petersson.eth

Send this man a Mac, for humanity!

@mitsuhiko @antirez oh nice. is this runnig 2 bit quant? tried it when deepseek-flash -q2 was brand new with oMLX - had unusable tool calling behavior - is this potentially fixed when using this new ds4 runtime?

Nice! @antirez merged my tool parameter streaming changes into ds4. Means you can now just install the pi extension and you get great behavior from ds4 out of the box. Moved the extension into a separate repo now. You “just” need a 128GB Mac now. github.com/mitsuhiko/pi-d…

i got some gremlins in my codex.. They are real!!

@b_fiive @mitsuhiko would it be feasible to add iOS Multipeer/Google nearby as custom transports?

@yellowhatcoder @mitsuhiko Custom transports is one of our newest features! We’ve been working on a Bluetooth low energy impl with a community member: github.com/mcginty/iroh-b…

I don't know who needs to know this: iroh is good stuff. iroh.computer

@micLivs @iamnotnicola @badlogicgames how does it compare to tmustier/pi-ralph-wiggum ?

@iamnotnicola @badlogicgames a completely different thing, auto research is an entire framework for generating hypothesis in pursue of a metric. pi-goal is closer to a ralph loop with an exit strategy. it reinjects your prompt on agent's turn end until the agent uses a tool to state "im done here"

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.

@KernNiko Schau Mal auf uebrig.at hab ich vibe gecoded. Ist eine nette Visualisierung der Steuer Situation in Österreich

Unpopular Opinion: Lohnkostensenkung - KÖSt Erhöhung geht in die richtige Richtung Durch KI werden zigtausende Arbeitsplätze wegfallen Abgaben auf Arbeit machen bis 60% des Steueraufkommens aus. Wie soll das in Zukunft funktionieren, ohne Maschinensteuer oder KÖSt Erhöhung?

@badlogicgames i use github.com/apetersson/pi-… - works well for nvidia, oMLX, LM studio and others that support the v1/models endpoint - but it only enumerates, no detailed specs sadly yet.

It would be lovely if inference engines would have endpoints that advertise: - all locally cached models and their specs (input modalities, context window size, thinking, tool caling) - loaded model(s) that would allow harnesses to easily enumerate things dynamically, instead of having users (or their agents) write silly config files.

first time hitting my 200€ Codex weekly limit. Hit it pretty hard those past days, that means great things are coming..

@FranzSchellhorn @lukasludens Ich weiß genau wann der Zeitpunkt kommen wird dass die Pensionsprivilegien massiv eingeschränkt werden: Wenn ich 65 bin.

Wir leben in einer Simulation. orf.at/stories/342801…

@thsottiaux I can't move sessions to different projects/folders.. Annoying for those important foundational threads that spin off a new idea/project. Also multi machine sync.

It’s the little things that matter, what are some small papercuts you have noticed in Codex? We’ll fix as many as possible in the next week.

@mitsuhiko Confession: I have not idea how to work the clipboard with tmux yet Tmuxcheatsheet.com is on auto complete for the letter t in my browser, and it's not going away soon.

I think tmux is great software for an agent. But how people can actually work day to day in tmux is beyond me. It's such a horrible UX and hack.

@Blueelectron4 @IAPonomarenko I am very sure the small margin towards the 2/3rd majority was triggered by JD, the reaper.

@IAPonomarenko It would be really funny of JD Vance's speech was the thing that tipped the balance against Orban.

The Hungarian people are now having a very real chance to do the world's funniest thing in a long time

Caveman speak for best agent. Hard times make strong programmers. Strong programmers make good times.

@thekitze TBF, Car wash benchmark must be in all the training sets by now

chatgpt 5.4 vs gemma 4 😬

@thekitze Of course not as the chat UI, open webui is the main UI for me, or openclaw for agentic work

@thekitze oMLX ftw. Best caching and multiple API keys make for a great LLM lan server.

is there a better lm studio alternative for chatting with local models? who the goat?

@badlogicgames I am, happy to contribute, have a look at llm360.ai a public good, which I am a big fan of.

we as software engineers are becoming beholden to a handful of well funded corportations. while they are our "friends" now, that may change due to incentives. i'm very uncomfortable with that. i believe we need to band together as a community and create a public, free to use repository of real-world (coding) agent sessions/traces. I want small labs, startups, and tinkerers to have access to the same data the big folks currently gobble up from all of us. So we, as a community, can do what e.g. Cursor does below, and take back a little bit of control again. Who's with me? cursor.com/blog/real-time…

Thanks so much for the exciting Hackathon and an amazing 26 hour long building streak, big thanks @techno0ptimist for organising, @steipete for your support and everyone for the amazing builder energy. Proud to be contributing to @openclaw security bit by bit.

Hosting Vienna's largest AI hackathon to date today!! 20k in prizes (5k€ cash 💶), 4 tracks (AI agents, OpenClaw, cybersec & bio 🧬 + special whimsy track 🦄) There's still time to join (we're coding overnight! 🦞✨)