granitepail

"Just" Kimi K2.5 with RL—people really misunderstand how difficult it is to achieve these reults with RL. Cursor's research team is highly respected within the industry, and what they've done is quite impressive.

Before people go jumping to conclusions about model theft, it's worth considering the possibility that they did reach an agreement with Moonshot which their researchers were not aware of. That would certainly explain the deleted tweets. Until Moonshot makes an official statement, I'm not particularly concerned.

To me, it’s very obvious that the problem is social media. To social media, AI slop is peak efficiency. The affordances and incentives of the network encourage its creation. I don’t care for the media slop, but eg television media has more or less been producing crap like this for a while.

I don’t think LLMs and video/image models are a negative at all. And it’s shocking to me that more people don’t share this viewpoint.

His JEPA family of models is a genuine step forward for SSL. Not the only approach, but a very insightful one. You’re very dismissive of his work.

If ya made it through all three of the sentences they wrote, you'd see the comment you replied to came around to it being reasonable to give families a break on group seating.

In my case, I’m paying for inference on the original models from e.g. Fireworks. So it’s not a quantization problem. The Qwen3 I was using was the new 458B (i think that’s the size?) model that was their top performer for code.

I agree with other comments that there are productive uses for them. Just not on the scale of o4-mini/o3/claude 4 sonnet/opus.

So imo open weights larger models from big US labs is a big deal! Glad to see it. Gemma models, for example, are great for their size. They’re just quite small.

While the benchmarks all say open source models Kimi and Qwen outpace proprietary models like GPT 4.1, GPT 4o, or even o3, my (and just about everyone I know's) boots on the ground experience suggests they're not even close. This is for tool calling agentic tasks, like coding, but also in other contexts (research, glue between services, etc). I feel like it's worth putting that out there--it's pretty clear there's a lot of benchmark hacking happening. I'm not really convinced it's purposeful/deceitful, but it's definitely happening. Qwen3 Coder, for example, is basically incompetent for any real coding tasks and frequently gets caught in death spirals of bad tool calls. I try all the OSS models regularly, because I'm really excited for them to get better. Right now Kimi K2 is the most usable one, and I'd rate it at a few ticks worse than GPT 4.1.

From the individual article summaries:

"The Editor-in-Chief and the publisher have retracted this article. The article was submitted to be part of a guest-edited issue. An investigation by the publisher found a number of articles, including this one, with a number of concerns, including but not limited to compromised editorial handling and peer review process, inappropriate or irrelevant references or not being in scope of the journal or guest-edited issue. Based on the investigation’s findings the Editor-in-Chief therefore no longer has confidence in the results and conclusions of this article."

More, I think, preying on the public's interest in UFOs to justify another massive military budget. All of the wild PR around UFOs with absolutely nothing of substance to back it seems to be a very effective PR campaign.

Maybe the backend was written in concrete.

Definitely. If you have the skills, working in a strong fund is the best way to get pay that scales with your contribution.

I haven't updated my LinkedIn in around five years. Definitely not a requirement. I do keep an up-to-date CV and just pass that along. Could always try putting n/a in the field and writing a note, or finding a hiring manager or someone in HR to mail directly.

From the hiring side--I've done quite a bit--I never thought twice about whether someone had a LinkedIn or CV. I do think it's lazy not to provide anything on one's background, though. It helps me guide the conversation towards a candidate's strengths.

Exactly. A cryptocurrency is driven by speculation. A treasury system helps manage flows of capital in and out of a business. Said company pays to use the proprietary system. Sure, they're both backed by a ledger, but Adjoint isn't turning profit by encouraging or even allowing people to speculate on the future value of their platform. They're selling a service.

I played a lot of M:tG at various times in my life, and games like Magic teach you a lot about these concepts. I would have never thought about it has I not had that experience! No shame in learning about the meta game!