bashtoni

My experience running quantised models is that they look the same at first but are all noticeably worse once the context size gets beyond anything trivial.

No, this is not a small model. I believe 4x Sparks is possible but 4.1 is pretty big for a flash model.

MiMo 2.6 Flash is the most capable model you can currently run on 2x Sparks without resorting to a quantised version of something.

"a minor update" they say before showing off benchmarks that display a substantial improvement.

You might not buy any of his products, but by visiting X you are a product he's selling to advertisers. The fact you block ads is not disclosed to the advertisers he sells to, but your regular visits absolutely are.

Russian social media app has backdoor. Who would have thought?

(Yes, I know they're technically Dubai based now)

Is AI going to put Mathematicians out of a job, or is it going to create many new jobs reviewing proofs it creates?

I'm not sure it's clear right now.

I love how much inside information Xiaomi are giving us on the every day challenges faced by AI labs. Before the release of MiMo 2.6 it was the RL dashboard, now this update.

Had to chuckle at this when outside the US the problem with EVs is that there are so many options people find it tough to choose.

Denza or Zeekr or Genesis or Tesla? Xpeng or Leapmotor or Geely or Kia or Hyundai or Mazda? BYD or MG or GWM or Chery or Jaecoo?

Plus the various Porsche, Audi and VW electric options, Subaru and Toyota, even the new Range Rover. There are far more electric options than petrol, and a lot of them are significantly better at the same or lower price point.

NZ isn't big enough to think about. Don't tell them because they get very upset.

Don't be evil -> What does "evil" really mean, anyway? -> Judgements about good and evil don't belong in business -> Evil is pretty profitable, actually

Are the Artificial Analysis benchmarks really worthwhile any more?

They really don't seem to match my real-world experience, and based on the comments I see I don't think that match most other people's either.

For example, Opus 5 was at the top for some time. My experience is that it's not noticeably better than Opus 4.8, and it definitely seems worse than Fable 5, which AA benchmarks put behind Opus 5. GPT 5.6-sol and Opus 5 seem pretty interchangeable, although Sol is noticeably better at finding problems in code, particularly edge cases.