@camsoft2000 You get small model thatβs good in solving your problems your way. Most of the harness can be removed because itβs basically flashed into parameters. This is hypothesis but my gut feeling is we are going there. Large models that we get rn for cheap are not sustainable.
@camsoft2000 Long term solution is having small open models (fit into 100gigs) that are fine tuned to solve your problems on your datasets. I.e. you build harness, skills etc,l that are about your way of development and your projects, collect and refine/tune datasets, do finetuning.
@zeeg And gpt-oss120b, when run properly, is better in coding/infra tasks than most/all other open weight models from china which are ten times bigger. Hope this gives you a clue on how immature, overhyped, and at the same time immensely potential the βlocal modelsβ space is.
@zeeg Not to mention that overall infra is just outright broken - the gpt-oss family, which was released 9 months ago, is still broken in most inference engines including llama.cpp, ollama, lm studio and requires hours of investigation and patching to run properly on common hardware.
If you're interested in how to rewrite an open source project like git yourself, we wrote a post on
- How to "code cannon" yourself
- A retro of our rewrite of git in zig
- Our theory as to why this works
https://t.co/I8AXannE3j
In 20 years, vibe coders will look at the Linux kernel repo the way we look at the pyramids. In awe, unable to imagine how they managed to drag all those giant stones and pile them up in the middle of the desert.
Peter Steinberger is joining OpenAI to drive the next generation of personal agents. He is a genius with a lot of amazing ideas about the future of very smart agents interacting with each other to do very useful things for people. We expect this will quickly become core to our product offerings.
OpenClaw will live in a foundation as an open source project that OpenAI will continue to support. The future is going to be extremely multi-agent and it's important to us to support open source as part of that.
@ivanfioravanti@PromptInjection I can still complete some tasks with it in airplane mode (literally) with proper prompting and it will not trash me with tables