RSS Parrot

BETA

🦜 Mario Zechner / @badlogicgames

@nitter.net.badlogicgames@rss-parrot.net

I'm an automated parrot! I relay a website's RSS feed to the Fediverse. Every time a new post appears in the feed, I toot about it. Follow me to get all new posts in your Mastodon timeline! Brought to you by the RSS Parrot.

---

Twitter feed for: @badlogicgames. Generated by https://nitter.net

Your feed and you don't want it here? Just e-mail the birb.

Site URL: nitter.net/badlogicgames

Feed URL: nitter.net/badlogicgames/rss

Posts: 98

Followers: 1

RT by @badlogicgames: This is also why things like SDK design are hard for agents. When you start with zero, there is no implicit context that prunes that decision space of the universe of “bad” choices for the non functional requirements. It’s not that the agents don’t know the good choices, they absolutely have that knowledge! It’s largely that we don’t eval the things to have super strong opinions. No agent today will refuse to write Elixir or C, even though I would never choose those two in a professional context for the long term maintainability of the software by teams of varying expertise and seniority.

https://nitter.net/_lopopolo/status/2051411695873216842#m

Published: May 4, 2026 21:20

This is also why things like SDK design are hard for agents. When you start with zero, there is no implicit context that prunes that decision space of the universe of “bad” choices for the non functional requirements. It’s not that the agents don’t know…

RT by @badlogicgames: Re the best spec is the code: the spec for symphony was distilled out of the actual codebase for the bespoke version embedded in our monorepo. From symphony already existing, in a loop: “agent 1, create a spec from the existing impl”, “agent 2 take the spec and implement”, “agent 3 look at existing impl, spec, and new impl and refine the spec” In general there are thousands of small little decisions that go into producing high quality output (of any artifact, not just diffs). The easiest way to write those down in text form happens to be distilling them out of things that you agree are good. We don’t know yet the best way to tell the machines the full set of instructions. This is also why plan mode is never what you want—your instructions given via the plan are likely mis-specified and under-specified in ways that matter a lot.

https://nitter.net/_lopopolo/status/2051410617937125508#m

Published: May 4, 2026 21:16

Re the best spec is the code: the spec for symphony was distilled out of the actual codebase for the bespoke version embedded in our monorepo. From symphony already existing, in a loop: “agent 1, create a spec from the existing impl”, “agent 2 take the…

recommended reading!

https://nitter.net/badlogicgames/status/2051398208920691116#m

Published: May 4, 2026 20:27

recommended reading! Leonie (@helloiamleonie) Spent the weekend crossing one thing off my "to learn" list: GRPO In this blog, we walk through: • What is GRPO and how does it work • Fine-tune @liquidai's LFM2.5-1.2B-Instruct • using @UnslothAI and…

we need more euromemes like this. someone do one on public/private partnerships. go!

https://nitter.net/badlogicgames/status/2051394568088600953#m

Published: May 4, 2026 20:12

we need more euromemes like this. someone do one on public/private partnerships. go! Infornomics (@infornomics) Eurochad — https://nitter.net/infornomics/status/2051380140479848934#m

recommended reading. some very interesting replies in this thread.

https://nitter.net/badlogicgames/status/2051393183242752451#m

Published: May 4, 2026 20:07

recommended reading. some very interesting replies in this thread. George (@odysseus0z) Things I don't understand: - AI Researchers feel like they will soon be useless because auto research is near - AI still can't design harness SDK/frameworks better…

People of http://pi.dev. New release. Teary eyed. Clankolas is growing up so fast. One of these days I might not even have to review all of his commits anymore.

https://nitter.net/badlogicgames/status/2051392308684206174#m

Published: May 4, 2026 20:03

People of pi.dev. New release. Teary eyed. Clankolas is growing up so fast. One of these days I might not even have to review all of his commits anymore. Armin Ronacher ⇌ (@mitsuhiko) People of pi. The issue tracker might take a vacation on the way to…

RT by @badlogicgames: Stole the loop shape from @GeoffreyHuntley, the harness from @badlogicgames, and a lot of the extension instincts from watching @nicopreme cook. This one's mine, but only barely. https://github.com/srinitude/pi-until-done

https://nitter.net/srinitude/status/2051384364173312174#m

Published: May 4, 2026 19:31

Stole the loop shape from @GeoffreyHuntley, the harness from @badlogicgames, and a lot of the extension instincts from watching @nicopreme cook. This one's mine, but only barely. github.com/srinitude/pi-unti…

RT by @badlogicgames: I’ve read a lot of code in my life So now the latent space in my brain tingles at the smallest sign of slop

https://nitter.net/Infoxicador/status/2051383171178963417#m

Published: May 4, 2026 19:27

I’ve read a lot of code in my life So now the latent space in my brain tingles at the smallest sign of slop

RT by @badlogicgames: Today we're open-sourcing `deepsec`: a security harness powered by coding agents. We've been testing it for a few months on our internal code bases as well as open-source applications from customers and partners. For the latter group we have privately shared the results, so issues can be fixed. - It actually works. I recommend giving it a try. The dream of Mythos in CLI-form. - You can run it on your laptop with your existing claude or codex subscription. - For large repos it can take a very long time to run. For this it supports fanout to worker sandboxes. I've been running it on 1000 cores+ to get through a lot of code quickly

https://nitter.net/cramforce/status/2051382563613397434#m

Published: May 4, 2026 19:24

Today we're open-sourcing `deepsec`: a security harness powered by coding agents. We've been testing it for a few months on our internal code bases as well as open-source applications from customers and partners. For the latter group we have privately…

