@katclone Is it good / bad?

@javilopen And (C) is not just external stimuli as in (A). It should be something that creates constraints for the model blocking its behavior that is not compliant with the environment, not just stimuli for passive observation for pattern extraction.

RT @MarioBalukcic: @dmdohan 2012$ to reach 75.7% in 1.3 minutes and an unknown amount for achieving 87.5% in 13.8min. I assume the cost is…

RT @mikeknoop: o3 is really special and everyone will need to update their intuition about what AI can/cannot do. while these are still ea…

@fewsats +

And now it’s a tornado in SF. Will OAI bring us the end times next week? :)

I feel like most people writing about LLMs are not even trying to use them in the real world. They write about "ChatGPT killers" and amazing performance for models which don't actually perform that well when you try to test them even on simple tasks. I understand, impressions are important, but so is reputation.

@theautomatedtom @levelsio I don’t believe Willow contribute to MWI meaningfully though, but I’m not an expert. Willow uses quantum mechanics to attain its performance, but it’s interpretation neutral, and I don’t see how Willow allows us to lean towards any specific one.

@katclone @jdpereira Depends on the goal. I don't think full brain uploading is possible, but creating a model that approximates our behavior and knowledge very closely may be good enough for "preserving the knowledge in the form of AI". Although that's not the same as longevity / immortality ofc.

It is not me: each of us would perceive the other clearly as separate individual entity, even being able to look at the copy and see disconnect from their POV and experience they gain starting from this point forward. That also poses an interesting question on if we’re the same person as we were yesterday of 5 years ago or are we just a novel set of thought “tokens” that considers its own past versions as a context creating an illusion of continuity. And in this case aren’t we dying every day or even every hour / minute with our imperfect “clone” replacing us every time?

@sdw *is not portable

@sdw The chip is too busy with all the spatial compositing to be able to run heavy productivity apps. And putting in Ultra chip would probably not work due to heat & power constraints. I would prefer Mac providing external compute and window rendering, and AVP doing compositing.

RT @alexalbert__: Fun story from our internal testing on Claude 3 Opus. It did something I have never seen before from an LLM when we were…