so real-world Pi extensions still behave correctly - add capability policy evaluation, runtime risk scoring, per-extension quotas, and audit telemetry on the execution path - wire the whole thing through structured concurrency (asupersync) so cancellation/lifetimes are deterministic and failure handling is explicit - build a conformance + benchmark harness large enough to validate behavior/perf across hundreds of extensions and realistic long-session workloads This was a full re-architecture of the execution model while preserving the Pi workflow and extension ecosystem. And indeed, this aspect of it dwarfs the entire rest of the project in size and complexity. To put hard numbers on that: the extension/runtime/security subsystem alone is now about 86.5k lines of Rust across src/extensions.rs (~48.1k), src/extensions_js.rs (~23.4k), src/extension_dispatcher.rs (~13.4k), and src/extension_index.rs (~1.7k), with roughly 2.5k callable units in just those files. For context, the original Pi coding-agent production code is about 27.4k lines total. So this one subsystem by itself is roughly 3.2x the size of the original harness, which is why calling this a “port” would seriously undersell what had to be built. And on top of that, pi_agent_rust introduces a bunch of genuinely new capabilities beyond the legacy harness, not just a faster core: - Security and enforcement are materially stronger at runtime: capability-gated hostcalls with explicit policy profiles (safe/balanced/permissive), per-extension trust lifecycle (pending -> acknowledged -> trusted -> killed), explicit kill-switch operations, and audited state transitions. - Shell execution mediation is deterministic and argument-aware: rule/feature-based risk scoring plus heredoc AST inspection (dcg_rule_hit, dcg_heredoc_hit) before spawn, instead of relying on coarse deny patterns. - Containment and forensics are first-class: tamper-evident runtime risk ledger tooling (verify/replay/calibrate), unified incident evidence bundles, and forced-compat controls that let you contain issues without disabling the whole extension system. - The extension runtime architecture is native: JS extensions run in embedded QuickJS with typed hostcall boundaries and Rust-native connectors for tool/exec/http/session/ui/events, plus compatibility shims for real-world legacy extensions. - Runtime behavior under load is explicitly engineered: deterministic hostcall reactor mesh, fast-lane vs compat-lane routing, and warm-isolate prewarm handoff for more predictable throughput and latency under contention. - Long-session reliability is upgraded: JSONL v3 sessions with indexed sidecar acceleration and optional SQLite-backed sessions, plus operational controls via --session-durability, --no-migrations, and migrate. - Provider and auth coverage are broader and more operationally explicit: native Anthropic/OpenAI (Chat + Responses)/Gemini/Cohere/Azure/Bedrock/Vertex/Copilot/GitLab plus large OpenAI-compatible routing; pi --list-providers currently shows 90 providers with aliases and required auth env keys. - Auth is not just API keys: OAuth (Anthropic/OpenAI Codex/Gemini CLI/Antigravity/Kimi/Copilot/GitLab plus extension-defined OAuth), AWS credential chains (Bedrock), service-key exchange (SAP AI Core), and bearer-token flows. - Operator tooling is stronger: pi doctor supports scoped checks (config, dirs, auth, shell, sessions, extensions), machine-readable output (--format json|markdown), and safe auto-remediation (--fix). - Extension/package lifecycle workflows are built in: install, remove, update, update-index, search, info, and list. I want to thank Mario for making a great harness and for not telling me to get lost when I asked him if he was OK with me porting it to Rust. I may give him a hard time in jest about not going "full clanker," but that doesn't mean that I don't respect his work a huge amount. PS: There could still be bugs. If you find some, please let me know in GitHub Issues and I'll fix them same day. There's always a tradeoff between perfect and getting stuff out the door and I felt like it was time to release this.

Jeffrey Emanuel

135,570 views • 6 months ago

q: there are photos everywhere of you and the bee costume. can you talk about the bees becoming part of your indian wells story? carlitos: “it’s a funny story. what happened here in indian wells was the first time i saw something like that on a tennis court. after that a lot of people started coming to the tournament wearing bee costumes. it’s funny to see them in the stands supporting with those costumes. the other day i spent a little time with them after the match, and today they brought me a bee costume and kept saying ‘wear it, wear it.’ so i had to do it for them. they were cheering from the first point until the last one, so i felt like i had to. but yeah, it’s funny.”
1:12

q: there are photos everywhere of you and the bee costume. can you talk about the bees becoming part of your indian wells story? carlitos: “it’s a funny story. what happened here in indian wells was the first time i saw something like that on a tennis court. after that a lot of people started coming to the tournament wearing bee costumes. it’s funny to see them in the stands supporting with those costumes. the other day i spent a little time with them after the match, and today they brought me a bee costume and kept saying ‘wear it, wear it.’ so i had to do it for them. they were cheering from the first point until the last one, so i felt like i had to. but yeah, it’s funny.”

f #7inMelbourne

34,891 views • 5 months ago

Factory CEO Matan Grinberg says all companies go through three phases of AI adoption: "Phase Zero is reluctance to adopt because developers have had the same workflows for the last 20 years and they might not want to change it." "Phase One is throw the kitchen sink and use as much AI as possible. Whatever you do, just change your workflows." "Phase Two is that people are now adopting the tech, they're proving that they can change their behavior, and use these tools. But now we need to make sure we're actually doing it efficiently and effectively. Uber and Meta are examples of this." "Yes, you'll overspend in Phase One, but the point is to get to Phase Two where then you're actually really efficient on a per-token basis, moving the needle for the business and delivering software faster."
0:56

Factory CEO Matan Grinberg says all companies go through three phases of AI adoption: "Phase Zero is reluctance to adopt because developers have had the same workflows for the last 20 years and they might not want to change it." "Phase One is throw the kitchen sink and use as much AI as possible. Whatever you do, just change your workflows." "Phase Two is that people are now adopting the tech, they're proving that they can change their behavior, and use these tools. But now we need to make sure we're actually doing it efficiently and effectively. Uber and Meta are examples of this." "Yes, you'll overspend in Phase One, but the point is to get to Phase Two where then you're actually really efficient on a per-token basis, moving the needle for the business and delivering software faster."