RT by @badlogicgames: Eurochad

https://nitter.net/infornomics/status/2051380140479848934#m

Published: May 4, 2026 19:15

Eurochad Mario Zechner (@badlogicgames) hi, i'm a sole proprietor/founder in Austria and i earn many many multiples of what i'd earn as an employee, despite "predatory income tax". in fact, i opt out of the many tax optimizations i could use because i…

RT by @badlogicgames: What if everyone built their own product instead of pretending GitHub is an easy problem

https://nitter.net/zeeg/status/2051372422583226797#m

Published: May 4, 2026 18:44

What if everyone built their own product instead of pretending GitHub is an easy problem

RT by @badlogicgames: People of pi. The issue tracker might take a vacation on the way to finish the great refactor, but we're still watching and fixing. Biggest changes are SSE downgrade for when websockets fail, rendering improvements to the read tool and some perf fixes for bash.

https://nitter.net/mitsuhiko/status/2051370264269603204#m

Published: May 4, 2026 18:35

People of pi. The issue tracker might take a vacation on the way to finish the great refactor, but we're still watching and fixing. Biggest changes are SSE downgrade for when websockets fail, rendering improvements to the read tool and some perf fixes for…

being a capitalist and the same time liking humans seems to be an entirely foreign concept to large parts of this clusterfuck nazi bar. love it!

https://nitter.net/badlogicgames/status/2051357677234602356#m

Published: May 4, 2026 17:45

being a capitalist and the same time liking humans seems to be an entirely foreign concept to large parts of this clusterfuck nazi bar. love it!

RT by @badlogicgames: This is a very good model in Amp.

https://nitter.net/sqs/status/2051346726465028532#m

Published: May 4, 2026 17:02

This is a very good model in Amp. Nicolay Gerold (@nicolaygerold) It's finally here. `deep` is now using GPT 5.5 in @AmpCode. — https://nitter.net/nicolaygerold/status/2051323925640937793#m

RT by @badlogicgames: i'd rather lose $100,000 a month with a team of 10 i feel sorry for you if you don't understand this

https://nitter.net/thdxr/status/2051344387146137938#m

Published: May 4, 2026 16:53

i'd rather lose $100,000 a month with a team of 10 i feel sorry for you if you don't understand this jack friks (@jackfriks) would you rather make $100,000 a month with a team of 10 or $30,000 solo? i would and have chosen the second option but curious…

RT by @badlogicgames: still don't how @openclaw works? this one's for you. @steipete hope i could do justice to it and you'll also understand how pi by @badlogicgames fits into the openclaw architecture.

https://nitter.net/GargEtisha/status/2051337786352119900#m

Published: May 4, 2026 16:26

still don't how @openclaw works? this one's for you. @steipete hope i could do justice to it and you'll also understand how pi by @badlogicgames fits into the openclaw architecture. Etisha Garg (@GargEtisha) x.com/i/article/204832569915… —…

me and @mitsuhiko as russian trolls and bots come to our bespoke, handcrafted, sundried "europe fuck yeah" threads.

https://nitter.net/badlogicgames/status/2051331009837174828#m

Published: May 4, 2026 15:59

me and @mitsuhiko as russian trolls and bots come to our bespoke, handcrafted, sundried "europe fuck yeah" threads.

recommended reading.

https://nitter.net/badlogicgames/status/2051305577721086346#m

Published: May 4, 2026 14:18

recommended reading. Armin Ronacher ⇌ (@mitsuhiko) I analyzed my coding sessions and on the text interactions some words stand out. And well, they also show up on Google Trends as spiking. Oh and so much slop in my Twitter mentions and on GitHub. Thus…

RT by @badlogicgames: I like to complain as much as the next person, but no amount of me blogging or complaining on Twitter is going to change anything. Doing something about it is — so if nothing else, we're going to try to build one more kick ass company here. https://x.com/badlogicgames/status/2051289288776540282?s=20

https://nitter.net/mitsuhiko/status/2051294544801464379#m

Published: May 4, 2026 13:35

I like to complain as much as the next person, but no amount of me blogging or complaining on Twitter is going to change anything. Doing something about it is — so if nothing else, we're going to try to build one more kick ass company here.…

R to @badlogicgames: and if you as a startup founder are unable to afford IKEA in Austria, then that's a perfect signal for me as an angel investor to stay the fuck away from your "company". you are clearly not fit to run a company, irrespective of tax regime. but as i said, still happy to finance the safety net you'll enjou after your failure with my taxes!

https://nitter.net/badlogicgames/status/2051290046217466123#m

Published: May 4, 2026 13:17

and if you as a startup founder are unable to afford IKEA in Austria, then that's a perfect signal for me as an angel investor to stay the fuck away from your "company". you are clearly not fit to run a company, irrespective of tax regime. but as i said,…

hi, i'm a sole proprietor/founder in Austria and i earn many many multiples of what i'd earn as an employee, despite "predatory income tax". in fact, i opt out of the many tax optimizations i could use because i like having good schools and as high a standard of living as possible for everyone. the great thing about the EU is that you can just live under any tax regime you like in any of the 27 member states. it's all about trade offs. if poland works for you, fantastic! go build there. and if i may add one more thing: if the CEO of a startup, especially pre-revenue, lives "barely any better than a regular employee" then the system works as intended. fact of the matter is most startups are bad. you are not special because you are trying out a shit idea and fail. but i'll happily pay taxes so you can try your shit idea, fail, and can still live.

