will_a113 @ will_a113 @lemmy.ml

Posts

40

Comments

271

Joined

2 yr. ago

Jimmy Wales (of Wikipedia fame) has been working on something like this for several years. Trust Cafe is supposed to gauge your trustworthiness based on other people who trust you, with a hand-picked team of top users monitoring the whole thing — sort of an enlightened dictatorship model. It’s still a tiny community and much of the tech has to be fleshed out more, but there are definitely people looking into this approach.

His complaint is hinged on the idea that GPT-4 is an AGI, so... I don't think this is going anywhere. It'll be interesting to hear what each side says constitutes an AGI though.

Who is your PM or senior assigning the tasks? You need to take this up with them -- everyone always needs a couple of quick hits in their back pocket. When you stall out grinding on task after impossible task it kills your motivation and productivity, and that's your boss's job to fix.

Do you do any kind of before/after testing of these to measure performance/accuracy changes? I've always wondered if there is some way to generalize the expected performance changes at different quantizations.

What's a tortoise?

It’s actually 1.58bits weirdly. The addition of 0 here was the significant change/improvement in this experiment. The paper isn’t too dense and has some decent tables that explain things fairly accessibly.

No.

It was kind of a difficult read for me - things just hit a little too close to home for me, and the resolution was too perfect. I'd still recommend it though - at the end of the day it's still Kim Stanley Robinson, and he is an absolute master of hard social-scifi.

Every time I read a story about some billionaire getting angry about their private jets being tracked I recall a part of the Kim Stanley Robinson novel Ministry for the Future, a (very) near-future tale about how a few global climate catastrophes wreak such havoc that regular people start taking extreme measures -- for example randomly shooting down passenger aircraft for months, causing the collapse of the air travel industry. I have to imagine that the 1%ers are thinking about that too now.

I wonder if content should carry some license automatically. Like if you agree to the TOS of an instance, your comments are automatically all licensed as CC:BY or CC:O or the more restrictive license of choice of the instance owner.

One thing I never thought about is how the longer a scientist is out of work (due to war, political instability, etc) the more likely it is that they’ll be lost to science (in other words, the less likely they are to return to scientific work when they can). That could potentially be like a chilling effect on steroids.

One of the video clips has Doom Guy shooting at a Baron of Hell :)

Are there any self-hosted twitter front ends? Something that would still use your account credentials (or maybe shared credentials) but would cut out all of the cruft, kinda like piped for YT?

"Texas needs to be less dependent on the federal government, not more. These politicians want to mismanage our electric grid just like they mismanage our border," the statement said.

I don't think it's objectively possible to be more mismanaged than the current Texan electrical grid.

My folks live on costal Long Island and have had 3 “100-year” flooding events in the last 10 years — and 3 10-year events since December.

Yep I’m sure that was the theory. Was just curious if anyone had been checking on how it’s going in practice.

Fair enough. ML ⊆ AI then. But these days when everyone talking breathlessly about AI taking away jobs they’re almost always taking about LLMs. This article is about ML in particular which is a different discipline with different applications.

Has there been any data on whether the approach is working?

ML =/= AI. There are legit uses for ML that don’t have anything to do with LLMs and the cloud. I worked on an ML project 3 or 4 years ago to listen for fan noise that might indicate that it was about to fail soon. We trained a tiny GAN on good and bad noises. It runs on a tiny CPU, locally. Highly specialized work, and I have to imagine there are and will continue to be lots of similar opportunities to bring efficiencies by getting computers to make good observations and decisions - even if only about “simple” things like “does this thing seem like it’s about to break?”

I read about 25% of the book this is based on before giving up (too manifesto-y for me, and needed a different/better editor), but the thought of coupling a book with a game like this is pretty interesting — get the point of your argument across to people who might otherwise never engage with it (if not for the title, anyway)