Ed Zitron: Investing a billion dollars to train a model that is 10% better at generating a picture of Garfield with an ak47 and giant cans for the low price of boiling an entire lake www.tomshardware.com/tech-industr...

Investing a billion dollars to train a model that is 10% better at generating a picture of Garfield with an ak47 and giant cans for the low price of boiling an entire lake www.tomshardware.com/tech-industr...

Also "I think things will just get better for reasons"

Yeah man that's how everything works right. No need to check this

It's all about Spacestar Ordering, my dudes.

They're writing the scenario of a sci-fi movie that we've already watched

A chemistry will happen to make the AI not suck ass

Shades of autonomous car boosterism circa 2015-16... "It can't be that hard, right? We'll get there next year, right?" Cackle.

Oh man I made so many people so mad for so long explaining why that shit was never gonna work

I got in to driving research right around the time of maximum boosterism... and as I worked on the human side of the problem, realized that we really didn't have a handle on that, much less on how to do it in silicon.

Hell, I fell into driving research by happenstance at the right (ish) time.

obligatory

[Narrator voice]: The improvements did not continue apace.

I wish they continued Apache

They will continue pear-shaped, don't worry

Someday AI will know what the shape of a pear is. Ok maybe not

Imagine, right, if there were all these spare computing resources just… doing nothing. Just requiring some investment. Thinking machines that could run rings around the fastest supercomputers at doing human stuff, just lacking resources and investment. Imagine if society invested in them.

if we put a tremendous amount of money into it now, we will be able to put an even more tremendous amount of money into it later

" For this reason, many tech companies, including Microsoft, are now considering modular nuclear power for their data centers." And given the quality they put into their products, nothing could go wrong here.

Have they ever defined what "better" means?

They said it means more gooder

The line must always go up

the algorithm will continue to get better forever at the same rate. i'm a genius

You could write a pretty good statistical law about the linearity of that development, im sure it will never hit a technical roadblock at any point

All we need is a pool of data greater than the entire sum of human knowledge to produce derivative works of middling quality, easy-peasy. Also: en.m.wikipedia.org/wiki/Moore's...

Moore's law - Wikipediaen.m.wikipedia.org

We should ask it, "What's the first & easiest thing to drop for mankind, in order to slow down global heating?", every now and then.

Yes let's tell the AI it's getting downsized That always goes well

Old enough to remember (when it happened on a s/w tv set). media0.giphy.com/media/v1.Y2l...

matthew broderick professor falken GIFmedia0.giphy.com Discover & share this Animated GIF with everyone you know. GIPHY is how you search, share, discover, and create GIFs.

One of the most persistent themes in the 80s was getting wrecked by omnipotent computers. Sadly, I think some of us raised by media ended up with the wrong takeaway from that. Me, I'm not automating my house or wandering in the woods at night, or moving into houses with creepy dolls. XD

"... wandering in the woods at night, or moving into houses with creepy dolls. XD" ..? – I think you'll be fine, just mind to leave your smartphone at its cozy little charging station 😉

(Not 'automating' the house is a no-brainer, actually. Maybe the ppl doing it just unconsciously hate their tenement on a scorched-earth sort of level ..?)

I am struggling to imagine a professional writer who doesn't know the word "apace" and then struggling to understand why this person is getting paid to write anything and THEN struggling to understand what the fuck his editor even DOES

Twice a pace. The proofreader in me goes ugh.

10% better goatse? In this economy????

Dude has “if” doing a lot more work than Kipling ever asked of it

This is clearly the correct and rational takeaway here

And if you question how long it's taking for anything concrete to materialize, it means you think your kid is stupid

This is some extraordinary bullshit, given how little technical knowledge of LLMs is necessary to refute this claim. Model training has nothing in common with how humans learn.

All this to avoid paying creatives a living wage.

Does the billion dollar go straight into AWS bills or does it include all the people who have to label everything?

Or we could be solving the climate crisis. Call me crazy.

Could they pleeeeeaaaaase just pay human artists and designers. Billions would mean the world to us, help stabilize and rejuvenate the economy, and NOT boil lakes.

👏 what👏 does👏 better👏 mean

Every time in the past few days I’ve hoped you’ve seen and enjoyed @acollierastro.bsky.social ‘s video about Zoom CEO’s AI takes. You’re both super fun to watch go off.

The most comical thing about this… it’s all pretty reverse John Henry, we are still an incredibly long way from human levels of capability at human energy use levels.

Read about "fluid intelligence" here, and tell me how LLMs compete. www.theatlantic.com/health/archi...

There Are Exceptionally Sharp Octogenarians. Biden Isn’t One.www.theatlantic.com The president’s age isn’t his problem.

Hey man. Yeah that's funny. Can I, uh, can I see the picture?

They can stop now. These models are already vastly superior at generating custom Hentai than I will ever be.

Can we just, in the words of Nik Suresh, just "f#@king piledrive" this person?

Excellent summary.

And if that doesn't happen, we'll just fake the ever-loving fuck out of it for long enough to bilk all the money left. I think the last man to make predictions this confident of future returns was Bernie someone, can't quite bring the fucking name to mind...

world's strongest "if"

Post