Shaun Smith

RT @Kimi_Moonshot: Congrats to the @cursor_ai team on the launch of Composer 2! We are proud to see Kimi-k2.5 provide the foundation. S…

@badlogicgames Agreed. Open weights means customising for your own needs and alignment. Why would you trust anyone else to do this on your behalf?

I, for one, don't think there's anything wrong about Cursor taking an open weights base model and turning it into a great agentic model (any licensing issues not withstanding, don't know the details). I want this to be the future for all of us.

Putting out a wish to the universe. I need more compute, if I can get more I will make sure every machine from a small phone to a bootstrapped RTX 3090 node can run frontier intelligence fast with minimal intelligence loss. I have hit page 2 of huggingface, released 3 model family compressions and got GLM-4.7 on a MacBook huggingface.co/0xsero My beast just isn’t enough and I already spent 2k usd on renting GPUs on top of credits provided by Prime intellect and Hotaisle. ——— If you believe in what I do help me get this to Nvidia, maybe they will bless me with the pewter to keep making local AI more accessible 🙏

@dominikhonnef When they did the $1,000 free credits when they launched the Claude Code it scooped up an API key and started using it. Issue filed, no response, out of pocket.

Anthropic's "Claude for Open Source" program has been an awful experience for me. I applied and got invited a couple days later. When I used the promo link to get free Max, it charged me $200, anyway. It has been 17 days of me trying to talk to a human to rectify that.

@badlogicgames Let's do it 🍻. The last one was great fun!

who here will be a AI Engineer London in April? I'm ready to have more pub visits.

@jxnlco what's most token dense?

Future of AI

@OpenAINewsroom @astral_sh Go team 🐍

We've reached an agreement to acquire Astral. After we close, OpenAI plans for @astral_sh to join our Codex team, with a continued focus on building great tools and advancing the shared mission of making developers more productive. openai.com/index/openai-t…

@arvidkahl I've found in my testing that it's perfectly usable for coding -- if I were on API rather than plan it would be my default choice.

If you do AI inference via OpenAI’s API, you should use the flex tier for half price. My requests always try to use flex tier first, and on 429 / 500 errors, I use the default service tier. 95% of my requests are flex. 2 tries flex, then fall back to standard. Massive cost cut.

@idosal1 I've played, can confirm it's fing cool.

The bigger IDE is multiplayer. AgentCraft now lets humans and agents collaborate in one shared workspace! ⚔️ See allies on one map. Share context. Hand off agent work across machines.

@liran_tal It's not even the best harness for Claude.

I've been saying this for so long. I'm not even sure the CC harness is at all better than the competition (codex, cursor), just folks uninformed

Loving this new way of looking at the Hugging Face Hub; Generative UI. Early days, but looking promising 😎

I'll publish a few quickstart packs over the next couple of days with development environments optimised for Codex and Hugging Face IP and code mode subagents.

OK, we know the drill by now. Proper llama.cpp support yesterday, now the best(?) coding and general purpose agent has GPT-5.4-mini/nano support. Oh, and MCP Server side migration to FastMCP3 (deprecating SSE transports). Web Search is very snappy with the mini model 🌐

@nicoritschel @AbeWheeler5 👀

@evalstate @AbeWheeler5 has something for this

What's a good simple MCP Apps testing tool that isn't MCP Jam?

@Shashikant86 (those are rolled up loops btw)

@Shashikant86 Definitely some variance in TTFT -- think that's probably the issue.

Codex is amazing and codes bug free softwares than any other agents. However, I would love to see the fast mode in the Codex so that I can get something done without needed the deep analysis of the code. It would be amazing to get "fast" mode in the codex for those who wants needs to get something done quickly when time is money. @thsottiaux @ah20im @romainhuet Something similar to what @AmpCode did like smart/fast mode etc and let users decides.

@Shashikant86 Yes, I'm trialling using spark for it again, but gpt-oss-120b with heavy guardrails still seems to beat it.

And that's the FastMCP 3.1.1 migration completed, with working Hugging Face OAuth and token passthrough, and elicitations and all that stuff! A wise man said "you're weird - I bet you can't just change the imports". They were right 🤣