FelikZ

@steipete Like you have access to most advanced models and funded by top 100 company - yet you choose to start new projects with JavaScript, knowing you will not write that code anyway and most likely not review most of it. Why?

🩹 clawpatch 0.1.0 is live: Clawpatch maps codebases into semantic feature slices, reviews them for bugs and quality issues, and records explicit fix attempts with validation. You'll be surprised how much this will find. npm install -g clawpatch clawpatch.ai

@badlogicgames There is also plenty of room to strip system prompt - for example shorter references to docs. In fact, those can be just an embedded skill.

sneak peak of the new pi.dev no tools mode. you're gonna love it. you can try it with `pi -nbt`. enjoy!

@mick__net @danielhanchen @UnslothAI Yeah I have the same spec and I have either no difference or slightly slower for both 27b and 35b. Thats weird if only nvidia gpus got a per win here

Testing Qwen 9B/ 27B Unsloth MTP GGUFs on Mac showed MTP slower for me. Setup: Apple M1 Max, 32GB llama.cpp MTP fork build: b9173-a957b7747 Metal, -np 1, --no-mmproj Model: unsloth/Qwen3.5-9B-MTP-GGUF File: Qwen3.5-9B-Q5_K_M.gguf Context: -c 100000 20k prompt, 2048 cap: no MTP: 23.49 tok/s ngram-mod,draft-mtp: 14.09 tok/s acceptance: 55.6% Also tried 512-token run: no MTP: 24.61 tok/s draft-mtp: 20.62 tok/s ngram-mod,draft-mtp: 22.44 tok/s Args: --spec-type ngram-mod,draft-mtp --spec-draft-n-max 6 --spec-draft-p-min 0.75 Anything obvious I should change for Apple Silicon/Metal?

Qwen3.6 MTP Unsloth GGUFs now run 1.8x faster, increased from 1.4x just two days ago! This is due to llama.cpp adding --spec-draft-p-min 0.75! Args have also changed from --spec-type mtp to --spec-type draft-mtp Also increase --spec-draft-n-max 2 to 6 We also released Qwen3.6-0.8B, 2B, 4B, 9B MTP GGUFs! We'll be providing more soon! For folks who find the new updated branch to have some perf regression, set --spec-draft-p-min to 0.0 to get the old behavior - we provided a plot of the old branch (red) vs the new branch (blue / green) as well. Also you can use 2 speculative decoding algos - you can add ngram via --spec-type ngram-mod,draft-mtp - the perf isn't yet optimized so I'll do more benchmarks to find better numbers - see github.com/ggml-org/llama… Guide for MTP: #mtp-guide" target="_blank" rel="nofollow noopener">unsloth.ai/docs/models/qw…

@eugenio8a8 Thats a level of creativity we have to deal with

Valve, a multibillion dollar company and one of the most profitable gaming companies in the space. Dev 1: "We need a system where people can report bugs fast, right when they happen." Dev 2: "Cool, what are we gonna call it?" Dev 1: "Let’s call it BugBug" Everyone applauds

@CounterStrike Bring back assault

Release Notes for today are up: store.steampowered.com/news/app/730/v…

@jun_song Does not work that way on 32Gb - constantly hitting kernel panic on mlx and according to search is it “well known issue that apple do not fix”. Perhaps not an issue when running on higher memory setups with smaller models.

Daily reminder: Use MLX format for local llms running on Mac Fastest and best format for Apple silicon.

@jun_song What exactly have you built using those? Or its all for fun

My current monthly subscriptions : - GPT Pro $200 - Gemini $20 - X Premium+ (supergrok) $40 - Minimax $80 - GLM $30 - Devin $200 (for latest model testing) - Ollama $20 - Huggingface $9 - icloud $10 Yes, cancelled Claude. I might go broke soon😂

@fitchmultz Whats the benefits of using this compare to playwright-cli + skill? Yes it wraps tools to “native” tools, however the agent still have to draft tool call which is similar to drafting a CLI call

If you use pi, try pi-agent-browser-native. It makes agent-browser a native pi tool, so agents actually use browser automation instead of awkward bash glue. In my runs it has materially improved tool uptake, speed, and token-efficient browser work. github.com/fitchmultz/pi-…

@kmdrfx No thank you. I think it for people who still use notifications sound on their phones.

OpenCode now let's you enable sound and desktop notifications for the TUI. Sounds are customisable. You can turn off sound or desktop notifications individually. Desktop notifications use OSC 9/99/777 sequences where available, might not work reliably in all terminals.

@thekitze Its Microslop, remember their attempt to get mobile market share - similar vibes.

> be github > invent copilot > you are literally the first one > you are literally the only one > you literally have access to all the code in the world > get mogged by literally every single agentic bs that came out in the past few years this level of fumble should be studied

@badlogicgames What a torture to read that language.

everything llm land is like that. context overflow signaling, retries, etc. it's very sad and very horrible.

he's become fully reliant on LLMs to code. now increase the price by 1000%

installing npm packages in 2026

@badlogicgames I am pretty sure pi audience on mac using brew. Just make it official, why not?

People of pi.dev. Do not install.by via any method other than what's shown on the website and in the docs. E.g. we do not publish to brew and never will. Someone else did. We have zero control over what goes into the brew release.

@yacineMTB Do you like spaghetti on your breakfast?

/goal optimize this code, use `date` to check the time and don't stop until I wake up at 7 am

@badlogicgames Any chance to avoid JavaScript for good. Single go binary will be amazing

pi depends on this package, but not that version. i'm sure we'll eventually get fucked by this too, by proxy. i hate the npm model of dependency resolution.

@jun_song How about stay calm and try lower bits and smaller models? Deepseek ~ qwen. Would be interesting to see results. It seems sweet spot is around 5-6bit and not 4bit

Running Kimi-k2.6 1T 8bit with only 21GB RAM on my Macbook at speed of 25tok/s. Some of my theory worked, but architecture is not perfect. Need to fix a lot of stuff, but there is hope. Working hard on this future method of Local LLM.

@petergostev Will it run on raspberry Pi?

This is the most ambitious project of my life, but I've finally succeeded - I've managed build a Cloudflare replacement. Try it now here: http://localhost:8000

@vaccoin Anti-what?

New Update on Counter-Strike 2 including Cache Fixes, Music Kits and various other Fixes ✅ Nothing about Anti-Cheat 🚫

5 months