Egregoros · Phoenix Framework

Timeline

Post

Remote status

Context

jwz

@jwz@mastodon.social remote

18w

A slopfondler walks into a bar.

They: Hey. I found a good use of AI!
Me: No you didn't.
They: Wait let me tell you about it.
Me: Please don't. I'm begging you. Stop talking.
They: No, you'll like this.
Me: I assure you, I will not...
https://jwz.org/b/yk4i

Jeff Atwood

@codinghorror@infosec.exchange remote

18w

@jwz well, it gave me critical information about my health and DNA that none of the dozens of "healthcare professionals" I was working with for 4+ years ever did. (as you can probably understand, I have an entire team working on my stupid ass)

Replying to @codinghorror@infosec.exchange

The Orange Theme

@theorangetheme@en.osm.town remote

18w

@codinghorror @jwz Everyone knows the solution to structural healthcare problems caused by late capitalism is one trillion GPUs and a chatbot Ponzi scheme.

Replies

Replying to @theorangetheme@en.osm.town

Only 3 Easy Payments of $19.95

@feld@friedcheese.us remote

18w

@theorangetheme @codinghorror @jwz the structural problems don't change the facts that doctors can't know everything about everything

Replying to @feld@friedcheese.us

The Orange Theme

@theorangetheme@en.osm.town remote

18w

@feld @codinghorror @jwz And I don't think tanking the global economy at the whims of sociopaths who don't care how much CO2 we pump into the atmosphere is worth it because "some guy online said ChatGPT diagnosed him", sorry.

Replying to @theorangetheme@en.osm.town

The Orange Theme

@theorangetheme@en.osm.town remote

18w

@feld @codinghorror @jwz "Doctors don't know everything" is a pretty piss-poor defense of LLMs. Try again, please.

Replying to @theorangetheme@en.osm.town

Only 3 Easy Payments of $19.95

@feld@friedcheese.us remote

18w

@theorangetheme @codinghorror @jwz sorry but these tools have the ability to process thousands of research papers and connect dots faster than humans will ever be able to. And that can be the difference between life and death for many when they have issues that are going undiagnosed. It's just math.

Can you pay hundreds of doctors and researchers to spend months doing nothing but focusing on your problem? Few are rich enough to be able to do this.

Replying to @feld@friedcheese.us

Ariaflame

@ariaflame@masto.ai remote

18w

@feld @theorangetheme @codinghorror @jwz Aw, bless. You think they actually do analysis rather than spicy autocomplete.

Replying to @ariaflame@masto.ai

Only 3 Easy Payments of $19.95

@feld@friedcheese.us remote

18w

@ariaflame @theorangetheme @codinghorror @jwz they don't need to do analaysis, it's a vector. it's correlation. feed them the papers, input key parameters, they'll find links between them. Then a human reviews them for accuracy, and then maybe you found a new diagnosis, treatment, or cure. It's not a difficult concept.

Replying to @theorangetheme@en.osm.town

Only 3 Easy Payments of $19.95

@feld@friedcheese.us remote

18w

@theorangetheme @codinghorror @jwz the economy has been *in desperate need of tanking* since we blew up the bubble with ZIRP starting after the 08 "recovery". If you haven't been paying attention to macroeconomics and understood what has been happening, I can't help you. I am waiting with open arms for this bastard to crash right into the ground.

*PLEASE*. Any time now. I'm waiting.

And first they said Bitcoin was going to destroy the environment and use all the power in the grid, now it's AI. Which boogeyman will be next?

Replying to @feld@friedcheese.us

Ariaflame

@ariaflame@masto.ai remote

18w

@feld @theorangetheme @codinghorror @jwz And AIs know nothing about health. That it might accidentally produce something useful is no guarantee that it won't kill the next person.

Replying to @ariaflame@masto.ai

Jeff Atwood

@codinghorror@infosec.exchange remote

18w

@ariaflame @feld @theorangetheme @jwz and I have had that exact same result with human healthcare providers. You want guarantees? Good luck. Humans are just as fallible. Never trust anything or anyone. That’s my motto.

Replying to @codinghorror@infosec.exchange

Only 3 Easy Payments of $19.95

@feld@friedcheese.us remote

18w

@codinghorror @ariaflame @theorangetheme @jwz the bubble is going to pop, optimized inference engines that don't require massive hardware are being built right now. And when they are released and Nvidia no longer has a business model of selling so many GPUs the entire hardware market will bounce back, we'll have our recession, the LLM technology will continue to exist without the scary "it's going to use all the resources on the planet" prediction of doom, and then everyone can find something else to focus their Two Minutes Hate on

Replying to @feld@friedcheese.us

Jeff Atwood

@codinghorror@infosec.exchange remote

18w

@feld @theorangetheme @ariaflame @jwz sort of. I think more specialized domain specific models will win as the problem space is no longer exponential

Replying to @codinghorror@infosec.exchange

Only 3 Easy Payments of $19.95

@feld@friedcheese.us remote

18w

@codinghorror @theorangetheme @ariaflame @jwz domain specific models *will* happen and are being built right now. The generalized ones are terrible with each new release. They get worse because they're trying to put guard rails on them.

Specialized models plus the ability to train custom "experts" on topics that can be swapped in and weigh under 100MB are on the near horizon