Yeah, I find it just works and reasonably priced. While local models are pretty competent now, it can do more complex tasks I find.

yeah that’s a perfect analogy actually

Yup, and a massive advantage in the amount of data available by virtue of having a much bigger population.

People really need to stop parroting this line uncritically. The process takes time because even when you’re distilling answers, you still need to actually do reinforcement training on the model. And given that Fable and GPT 5.6 just came out there simply hasn’t been much time to do that. However, models like Kimi also do better than Fable or GPT on a lot of tasks, which means it’s not just distillation but also difference in architecture. You can watch this talk from Kimi founder to see how Kimi was actually trained and why it performs well.

It’s also absolutely hilarious that people think only Chinese companies use distillation, as if Anthropic or OpenAI are above that or something. Not to mention that they basically ignored copyrights on all the data the siphoned and are now crying that people aren’t respecting their terms of use.

The reality is that China tops the world in artificial intelligence publications today. Chinese labs have come up with a bunch of genuine innovations: GRPO, auxiliary loss free MoE load balancing, MLA, muon optimizer, and a bunch of other ones. The Deepseek papers are really well written, this isn’t just sneaking a peek at a peer. Anybody who thinks China is simply distilling glorius American models is not engaging with reality.

This whole narrative has just been a massive cope.

Not really, while it’s supposed to be released on the 27th, you need massive hardware to run it. So, it won’t be useful to individuals. The problem from Anthropic perspective is that other providers will be able to run Kimi at that point. Moonshot has already closed subscriptions because they’re at capacity, so they actually benefit from more companies running and offering Kimi since they can only serve so many requests themselves. On the other hand, Anthropic has to foot the bill for every instance of Claude themselves.

Try actually reading what I wrote.

It would be cool in a sense that could could have a robot able to do these things by hand I guess.

Nobody is talking about replacing dishwashers and laundry machines with humanoid robots here.

gonna be awesome

Great, now you need two separate supply chains for each robot, and you have to keep to sets of parts to service them, and then if the thing you were doing changes to a different thing the robot becomes useless.

There are smaller models like Qwen 3.6 27b that can run with as low ast 16gb VRAM. The larger models are largely used either by companies running them on prem, or research institutions.

the thing to remember is that these models do not block anything for the people anthropic prefers which is basically the same thing that happened in the 90s with encryption export controls

I think it’s mostly that Iran was going through the motions with the MoU and didn’t want to escalate during that time.

The process takes time because even when you’re distilling answers, you still need to actually do reinforcement training on the model. And given that Fable and GPT 5.6 just came out there simply hasn’t been much time to do that. However, models like Kimi also do better than Fable or GPT on a lot of tasks, which means it’s not just distillation but also difference in architecture. You can watch this talk from Kimi founder to see how Kimi was actually trained and why it performs well.

It’s also absolutely hilarious that people think only Chinese companies use distillation, as if Anthropic or OpenAI are above that or something. Not to mention that they basically ignored copyrights on all the data the siphoned and are now crying that people aren’t respecting their terms of use.

yup, good news is all the files are out there, and not going anywhere :)

You can’t distill models that fast lmfao. People just keep parroting this without having any clue how any of this works.

I find traditional OCR apps are just qualitatively worse. This tool understands complex tables, extracts mathematical formulas, predicts reading order, and outputs nicely formatted Markdown. So, it’s not just blindly scraping flat text from a document.

Aside from agentic coding, this seems to be the most interesting niche to me. There are now lots of models for doing stuff like tts, stt, ocr, and it’s really handy for sifting through documents, extracting transcripts, etc. And it’s all stuff you can run locally on moderate hardware.

so brave

about as intelligible comment as I’ve come to expect from you