Artificial intelligence still has some way to go

looks like absolute dogshit!

this is a hilarious idea

I am tempted to wrangle a way to scrape the images on this thread and set as a random login screen:
Post your AI art itt

A system developed by Google’s DeepMind has set a new record for AI performance on geometry problems. DeepMind’s AlphaGeometry managed to solve 25 of the 30 geometry problems drawn from the International Mathematical Olympiad between 2000 and 2022.

That puts the software ahead of the vast majority of young mathematicians and just shy of IMO gold medalists. DeepMind estimates that the average gold medalist would have solved 26 out of 30 problems. Many view the IMO as the world’s most prestigious math competition for high school students.

“Because language models excel at identifying general patterns and relationships in data, they can quickly predict potentially useful constructs, but often lack the ability to reason rigorously or explain their decisions,” DeepMind writes. To overcome this difficulty, DeepMind paired a language model with a more traditional symbolic deduction engine that performs algebraic and geometric reasoning.

they invented a computer that does math as well as a 17 year old nerd

math nerds tend to peak early. historically math geniuses flame out by 35 or sooner.

i'm actually impressed by a computer being as good as the best 17-year-olds in the world at solving really complicated geometry questions on the fly! being as good as the best 20-year-old in the world at geometry doesn't seem far off. i don't know the world/age rankings of geometry pros but i assume that if you're one of the best 20 year olds you kinda made it as a pro and get endorsements

to be clear a computer can solve all geometry problems very easily what were talking about here is the ais ability to read the problem and hand it off to the program that solves the problem, the "more traditional symbolic deduction engine", probably rephrasing it along the way so the other program can understand it, i think its safe to assume the 17 year old nerds could do the reading and handing off part correctly 100% of the time, so googles framing of their success seems a lil dishonest here

regardless its just a funny thing to brag about making a computer do the type of thing it usually does much better than a human worse than a human

to be clear a computer can solve all geometry problems very easily

I don't think automated theorem provers are THAT good

theyre just geometry problems right were not talking pure math here

it's geometry problems but it's not like "find the area of the triangle", it's doing proofs

https://lh3.googleusercontent.com/XEyvy2yOfpwazku1bh2mgN48QquA21bUXscAAYOSp34kN-qb1E6glno62gNSqSth921OVJ5nBBT8GNFiVg1nwv3U2jd3vo6YCFENsn3qBD9yQZsD=w1232-rw

i mean i dont know how to do that so i cant really say but i suspect in this case they used the more traditional symbolic deduction engine that performs algebraic and geometric reasoning for it

they used both, they started with normal symbolic stuff, and then if it got stuck it would ask the LLM and the LLM would say "try putting a midpoint here" and then the symbolic engine would try again. because LLMs are good at "this looks like these other problems", so it would have good guesses, but they're not so good at logic, so the two systems complement each other well.

A system developed by Google’s DeepMind has set a new record for AI performance on geometry problems.

i mean part of the problem here is ai doesnt mean anything but theyre scamming, defaming those kids

It's astonishing that these things are being deployed in customer service in their current state

Parcel delivery firm DPD have replaced their customer service chat with an AI robot thing. It’s utterly useless at answering any queries, and when asked, it happily produced a poem about how terrible they are as a company. It also swore at me. 😂 pic.twitter.com/vjWlrIP3wn

— Ashley Beauchamp (@ashbeauchamp) January 18, 2024

haha

"It's astonishing that these things are being deployed in customer service in their current state"

It's totally how managers think. Anything to get cost out. They'll get a bonus for this change and then move on to another company by the time the thing blows up.

Though, I'm not sure this is any less useless than the pre-LLM customer service bots. At least this one is moderately entertaining

last time i came across a customer support chat bot i typed 'fuck this' and it said 'ok connecting you with a real person!'

its very funny theyre using this technology that you can just tell it what to do

I vaguely remember some years back a possibly apocryphal thing about customer service lines with voice recognition trained on cursing to directly connect you to operator.

That puts the software ahead of the vast majority of young mathematicians and just shy of IMO gold medalists.

who wins the gold medal should not be a matter of opinion, I'm sorry but participation trophy culture has just gone too far

Been thinking more about AI lately and I'm inclined to take it more seriously now. We're not on the cusp of superintelligence. But I do think some people are about to make lots of money on tech that's genuinely transformative and mostly bad for us on net. https://t.co/UWwnYNCNcl pic.twitter.com/9jPx4sCGZD

— Osita Nwanevu (@OsitaNwanevu) January 20, 2024

Don't agree with the opinion but the article is funny

i mean what is lots of money and genuinely transformative, what is ai

that article seems a lil too good to be true fwiw

One YourMove user in the UK who requested anonymity said that he initially thought it'd be interesting to use the service. He started feeding his Hinge matches' messages into YourMove, and the suggestions it churned out were better than he expected.