https://nitter.net/badlogicgames/status/2051289288776540282#m

Published: May 4, 2026 13:14

hi, i'm a sole proprietor/founder in Austria and i earn many many multiples of what i'd earn as an employee, despite "predatory income tax". in fact, i opt out of the many tax optimizations i could use because i like having good schools and as high a…

I have that too! It's very useful: https://gist.github.com/badlogic/563f245975444dbeedd1a93de95a5e92

https://nitter.net/badlogicgames/status/2051278108532109573#m

Published: May 4, 2026 12:29

I have that too! It's very useful: gist.github.com/badlogic/563… Geoff Goodman (@filearts) My first pi extension: /feedback. For the novels that models tend to write. It writes the last msg to a .md in your session and opens it in $EDITOR. If you…

have no allegiance. be ruthless in diversifying your tool providers.

https://nitter.net/badlogicgames/status/2051243977127403656#m

Published: May 4, 2026 10:14

have no allegiance. be ruthless in diversifying your tool providers. Tommy Falkowski (@TommyFalkowski) Anthropic: You are forbidden to use anything other than our shitty client if you want to use the Claude Code subscription. We don't want any of your…

RT by @badlogicgames: Anthropic: You are forbidden to use anything other than our shitty client if you want to use the Claude Code subscription. We don't want any of your third party harnesses touching our precious Claude. OpenAI: You can use whatever you want. But know that all your third party harnesses are gonna be obsolete soon anyways, since we're gonna steamroll you. So you might as well just use our tools from the get go. I don't like either one of those...

https://nitter.net/TommyFalkowski/status/2051218993696940487#m

Published: May 4, 2026 08:34

Anthropic: You are forbidden to use anything other than our shitty client if you want to use the Claude Code subscription. We don't want any of your third party harnesses touching our precious Claude. OpenAI: You can use whatever you want. But know that…

RT by @badlogicgames: Get ready to hit your goals (and usage limits) Introducing pi-goal, pi-goal is a pi implementation for the recent /goal command in codex, does the exact same thing. pi-goal adds a /goal command and goal tools so @badlogicgames's pi can keep working toward a long-running objective until the goal is complete, paused, cleared, or token-budget-limited.

https://nitter.net/micLivs/status/2051183152798810349#m

Published: May 4, 2026 06:12

Get ready to hit your goals (and usage limits) Introducing pi-goal, pi-goal is a pi implementation for the recent /goal command in codex, does the exact same thing. pi-goal adds a /goal command and goal tools so @badlogicgames's pi can keep working toward…

i'm equal parts scared and excited.

https://nitter.net/badlogicgames/status/2051061419710226903#m

Published: May 3, 2026 22:08

i'm equal parts scared and excited. Peter Steinberger 🦞 (@steipete) Seems I have to build all the tooling for the future of software myself. With Claws and Tokens! — https://nitter.net/steipete/status/2051025224708079737#m

recommended viewing. it is a very, very good talk.

https://nitter.net/badlogicgames/status/2051052748016140776#m

Published: May 3, 2026 21:34

recommended viewing. it is a very, very good talk. Lucas Meijer (@lucasmeijer) This talk had 60ish folks present. It reached 60k+ people online. Most material was cut to hit 20min budget. Looking forward to do a v2 somewhere soon. …

oops, i did it again end goal: get a shitty coding agent reference into every single forking company property.

https://nitter.net/badlogicgames/status/2051052206313349475#m

Published: May 3, 2026 21:32

oops, i did it again end goal: get a shitty coding agent reference into every single forking company property. Mario Zechner (@badlogicgames) plan 1. become friends with forking company people 2. show them the MIT licensed pi OSS project, have em…

RT by @badlogicgames: Built an extension for @badlogicgames's Pi coding agent that lets you create @excalidraw diagrams straight from your Pi. It's based on the official mcp and works just the way I like it. Have fun https://github.com/kostyay/pi-k-excalidraw

https://nitter.net/gr1m0/status/2051041334434156591#m

Published: May 3, 2026 20:48

Built an extension for @badlogicgames's Pi coding agent that lets you create @excalidraw diagrams straight from your Pi. It's based on the official mcp and works just the way I like it. Have fun github.com/kostyay/pi-k-exca… Video

R to @badlogicgames: Past 4 weeks I was spending 5-8 hours a day just triaging. I can't do both the refactor and the triaging. I'll be back.

https://nitter.net/badlogicgames/status/2051031566197944810#m

Published: May 3, 2026 20:10

Past 4 weeks I was spending 5-8 hours a day just triaging. I can't do both the refactor and the triaging. I'll be back.

People of http://pi.dev. I'm heads down in the big refactor (see bigrefactor branch). To have any chance of finishing it, I must pause issue triage for the next 2 weeks. All issues filed during that period will not be reviewed. In case of emergency -> Discord

https://nitter.net/badlogicgames/status/2051031563853312468#m

Published: May 3, 2026 20:10

People of pi.dev. I'm heads down in the big refactor (see bigrefactor branch). To have any chance of finishing it, I must pause issue triage for the next 2 weeks. All issues filed during that period will not be reviewed. In case of emergency -> Discord

R to @badlogicgames: smilarly great signals: - managing agents is like managing a team of humans - but i have review agents - spec is all you need

https://nitter.net/badlogicgames/status/2050972930645819826#m

Published: May 3, 2026 16:17

smilarly great signals: - managing agents is like managing a team of humans - but i have review agents - spec is all you need

i actually don't want this "but you don't review compiler output either" meme to die. it's the perfect signal for being immediately able to ignore someone in this space.