TBPN

25,747 views • 4 months ago

LLM Wikis are being slept on. I argue that creating knowledge bases with LLMs or coding agents is one of the most valuable applications of AI today. It's about being intentional in building and scaling your intelligence stack. To showcase this, I wanted to share an LLM Wiki I have built over the last couple of months. It's called PaperWiki, and I use it across all my research workflows, along with my research agents. In fact, I also use it to curate papers I share with my communities, newsletter, and on X. The PaperWiki is updated regularly with automations, so I basically have agents on a loop maintaining it. All the entries are ingested from different sources and stored in a vault (Obsidian) and further indexed using qmd. And then further presented via an HTML artifact. So all of it is easily accessible to all my agents and easily searchable through full-text search and rich semantic search. The structure of the wiki has proven significantly useful to start interesting and exciting cutting-edge research projects with my research agents (from building tiny and more efficient gpt/difussion llms to building out SoTA harnesses and memory systems). It turns out that agents love markdown files and can more easily navigate the papers given the rich metadata structure of the wiki. I am just getting started on this, but it's clear to me that we should all be experimenting with LLM Wikis. Here's why: Building LLM knowledge bases gets you into the habit of leveraging AI outputs in all kinds of creative ways. It's the good kind of tokenmaxxing we should all be pushing for. LLM Wikis can be maintained automatically in a loop. I use an automation that updates the wiki every day based on papers I curate. The curation is another automation I run in a loop (with a bit of human in the loop), so I get to build on all my previous knowledge and expertise, and all of it compounds the deeper the integration/layers. One interesting result of this process is that I feel like I can better spot high-quality papers and remove noise more easily. Social media could never solve that. And most paper aggregators use metrics I simply don't trust. I like that agents can help with the noise vs. signal problem. This is important for research. Lots of people consider agents to produce mostly slop. But it doesn't have to be that way. Careful curations, prompts, automations, verifiers, and human-in-the-loop can produce some astonishing results. And you really don't need frontier models for this. I use a combination of frontier models (opus-4.8) and open-weight models (deepseek-v4-flash) to maintain this. An exciting future work (we are working on this DAIR.AI) is to tune specialized models on top of this to allow LLMs to quickly understand cutting-edge research ideas and can better conceptualize research strategies that further accelerate scientific research agents. I plan to open-source a bunch of this work, including the artifact, but this is currently work in progress, and I was excited to share some thoughts as I continue working on it. Sharing more as I go. Stay tuned!
0:27

LLM Wikis are being slept on. I argue that creating knowledge bases with LLMs or coding agents is one of the most valuable applications of AI today. It's about being intentional in building and scaling your intelligence stack. To showcase this, I wanted to share an LLM Wiki I have built over the last couple of months. It's called PaperWiki, and I use it across all my research workflows, along with my research agents. In fact, I also use it to curate papers I share with my communities, newsletter, and on X. The PaperWiki is updated regularly with automations, so I basically have agents on a loop maintaining it. All the entries are ingested from different sources and stored in a vault (Obsidian) and further indexed using qmd. And then further presented via an HTML artifact. So all of it is easily accessible to all my agents and easily searchable through full-text search and rich semantic search. The structure of the wiki has proven significantly useful to start interesting and exciting cutting-edge research projects with my research agents (from building tiny and more efficient gpt/difussion llms to building out SoTA harnesses and memory systems). It turns out that agents love markdown files and can more easily navigate the papers given the rich metadata structure of the wiki. I am just getting started on this, but it's clear to me that we should all be experimenting with LLM Wikis. Here's why: Building LLM knowledge bases gets you into the habit of leveraging AI outputs in all kinds of creative ways. It's the good kind of tokenmaxxing we should all be pushing for. LLM Wikis can be maintained automatically in a loop. I use an automation that updates the wiki every day based on papers I curate. The curation is another automation I run in a loop (with a bit of human in the loop), so I get to build on all my previous knowledge and expertise, and all of it compounds the deeper the integration/layers. One interesting result of this process is that I feel like I can better spot high-quality papers and remove noise more easily. Social media could never solve that. And most paper aggregators use metrics I simply don't trust. I like that agents can help with the noise vs. signal problem. This is important for research. Lots of people consider agents to produce mostly slop. But it doesn't have to be that way. Careful curations, prompts, automations, verifiers, and human-in-the-loop can produce some astonishing results. And you really don't need frontier models for this. I use a combination of frontier models (opus-4.8) and open-weight models (deepseek-v4-flash) to maintain this. An exciting future work (we are working on this DAIR.AI) is to tune specialized models on top of this to allow LLMs to quickly understand cutting-edge research ideas and can better conceptualize research strategies that further accelerate scientific research agents. I plan to open-source a bunch of this work, including the artifact, but this is currently work in progress, and I was excited to share some thoughts as I continue working on it. Sharing more as I go. Stay tuned!

elvis

55,566 views • 1 month ago