Yup. It'll also be handy knowing that you'll never have an AI actor get "cancelled" for whatever random reason might come along.
FaceDeer
joined 2 years ago
You're welcome to try other methods but LLMs seem to be working best so far.
With a decompiler it should be pretty straightforward to automatically check for "hallucinations," the compiled code is still right there and you can compare the decompiled logic to the original.
The time at which the source code was lost is irrelevant for decompilation, decompilation uses the binary files. Those are the files that are out there being played right now.
Until recently decompilers tended to produce rough and useless code for the most part, but I'm looking forward to seeing what modern LLMs will bring to decompilation. They could be trained specifically for the task.
It's very hard for an established prediction market to go all the way to either 0% or 100%. There'll always be someone who bought in earlier on the losing side and isn't bothering to cash out for the handful of pennies they might be able to theoretically get back now because it there's nobody who's actually buying.
They're not using it as a "verified source", except in that they're saying "this is the source."
It's actually not a bad approach. Prediction markets (which betting sites are just a form of) are often very good at reflecting what the current well-informed belief on a subject is, because people who make bets on ill-informed beliefs quickly end up not having money any more and thus not betting any more. It's just important to bear in mind that it's not literally saying "this is what's going to happen," it's saying "this is what well-informed people currently believe is going to happen based on current information."
Workarounds for those sorts of limitations have been developed, though. Chain-of-thought prompting has been around for a while now, and I recall recently seeing an article about a model that had that built right into it; it had been trained to use tags to enclose invisible chunks of its output that would be hidden from the end user but would be used by the AI to work its way through a problem. So if you asked it whether cats had feathers it might respond "Feathers only grow on birds and dinosaurs. Cats are mammals. No, cats don't have feathers." And you'd only see the latter bit. It was a pretty neat approach to improving LLM reasoning.
And they're overlooking that radionuclide contamination of steel actually isn't much of a problem any more, since the surge in background radionuclides caused by nuclear testing peaked in 1963 and has since gone down almost back to the original background level again.
I guess it's still a good analogy, though. People bring up Low Background Steel because they think radionuclide contamination is an unsolved problem (despite it having been basically solved), and they bring up "model collapse" because they think it's an unsolved problem (despite it having been basically solved). It's like newspaper stories, everyone sees the big scary front page headline but nobody pays attention to the little block of text retracting it on page 8.
Which is actually a pretty good thing.
I wouldn't call it a "dud" on that basis. Lots of models come out with lagging support on the various inference engines, it's a fast-movibg field.
Why does the rule need to be specific to data centers? Why not just try to encourage renewable energy in general?
Wow, not what I was expecting. He didn't even look wounded. The level of despair those Russians must be experiencing is incomprehensible. Here's hoping that it helps cause the front to collapse sooner rather than later.
It's a process that's been going on for a long time. It used to be that if you wanted to listen to music you needed a human artist to physically play it for you, but recordings have been normalized for so long that nobody gives it a second thought.
Heck, this is computer games we're talking about. Much of the performance is inherently "robotic" on some level already.