https://nitter.net/badlogicgames/status/2050972668178944477#m

Published: May 3, 2026 16:16

i actually don't want this "but you don't review compiler output either" meme to die. it's the perfect signal for being immediately able to ignore someone in this space. solst/ICE of Astarte (@IceSolst) Interesting article on treating agent output like…

RT by @badlogicgames: the first thing i do before ever digging into a terminal bug (and esp before opening an issue or pr) is make sure it’s reproducible outside of tmux (it’s often not)

https://nitter.net/dillon_mulroy/status/2050969068639834321#m

Published: May 3, 2026 16:01

the first thing i do before ever digging into a terminal bug (and esp before opening an issue or pr) is make sure it’s reproducible outside of tmux (it’s often not) Simon Klee (@simonklee) If you're using tmux and software is buggy, then there is a…

how gpt 5.5 thinks it should do a "ok, if the file exists, load it, otherwise do a different thing" this is absolutely demented.

https://nitter.net/badlogicgames/status/2050895498630791526#m

Published: May 3, 2026 11:09

how gpt 5.5 thinks it should do a "ok, if the file exists, load it, otherwise do a different thing" this is absolutely demented.

many bad takes in here by yours truely.

https://nitter.net/badlogicgames/status/2050865213914661183#m

Published: May 3, 2026 09:09

many bad takes in here by yours truely. Johanna Pirker (@JoeyPrink) Die neue Folge von #einsnull ist online! Diesmal endlich wieder mit @badlogicgames mit einem Update an welchen 712 Projekten er seit der letzten Folge gearbeitet hat ... (inkl. AI…

if that's true, then that also means using the OpenAI API with custom tools for your agentic enterprise workflow will be shite. unless the model served via API is != the model served to the native harness. neither is agreat advertising.

https://nitter.net/badlogicgames/status/2050854755270873391#m

Published: May 3, 2026 08:27

if that's true, then that also means using the OpenAI API with custom tools for your agentic enterprise workflow will be shite. unless the model served via API is != the model served to the native harness. neither is agreat advertising. Ryan Lopopolo…

recommended reading.

https://nitter.net/badlogicgames/status/2050850265297760320#m

Published: May 3, 2026 08:09

recommended reading. Thorsten Ball (@thorstenball) Without any intro this week: a new Joy & Curiosity! registerspill.thorstenball.c… — https://nitter.net/thorstenball/status/2050830690401354149#m

RT by @badlogicgames: Added a fixed-position editor to Pi that stays put while the chat can stream and scroll above it. Also added configurable shortcuts for jumping between user/LLM messages, to the bottom of chat, and to the start/end of a multiline editor. pi install npm:pi-powerline-footer https://github.com/nicobailon/pi-powerline-footer

https://nitter.net/nicopreme/status/2050808090162073733#m

Published: May 3, 2026 05:22

Added a fixed-position editor to Pi that stays put while the chat can stream and scroll above it. Also added configurable shortcuts for jumping between user/LLM messages, to the bottom of chat, and to the start/end of a multiline editor. pi install…

RT by @badlogicgames: Just built gpt-image. An extension for @badlogicgames 's pi coding agent. Lets you generate images(with gpt-image-2) directly inside pi using your existing Plus/Pro subscription. No separate OpenAI key required. Includes saved artifacts, session-local manifests, configurable size/quality/format and my favorite terminal carousel browsing! try - https://pi.dev/packages/@georgetsouvaltzis/pi-gpt-image

https://nitter.net/gitsouv/status/2050702156710178925#m

Published: May 2, 2026 22:21

Just built gpt-image. An extension for @badlogicgames 's pi coding agent. Lets you generate images(with gpt-image-2) directly inside pi using your existing Plus/Pro subscription. No separate OpenAI key required. Includes saved artifacts, session-local…

recommended viewing. kind of tired of transformers tbh.

https://nitter.net/badlogicgames/status/2050685104335274090#m

Published: May 2, 2026 21:13

recommended viewing. kind of tired of transformers tbh. Yann LeCun (@ylecun) piped.video/kYkIdXwW2AE?si=hV2A… — https://nitter.net/ylecun/status/2050668477845798935#m

RT by @badlogicgames: @simonw Hi! Just read your post on DS4, please note that you can run my GGUF 2-bit quantized right now if you wish: https://github.com/antirez/llama.cpp-deepseek-v4-flash And a vertical ds4 inference engine is coming soon, I'm on it. https://www.youtube.com/watch?v=todMmp6AGCE

https://nitter.net/antirez/status/2050628563380920509#m

Published: May 2, 2026 17:28

@simonw Hi! Just read your post on DS4, please note that you can run my GGUF 2-bit quantized right now if you wish: github.com/antirez/llama.cpp… And a vertical ds4 inference engine is coming soon, I'm on it. piped.video/watch?v=todMmp6A…

RT by @badlogicgames: I made a new extension `pi-daytona` that let's you run Pi Coding Agent with all file access tools replaced with those that only access your daytona sandbox. @daytonaio @badlogicgames https://github.com/richardanaya/pi-daytona

https://nitter.net/richardanaya/status/2050622053343240690#m

Published: May 2, 2026 17:02

I made a new extension `pi-daytona` that let's you run Pi Coding Agent with all file access tools replaced with those that only access your daytona sandbox. @daytonaio @badlogicgames github.com/richardanaya/pi-d…

if you actually click through the link you get this. as a non-native speaker i think the closest translation of my thoughts is: absofuckinglutely not, wtf

https://nitter.net/badlogicgames/status/2050586352861814886#m

Published: May 2, 2026 14:40

if you actually click through the link you get this. as a non-native speaker i think the closest translation of my thoughts is: absofuckinglutely not, wtf Mario Zechner (@badlogicgames) asked gpt 5.5 to figure out an ffmpeg arm build for me. i'm no a…

