sambejk

I think it is unlikely that this would be possible right now. The complexity of the driver and how it interplays with the hardware makes this really difficult. However wondering why you would need this as they are pretty good and free?

I'm a bit skeptical. This is just a single benchmark with an already capable (&closed-source) model. Would have been cooler if there was a more recent benchmark with an open-source model.

Schools have to choose between giving kids access to AI tooling (which has been proven to be terrible for education) or basic internet search.

I don't expect anything big to change just because there is one election. The entire judicial system is still very much MAGA.

It feels like we have overfitted on safety by a long mile. This is how labs that care less about safety (a.k.a. Chinese & Grok) will probably be able to catch up in the future.

Google has long forgotten their old motto: "Don't be evil", but this is a new extreme. Luckily we have things like Duckduckgo

I think calculator is not part of the labels in cifar-10. Pretty normal as it has ‘only’ 90% accuracy. Things could improve quickly though

Very cool to see practical applications of this technology!