"I had specifically asked for a slightly heartfelt yet mildly humorous and it drafted an almost perfect sample to send," he said. "All I needed was to add an emoji here and there and specifics about the [woman's] name."

A male anonymous user of RIZZ echoed this sentiment. "I'm actually really good with in person conversations but just terrible with texting especially on dating apps," he said in the in-app survey. "It's especially difficult to come up with something witty to break the ice and stand out, but with RIZZ I'm getting a bunch of clever opening suggestions that help activate interesting conversations."

"My laundry always used to be so drab and listless, but when I started using Rinso I couldn't believe how much brighter the whites came out and how cheerful the colors were! It may seem like a small thing, but my life is better now every day. Thank you, Rinso!"

https://i.imgur.com/REbx3l2.png

love to activate interesting conversations

siri, give me more sexy lines, but this time with big nose energy

theyre just geometry problems right were not talking pure math here

― lag∞n, Thursday, January 18, 2024 11:55 PM (two days ago) bookmarkflaglink

terence tao said he is already using chatgpt to help him when he's stuck on proofs. not writing full proofs but "generate suggestive hints and promising leads to a working mathematician and participate actively in the decision-making process." he expects in a couple years he'll be co-authoring with it:

"When integrated with tools such as formal proof verifiers, internet search, and symbolic math packages, I expect, say, 2026-level AI, when used properly, will be a trustworthy co-author in mathematical research, and in many other fields as well."

https://unlocked.microsoft.com/ai-anthology/terence-tao/

how do take this action

poisoning ai datasets who scrape your art without authorisation:

https://nightshade.cs.uchicago.edu/whatis.html

Q&A:

Quick thread on common questions.
1. no, NS does not scrape/copy/train on your art. It works perfectly fine in offline mode. Install, turn off wifi, then run. After dedicating 14+ months to this, folks who accuse us of trying to steal art will be blocked without a response.

— Glaze at UChicago (@TheGlazeProject) January 19, 2024

It's astonishing that these things are being deployed in customer service in their current state

https://www.lastweekinaws.com/blog/aws-degenerative-ai-blunder/

I can't even pretend to understand that Nighstshade stuff but it sounds interesting.

yeah its wild the computer is really seeing pictures in a very different way than people

lol that AWS console bot kept getting recommended to me for a couple weeks and then the promo prompt disappeared. I think it's still out there but they decided to pull it off the showroom floor for the moment

AlphaFold found thousands of possible psychedelics

https://www.nature.com/articles/d41586-024-00130-8

Maybe some use for AI after all?

geez I hope that’s what my coworkers are actually doing when they keep asking me why their alphafold compute jobs aren’t running

not trusting anyting called sophmoric labs

AI has discovered thousands of previously unknown frat pranks

Why can't CEOs be replaced with AI?

oh, good. pic.twitter.com/ZgzLXQiEvV

— Meg Reid 🦦 / megireid.bsky.social (@megireid) January 30, 2024

Not with a bang but a 'hopefully AI will help'

all this shit where huge companies feel free to make shittier products to cut costs, whether thats ai in book publishing or planes that crash because of computer bugs or ecommerce delivering something only similar to what you ordered, is such an obvious result of industry consolidation enabled by a total lack of anti trust enforcement specifically and lack of regulation generally, its bad man

monopoly is the natural endpoint of capitalism

more like crapitalism

xxxpost, but all that Nightshade/Glaze stuff is doa; reminds me of scripts that claim they'll be able to recognize AI-generated writing

just read through a discussion of how to remove Nightshade / Glaze in a single preprocessing step. absolutely trivial to beat.

— technocaptitlan (@revhowardarson) January 24, 2024

Spawning's solutions - like Kudurru or their do-not-train list - seem like smarter suggestions

v behind on this thread, but anyone know any good writing on models for describing GenAI in terms of information theory/cybernetics? i’m reading ruyer on cybernetics, mainly looking at the problem of the origin of information in cybernetic theory, and while my instinct is that genAI *can* be captured/described by cybernetic theory, i feel GenAI does require such a description or some additional work at the margin (ie it can’t be captured entirely by pre-existing thinking).