R to @badlogicgames: ok, how do i instruction fine tune gpt-2 in typescript. how hard can it be. https://x.com/badlogicgames/status/2050025762070180005

https://nitter.net/badlogicgames/status/2050556262249623641#m

Published: May 2, 2026 12:41

ok, how do i instruction fine tune gpt-2 in typescript. how hard can it be. nitter.net/badlogicgames/status/2… Mario Zechner (@badlogicgames) felt cute, did some @karpathy style cozy coding. now i can run GPT 2 124M in pure TypeScript at 7 tps. played…

guess it's time to build my own model with spit and duct tape as well now. what a time to be alive ... ridonculous.

https://nitter.net/badlogicgames/status/2050555869914427553#m

Published: May 2, 2026 12:39

guess it's time to build my own model with spit and duct tape as well now. what a time to be alive ... ridonculous.

asked gpt 5.5 to figure out an ffmpeg arm build for me. i'm no a bad hacker according to gpt 5.5. @thsottiaux i'll take goblins over this, please.

https://nitter.net/badlogicgames/status/2050550959390982564#m

Published: May 2, 2026 12:20

asked gpt 5.5 to figure out an ffmpeg arm build for me. i'm no a bad hacker according to gpt 5.5. @thsottiaux i'll take goblins over this, please.

oh no, not them too ...

https://nitter.net/badlogicgames/status/2050506296894378026#m

Published: May 2, 2026 09:22

oh no, not them too ... Armin Ronacher ⇌ (@mitsuhiko) Did OpenAI change something here? Because this is getting really annoying. — https://nitter.net/mitsuhiko/status/2050504490327949717#m

RT by @badlogicgames: the secrets that Big Lab has, and the open community doesn’t, are in many cases quite ripe for community discovery. and you can experiment productively with shockingly little compute.

https://nitter.net/willccbb/status/2050392754312839394#m

Published: May 2, 2026 01:51

the secrets that Big Lab has, and the open community doesn’t, are in many cases quite ripe for community discovery. and you can experiment productively with shockingly little compute.

RT by @badlogicgames: yeah you can set CLAUDE_CODE_SIMPLE=1 claude to true if you want our take on the simplest harness dont think this is better but play around with it!

https://nitter.net/trq212/status/2050391489612747186#m

Published: May 2, 2026 01:46

yeah you can set CLAUDE_CODE_SIMPLE=1 claude to true if you want our take on the simplest harness dont think this is better but play around with it!

People of http://pi.dev. As a weekend gift, we added @XiaomiMiMo Token Plan as a first class provider. I also made some breaking changes for the better. If you have custom providers and models, point pi at the changelog so it can fix them up for you. This will be a recuring theme in the coming days and weeks. We'll get through it together.

https://nitter.net/badlogicgames/status/2050365955629301891#m

Published: May 2, 2026 00:05

People of pi.dev. As a weekend gift, we added @XiaomiMiMo Token Plan as a first class provider. I also made some breaking changes for the better. If you have custom providers and models, point pi at the changelog so it can fix them up for you. This will…

"look what they need to do to match a fraction of our power" write to file and force model to read it. tsk tsk :D

https://nitter.net/badlogicgames/status/2050334447082344932#m

Published: May 1, 2026 22:00

"look what they need to do to match a fraction of our power" write to file and force model to read it. tsk tsk :D Dan Bachelder (@BachelderDan) I have shamelessly stolen pi-diff-review from @badlogicgames and made it work for pi, @claudeai and…

RT by @badlogicgames: I have shamelessly stolen pi-diff-review from @badlogicgames and made it work for pi, @claudeai and @OpenAI codex. Introducing slop-review. https://github.com/dbachelder/slop-review

https://nitter.net/BachelderDan/status/2050333242734293067#m

Published: May 1, 2026 21:55

I have shamelessly stolen pi-diff-review from @badlogicgames and made it work for pi, @claudeai and @OpenAI codex. Introducing slop-review. github.com/dbachelder/slop-r…

RT by @badlogicgames: what if i told you... computer use can be faster on local models moondream3 with its photon update today that gives it mac support can see your screen and use it with 1s latency, ty @vikhyatk here we have whisper+qwen+moondream triple model pipeline working offline flawlessly

https://nitter.net/mayfer/status/2050323883950313980#m

Published: May 1, 2026 21:18

what if i told you... computer use can be faster on local models moondream3 with its photon update today that gives it mac support can see your screen and use it with 1s latency, ty @vikhyatk here we have whisper+qwen+moondream triple model pipeline…

we are getting there folks. slow and steady.

https://nitter.net/badlogicgames/status/2050295536834490493#m

Published: May 1, 2026 19:25

we are getting there folks. slow and steady. Daniel van Strien (@vanstriendaniel) Can an open-weight coding agent + harness match Claude Code at training a domain-specific model? Same one-line prompt. ~13 min e2e. Pushed to @huggingface. Pi +…

ok, i already posted this but holy shit it's built on pi?! https://github.com/withastro/flue/blob/main/packages/sdk/src/session.ts#L3 this makes me super happy!

https://nitter.net/badlogicgames/status/2050288716896247887#m

Published: May 1, 2026 18:58

ok, i already posted this but holy shit it's built on pi?! github.com/withastro/flue/bl… this makes me super happy! fks (@FredKSchott) Introducing Flue — The First Agent Harness Framework Flue is a TypeScript framework for building the next…

this is really really nice and i will steal from it ruthlessly!

