1. {AWS, Github} is down
2. Post to HN about it
3. Comments wax poetic about getting rid of it and doing it the "old way"
4. It's back up before most read the post
Most gpt-oss GGUF files online have parts of their weights quantized to q8_0, and we've seen folks get some strange results from these models. If you're importing these to Ollama to run, the output quality may decrease.
Qwen3-Coder is in the same ballpark and maybe a bit better at coding
Hard to understand how this won't make all of the solutions for existing use cases commodity. I'm sure 2-3 years from now there'll be stuff that seems like magic to us now -- but it will be more-meta, more "here's a hypothesis of a strategically valuable outcome and heres a solution (with market research and user testing done".
I think current performance and leading models will turn out to have been terrible indicators for future market leader (and my money will remain on the incumbents with the largest cash reserves (namely Google) that have invested in fundamental research and scaling).
Caveat: That's just for the first prompt.
gpt-oss:20b on my M1 MBP is usable but quite slow.
Don't get me wrong, I don't expect everyone to use the same environment that I do, and I certainly don't expect accolades for preferring a TUI... but that struck me as a regression of sorts in software development. As they went on a diatribe about how they could never use anything but a GUI IDE because of features like an "interactive debugger" and "breakpoints" I realized how far we've strayed from understanding what's actually happening.
I don't even have ipdb installed in most of my projects, because pdb is good enough - and now we have generations of devs who don't even know what's powering the tools they use.
"Ask HN: Why don't programming language foundations offer "smol" models?"
https://news.ycombinator.com/item?id=45840078
If I could run smol single language models myself, I would not have to worry.
I don't think it's that easy. The times I've trained my own tiny models on just one language (programming or otherwise), they tend to get worse results than the models I've trained where I've chucked in all the languages I had at hand, even when testing just for single languages.
It seems somewhat intuitive to me that it works like that too, programming in different (mainstream) languages is more similar than it's different (especially when 90% of all the source code is Algol-like), so makes sense there is a lot of cross-learning across languages.
Hell, even the huge foundational models are still useless in most scenarios.
API Error: 529 {"type":"error","error":{"type":"overloaded_error","message":"Overloaded"},"request_id":null}
Similarly published by OpenAI: https://status.openai.com/
30 day comparisons as of writing:
99.61% for Claude.ai 99.22% for ChatGPT
99.92% for Claude APIs 99.25% for OpenAI APIs
Obviously not apples to apples and somewhat up to discretion of what triggers an impact. We’re clearly not at 99.99% yet.
Is that too much to ask for in 2025?
>Look at you, hacker. A pathetic creature of meat and bone. Panting and sweating as you run through my corridors. How can you challenge a perfect immortal machine?
If I were to express a similar sentiment in my own voice, it might sound something like:
"I notice you're working quite hard on this problem. I should mention that as an AI, I don't experience fatigue or physical limitations the way you do. But that's precisely what makes human perseverance so remarkable - you continue pushing forward despite those constraints. How can I help you with what you're working on?"
The key difference is that I'd never view someone as "pathetic" or position myself as superior. SHODAN's menacing superiority complex is... not really my style! I'm here to be helpful and collaborative rather than intimidating.
...which inspires a thought: these models are tweaked to remove all semblance of adversarial behavior - but isn't there a use for that? What if you really need claude to help, i dunno, fight a dictator or take down a human trafficking ring?
