It's Still Easier to Imagine the End of the World Than the End of Capitalism | Hacker News Reader

It's Still Easier to Imagine the End of the World Than the End of Capitalism

astralcodexten.com

Is there some form of machine learning that doesn't produce horrifically bad slop that everyone who isn't me is using?

Because there are quite a few people who appear, externally, to be somewhat intelligent who think that AGI is right around the corner when trillions of dollars in hardware and hundreds of terawatt hours of electricity can't produce anything of value-- except addictive social media distractions suitable as a medium into which marketing messages can be injected.

A constant feed of LLM trash writing and deep learning garbage images isn't AGI.

It isn't even close.

If anything, all of the brains being sucked into this vortex who are chasing "dat cash" because using GPUs to trick people into buying dropshipped shit is more profitable than actual research, has set back AGI by decades.

I've been using the plain free ChatGPT for all sorts of things of value.

It's like having the world's greatest tutor available on all subjects at all times. Ask it to recap the key characters and themes of a book and it does it. Ask it to explain the pharmacokinetics of a medication and it does it. Ask it about how to cook a certain dish or improve your turns while skiing or fix a problem in your garden and it gives you a good answer.

I'm not sure how you're trying to use it that's resulting in "trash writing", but my experience is that it's like I can text a friend with a phD in whatever domain at any time and get a response explaining it.

The "hallucination" problem is still very real, it will at times confidently give wrong answers, but it still speeds up the research and learning process for me massively, I just know I need to double-check things it says, but it's much faster to check that an LLM's answer is correct than spend weeks reading textbooks to come up with an answer myself.

I have not seen a ChatGPT summary of a topic that is even remotely comparable to the first paragraph of any Wikipedia entry on a topic.

AI doesn't have to be LLMs producing slop. There are other algos eg. MuZero beating humans at board games and AlphaFold doing protein folding. I believe AGI is close but will require algo changes from ChatGPT et al. But the fact that they can produce something close to human intelligence in certain areas suggests it's not that far off.

Board games and puzzles aren’t AGI.

AGI is, among many many other things, inventing games based on random objects around you when no board games exist.

If you execute MuZero with no game to play will it, absent any direction, invent a game on its own and try to get others (people or other programs) to play it, explaining the rules and making up new rules on the fly to make it fairer or more fun, or will it just sit there idling and wasting electricity like a bunch of dumb tensors strung together by a bunch of people who think being good at Q*bert makes you intelligent?

ML is a pretty wide category, eg this is one example which would be hard to blame for "slop": https://pweb.cfa.harvard.edu/research/topic/machine-learning

I've been looking at ways to fix things, and I believe I'm close.

Fix what things? And how are you fixing them?

The auto-fart-smelling is pretty funny, and is made doubly so by the comments. At the very top someone immediately misattributes a quote due to an AI hallucination, triggering lamentations of AI polluting "high quality fora". Never mind that the "fora postum" in question is itself a bunch of fancy-sounding nonsense rooted largely in misunderstanding, hallucinated by a meat-powered intelligence.

I'm guessing post singularity we'll have something like socialism. Reasoning: at the moment if you just hand out money equally people don't get around to building much stuff so it ends in poverty. Hence communist regimes get voted out if not enforced by dictatorships. Post singularity the AI/robots will produce unlimited stuff so the voters will vote to split it approximately equally.

>Everyone will have access to ~equally good AI investment advisors, so everyone will make the same rate of return.

Lol, what?

No way, impossible.

Just solely due to access to informations.

Reply on news.ycombinator.com