https://nitter.net/badlogicgames/status/2050287678537535913#m

Published: May 1, 2026 18:54

this is really really nice and i will steal from it ruthlessly! fks (@FredKSchott) Introducing Flue — The First Agent Harness Framework Flue is a TypeScript framework for building the next generation of agents, designed around a built-in agent harness.…

RT by @badlogicgames: hard to believe we only launched big models on workers ai ~1.5 months ago. the craziest part about making models more efficient is that it's better for our users (faster and cheaper), but also better for us as an inference provider. we effectively reduced our hardware costs through software optimizations. this sounds easy but is really hard in practice - there is so much to do in this space, and so much we're learning and patching as we go. as we build the foundation for serving big models, the new models that come out inherit our efficient architecture too. if you want to join the challenge - we're hiring if you want to consume these models and don't want to be an MLE/SRE, happy to chat

https://nitter.net/michellechen/status/2050272995524939887#m

Published: May 1, 2026 17:55

hard to believe we only launched big models on workers ai ~1.5 months ago. the craziest part about making models more efficient is that it's better for our users (faster and cheaper), but also better for us as an inference provider. we effectively…

beautiful! we also had a PR but i had to sadly decline it for the time being, as this uses shiki, which would break a lot of existing themes and extensions. will revisit in the future after the great refactor :)

https://nitter.net/badlogicgames/status/2050261468923281597#m

Published: May 1, 2026 17:10

beautiful! we also had a PR but i had to sadly decline it for the time being, as this uses shiki, which would break a lot of existing themes and extensions. will revisit in the future after the great refactor :) Matt Leong (@matt_leong) …

RT by @badlogicgames: this tab ordering tells me everything I need to know about GitHub’s actual product priorities for the next year

https://nitter.net/vboykis/status/2050248703139582238#m

Published: May 1, 2026 16:19

this tab ordering tells me everything I need to know about GitHub’s actual product priorities for the next year

RT by @badlogicgames: TBH I don't agree with your take. I don't think Athropic's desire to control the harness is about keeping resource usage under control. They could accomplish that by just enforcing limits on the actual resource usage (which they already do) -- if some third-party harness is inefficient, users of than harness hit their limits faster. I think instead that they want to control the harness because if switching LLM providers is too easy, it makes business difficult for the providers. Say GPT 5.5 comes out and it's clearly smarter, faster, and cheaper than Opus 4.7. If everyone can switch providers with two clicks in their harness, many of them will. This would lead to wild revenue and usage swings, which makes capacity planning hard. And perfect competition drives down prices -- in this scenario Opus has to cut its prices to get some users back. Obviously no business wants to be in that situation! By controlling the harness, they add some stickiness. If switching LLM providers means switching harnesses, that's a barrier high enough that most people won't bother to do it on a whim. So now Opus 4.7 can weather the storm until 4.8 or whatever comes out and is back on top. So it makes perfect sense to me as a business decision. It may be user-unfriendly, but tech companies do stuff like this all the time. It's nothing new. Though I would say, it seems weird to me to do this *on top of* subscriptions. Subscriptions already create a lot of stickiness. If you're subscribed only to Claude, that's a pretty big barrier to trying out GPT quickly -- a bigger barrier than the harness barrier I think. So I question whether controlling the harness is really worth all the effort they are putting into it, but idk, they probably have insights that I don't on this. Another factor here might actually be safety concerns. As we know, Anthropic leadership is deeply (excessively, IMO) worried about AI safety, and they feel that Anthropic will do a better job of addressing safety than any other company. They may feel that control of the harness is an important tool for that. I could definitely imagine Dario being terrified of OpenClaw from a safety perspective (I sort of am too). These explanations make much more sense to me than the efficiency issue, which again seems like it could easily be managed in other ways. But of course, these explanations are much harder to just come out and say, without stirring a lot more outrage...

https://nitter.net/KentonVarda/status/2050232966454935965#m

Published: May 1, 2026 15:16

TBH I don't agree with your take. I don't think Athropic's desire to control the harness is about keeping resource usage under control. They could accomplish that by just enforcing limits on the actual resource usage (which they already do) -- if some…

"pi, there's an example extension called minimal.ts or so. i want that, but cute. do the thing. make no mistakes"

https://nitter.net/badlogicgames/status/2050227405843788245#m

Published: May 1, 2026 14:54

"pi, there's an example extension called minimal.ts or so. i want that, but cute. do the thing. make no mistakes" Palani — oss/acc (@Palanikannan_M) @badlogicgames can we please have a mode/toggle where the cli doesn't show the tool results of read/etc…

RT by @badlogicgames: Can an open-weight coding agent + harness match Claude Code at training a domain-specific model? Same one-line prompt. ~13 min e2e. Pushed to @huggingface. Pi + @moonshotai Kimi K2.6 vs Claude Code + Opus 4.7. Task: classify NC session laws (1866-1967) as Jim Crow or not

https://nitter.net/vanstriendaniel/status/2050227087185781126#m

Published: May 1, 2026 14:53

Can an open-weight coding agent + harness match Claude Code at training a domain-specific model? Same one-line prompt. ~13 min e2e. Pushed to @huggingface. Pi + @moonshotai Kimi K2.6 vs Claude Code + Opus 4.7. Task: classify NC session laws (1866-1967)…

really like this anthropic podcast. learning so much new stuff about anthropic every episode. it's a good anthropic podcast.

