"I think it's a long way off"

Welp, I hope this one doesn't age similarly: https://www.youtube.com/watch?v=IFe9wiDfb0E

Come to think of it though, as long as civilization has existed, libraries have been a thing. Or well, monasteries or some kind of place that kept records and copied books. And before that you just asked the oldest person in your village/tribe and hoped they knew I guess. It's weird that access to knowledge has basically been unchanged for the past ~10k or so years until now.

> - Tom Scott, 8 years ago

Ah, research in the pre-LLM era was quite different. You actually had to know which questions to ask.

Going one level down further, it's really hard to imagine what it was like doing research into something you have zero clue about in the pre-Google era. I guess you just asked a librarian and hoped they'd point you roughly towards the right section and then just started to read random books?

There's also a second aspect to it, just in terms of RLHF mechanisms. If you've ever experimented with VLA models (i.e. vision input + text task = robotic arm motion output), they tend to need all the training examples of the robotic arm being motionless removed entirely, otherwise the model simply learns that staying still is rewarded and proceeds to never do anything at all. You successfully train the laziest bot in the universe. I wouldn't be surprised if something similar happens to LLMs if reinforcement learning is involved in the instruct tuning process. If no is a valid answer, why ever do anything?

Bread-first search, is it?

Since when has low effort become a selling point anyhow?

He does not understand his own motivations? Like he doesn't want to automate away his entire very expensive workforce and bunker/yacht service staff with robots. There is a pretty clear endgame here.

Aka confirmation bias.

We like explanations that fit what we expect, even if they're completely wrong.

I mean, turning a server rack into a phone just isn't something that makes any sense, so it didn't seem like a valid option to consider. I've thrown it at a few LLMs now to see what happens and they also make the same assumption with no extra info, that it's a server now being hosted on a phone. That's the statistics talking. It's very unlikely to be the opposite.

The list of conditions that qualify for what will most likely have to be redone completely from scratch.

Universities are entirely elective, why would they need to cater specifically to disabilities? Driving schools don't make exceptions for the blind. There are skills that people expect you to have with a certain kind of diploma, and it reflects poorly on the school if they don't have a standard to hold people to. Not being able to present your work in a basic way is one of those mandatory skills that imo makes you unqualified to graduate if you don't grasp it.

I'm not saying you can't make an exception for a person or two who is clearly as good or better than the rest otherwise, but people will always try to misuse as many exceptions as they can, en masse, as an excuse to just avoid doing required work.

> Monitor clipping. review_watched produces PathStatus and nothing else. It never feeds solve. Whatever it does to legs cannot reach the search.

(that's after being told twice to not use shorthand jargon nor reference the code directly)

Like the Fable ban stunt, I wouldn't put it pass Anthropic to kneecap Opus deliberately to drive more people to their more expensive option.