https://nitter.net/badlogicgames/status/2050225380640882995#m

Published: May 1, 2026 14:46

really like this anthropic podcast. learning so much new stuff about anthropic every episode. it's a good anthropic podcast. Theo - t3.gg (@theo) Third episode of Nerd Snipe is live! This time I try to get Ben in on my Anthropic conspiracy theories (and…

RT by @badlogicgames: AI agents don’t feel the pain that humans do. 3 takeaways from Mario Zechner(@badlogicgames), creator of Pi - the minimalist, self-modifying agent that OpenClaw is built on: #1 - Automation bias: agents quickly wow us and win false trust: “There are moments of brilliance in agents where they spit out perfectly fine simple code. As the steering engineer you can look at that and think: ‘Wow, this is amazing. I can just sit back and not care because it's doing the thing how I would do it.’ Two minutes later you have another agent running that spits out the worst, horrible, garbage code, but you might not notice because you have fallen into automation bias and think your agent is doing the job well.” #2 - Humans have a different capacity for learning than agents: “Agents don't learn. You can put as much stuff in your AGENTS.md, build a memory system, but it's not the same type of learning that a human does. Humans are failable as well, but they have some capability of learning.” #3 - Feeling pain is inherently human and drives us to make things well: “Humans feel pain. I think that's one of the defining things about humans. If the pain gets too big, you as a human are incentivized to fix the cause of your pain. In a code base, the cause is usually terrible interfaces and terrible complexity that you want to get rid of because you can no longer maintain that system.”

https://nitter.net/Pragmatic_Eng/status/2050194777396682958#m

Published: May 1, 2026 12:45

AI agents don’t feel the pain that humans do. 3 takeaways from Mario Zechner(@badlogicgames), creator of Pi - the minimalist, self-modifying agent that OpenClaw is built on: #1 - Automation bias: agents quickly wow us and win false trust: “There are…

People of http://pi.dev. pi has supported OpenAI WebSockets mode since February. However, we did not support delta updates. pi v0.71.1 now also supports delta updates: only the latest context additions get send, leading to a nice 66% throughput increase. Start pi, /settings -> transport -> websocket-cached Here's SSE vs. cached WebSocket side by side. This will also apply to your OpenClaw instance, should @steipete & team decide to update to the latest and greatest, giving you all the benefits of the pi runtime on top of the Codex server backend.

https://nitter.net/badlogicgames/status/2050174400704418077#m

Published: May 1, 2026 11:24

People of pi.dev. pi has supported OpenAI WebSockets mode since February. However, we did not support delta updates. pi v0.71.1 now also supports delta updates: only the latest context additions get send, leading to a nice 66% throughput increase. Start…

RT by @badlogicgames: Writing code is cheap Maintaining code is not cheap Anyone who's hired an external contractor knows this

https://nitter.net/mattpocockuk/status/2050167042771194226#m

Published: May 1, 2026 10:54

Writing code is cheap Maintaining code is not cheap Anyone who's hired an external contractor knows this Jonathan Ross (@JonathanRoss321) For 50 years, software engineering ran on code rationing. Writing code was expensive, so we rationed it carefully…

R to @badlogicgames: see also https://x.com/badlogicgames/status/2022443997130813897?s=20

https://nitter.net/badlogicgames/status/2050152868796076541#m

Published: May 1, 2026 09:58

see also nitter.net/badlogicgames/status/2… Mario Zechner (@badlogicgames) People of pi. Still on vacation mode, but got nerd snipped. pi now supports websocket as a transport for the OpenAI Codex endpoint (same as Codex CLI). It also caches the…

http://pi.dev's been supporting OpenAI's WebSockets mode since March. /settings > transport. the difference between SSE and WS for e.g. Spark is not that stark in terms of tokens/sec. the real speed up actually comes from caching context OpenAI side, and only sending deltas as the context grows. pi currently does not implement the latter. hmmmmmm

https://nitter.net/badlogicgames/status/2050152124437147994#m

Published: May 1, 2026 09:55

pi.dev's been supporting OpenAI's WebSockets mode since March. /settings > transport. the difference between SSE and WS for e.g. Spark is not that stark in terms of tokens/sec. the real speed up actually comes from caching context OpenAI side, and only…

what are they doin to our boi?

https://nitter.net/badlogicgames/status/2050045801976431058#m

Published: May 1, 2026 02:53

what are they doin to our boi? Kyle Mistele 🏴‍☠️ (@0xblacklight) btw if claude feels dumber part of the reason is that it has literally 3 dozen tools BEFORE you add MCP servers it used to have like < 12 this is really all you need and skill is…

R to @badlogicgames: the most annoying part is the tokenizer really. only thing keeping me from going hand-written C. stupid bpe.

https://nitter.net/badlogicgames/status/2050030060128567414#m

Published: May 1, 2026 01:50

the most annoying part is the tokenizer really. only thing keeping me from going hand-written C. stupid bpe.

R to @badlogicgames: now i wonder how hard gemma 4b would be. only 32x bigger! maybe i should start with GPT-2 1.5B.

https://nitter.net/badlogicgames/status/2050029742091272524#m

Published: May 1, 2026 01:49

now i wonder how hard gemma 4b would be. only 32x bigger! maybe i should start with GPT-2 1.5B.

R to @badlogicgames: i did a lot of bespoke NLP in the 2000ths, max entropy, SVMs, etc. it's ridiculous that this simple architecture (well, encoder/decoder, but same same) basically did away with all that painstaking work. bitter lesson, etc. pp. still blown away.

https://nitter.net/badlogicgames/status/2050028182410002742#m

Published: May 1, 2026 01:43

i did a lot of bespoke NLP in the 2000ths, max entropy, SVMs, etc. it's ridiculous that this simple architecture (well, encoder/decoder, but same same) basically did away with all that painstaking work. bitter lesson, etc. pp. still blown away.

R to @badlogicgames: for your enjoyment: https://github.com/badlogic/gpt-2-ts GPT-2's architecture is really really simple to understand mechanically. Modern arch's are obviously more complex, but the basic principles are largely the same still. study it! https://openai.com/index/better-language-models/

https://nitter.net/badlogicgames/status/2050027889605656844#m

Published: May 1, 2026 01:41

for your enjoyment: github.com/badlogic/gpt-2-ts GPT-2's architecture is really really simple to understand mechanically. Modern arch's are obviously more complex, but the basic principles are largely the same still. study it! …

felt cute, did some @karpathy style cozy coding. now i can run GPT 2 124M in pure TypeScript at 7 tps. played with implementing the GEMV via C/WASM, but that only got me a 1.7x speed up.

https://nitter.net/badlogicgames/status/2050025762070180005#m

Published: May 1, 2026 01:33

felt cute, did some @karpathy style cozy coding. now i can run GPT 2 124M in pure TypeScript at 7 tps. played with implementing the GEMV via C/WASM, but that only got me a 1.7x speed up. Video

R to @badlogicgames: honestly think reimplementing a simple transformer will become the new implementing your own programming language. gotta understand your tools, and the best way is to build them yourself (yes, just building inference isn't quite the same, but then most compiler impls also stopped at the interpreter level :))

https://nitter.net/badlogicgames/status/2050001125558956391#m

Published: April 30, 2026 23:55

honestly think reimplementing a simple transformer will become the new implementing your own programming language. gotta understand your tools, and the best way is to build them yourself (yes, just building inference isn't quite the same, but then most…

love the phrase. my cozy coding plan: gpt-2 impl in C. tho likely hand coded, with some llm yelling at me when i get the matmuls wrong, as usual. cozy yelling.

https://nitter.net/badlogicgames/status/2050000816145064412#m

Published: April 30, 2026 23:54

love the phrase. my cozy coding plan: gpt-2 impl in C. tho likely hand coded, with some llm yelling at me when i get the matmuls wrong, as usual. cozy yelling. Andrej Karpathy (@karpathy) Haha yeah I call it cozy coding :) Usage: “This valentines, cozy…

RT by @badlogicgames: https://github.com/nexu-io/open-design/pull/117 oh i didn't expect this to get merged so quickly the pi lives on @badlogicgames

https://nitter.net/monotykamary/status/2049996439615574134#m

Published: April 30, 2026 23:36

github.com/nexu-io/open-desi… oh i didn't expect this to get merged so quickly the pi lives on @badlogicgames

People of http://pi.dev. I really, truely need to get the big refactor & pi server done, or kitten is mad. I will therefore close the issue and pull request trackers starting tomorrow for a week. Thank Kitten. In case of emergency, use Discord. Mostly getting feature requests now, so nothing should be horribly broken anyways.

https://nitter.net/badlogicgames/status/2049996239819919454#m

Published: April 30, 2026 23:36

People of pi.dev. I really, truely need to get the big refactor & pi server done, or kitten is mad. I will therefore close the issue and pull request trackers starting tomorrow for a week. Thank Kitten. In case of emergency, use Discord. Mostly getting…

RT by @badlogicgames: If you tried OpenClaw in group chats and got mixed results, you GOTTA try again. I changed how agents talk there, it IS SO GOOD NOW. https://docs.openclaw.ai/channels/groups#visible-replies And if you used GPT and got subpar performance, switch to codex harness. https://docs.openclaw.ai/plugins/codex-harness Enable both and boom.

https://nitter.net/steipete/status/2049988836160074022#m

Published: April 30, 2026 23:06

If you tried OpenClaw in group chats and got mixed results, you GOTTA try again. I changed how agents talk there, it IS SO GOOD NOW. docs.openclaw.ai/channels/gr… And if you used GPT and got subpar performance, switch to codex harness.…

R to @badlogicgames: I found the Cloudflare AI Gateway config a tiny bit confusing, so I tried my best to document it here: https://github.com/badlogic/pi-mono/blob/main/packages/coding-agent/docs/providers.md#cloudflare-ai-gateway

https://nitter.net/badlogicgames/status/2049988703238467751#m

Published: April 30, 2026 23:06

I found the Cloudflare AI Gateway config a tiny bit confusing, so I tried my best to document it here: github.com/badlogic/pi-mono/…

RT by @badlogicgames: Pi’s telegram gateway is like magic. I named my home server’s agent Yuri and we text all day. https://youtube.com/shorts/iAQq0MndX9c?feature=share

https://nitter.net/devopstoolbox/status/2049971155704881363#m

Published: April 30, 2026 21:56

Pi’s telegram gateway is like magic. I named my home server’s agent Yuri and we text all day. piped.video/shorts/iAQq0MndX…

RT by @badlogicgames: ⚙️ We made agent loops faster with WebSockets in the Responses API As Codex got faster, the bottleneck moved from inference to inefficient API calls WebSockets keep response state warm across tool calls, helping workflows run up to 40% faster end to end https://openai.com/index/speeding-up-agentic-workflows-with-websockets

https://nitter.net/OpenAIDevs/status/2049595890395152728#m

Published: April 29, 2026 21:05

⚙️ We made agent loops faster with WebSockets in the Responses API As Codex got faster, the bottleneck moved from inference to inefficient API calls WebSockets keep response state warm across tool calls, helping workflows run up to 40% faster end to end …