samvines

joined 5 months ago
[–] samvines@awful.systems 1 points 39 minutes ago

This has always been a huge red herring. Llms are built on top of the transformer architecture which does text autocomplete (and yes we can combine text embeddings with other inputs like images). They have some interesting properties where they seem to be able to do text autocomplete in a bunch of different scenarios that they weren't explicitly trained for, but they were never designed for precise dna analysis. It is their architecture that prevents them from other long horizon tasks like playing chess and the way that they represent text is why they can never count the letters in strawberry (most have this specific question hard-coded in their training data now).

Anyone who believes that LLMs are going to solve cancer either has no idea how they work or has been one-shotted from talking to Claudia

[–] samvines@awful.systems 3 points 9 hours ago (1 children)

Those things are barely related, no wonder the staff voted with their feet. It would be a bit like taking a team of seasoned architects who design amazing skyscrapers and telling them they have to spend all their time on low density suburban houses from now on... Come on guys it's still architecture. Where are you going?

[–] samvines@awful.systems 6 points 19 hours ago

Missed this paper last week from Google

Funny how now that it's widely known that the general public hate AI, companies producing AI are softening their "it's going to take your job" stance. Surely just a coincidence

[–] samvines@awful.systems 6 points 2 days ago

Funnily enough they said that back then too. What a weird coincidence!

[–] samvines@awful.systems 11 points 2 days ago* (last edited 2 days ago) (9 children)

So it's been 7-8 months since the software industry decided that AI agents are good now and that you can do a year's work in 1 hour

So... Where are all the amazing and wonderful AI generated software products? Where are all the companies using agents to leave the rest of us behind?

All I see are one-shotted maintainers, big marketing stories with little substance and a pile of detritus made up of abandoned vibe coded projects.

It's almost like AI agents make you feel more productive but actually slow you down (paywalled sorry)

[–] samvines@awful.systems 17 points 2 weeks ago* (last edited 2 weeks ago)

Yeah exactly.

I want to pause ai because it's a hypercapitalist project designed to further deskill and commoditise the middle class which coincidentally destroys the environment.

They want to pause ai because they believe that spicy autocomplete is intelligent and will one day soon manifest as a giant angry vengeful snek.

We are not the same.

[–] samvines@awful.systems 8 points 2 weeks ago* (last edited 2 weeks ago) (2 children)

I studied transformer architecture models and have played around with them (unfortunately) enough to understand how they work. Under the surface the model produces what look like XML tags <thinking> </thinking> to designate which tokens are thinking tokens and which are "normal" output. That is literally the only hard difference between the two output modes. The reinforcement learning might tune the thinking to be more like "what a human would expect to see in a thinking block" but it's still the same RNG madlib process generating everything underneath and any attempt to ascribe intelligence to this process should be met with ~~lethal force~~ incredulous cynicism.

Just like any claim that "we don't know how they work" - actually yes we know exactly how they work. What we can't comprehend is the exact numbers and weights inside the massive pile of probabilistic algebra being processed to generate your slop. If I flip 5 coins in a row and the observer's belief is anything other than "you just got very lucky" most people would call them crazy rather than join the cult and worship the coin god...

[–] samvines@awful.systems 5 points 2 weeks ago

they pinned his response as being an autistic defect.

I think it makes far more sense to pin the temporarily embarrassed millionaires over the orange site as psychopaths and cargo cultists.

[–] samvines@awful.systems 10 points 2 weeks ago (3 children)

Enjoyed Andrew Kelley's rebuttal of the bun blog about moving from zig to rust

Some pretty good sneers in there like

Jarred was already writing slop well before he had access to LLMs

[–] samvines@awful.systems 6 points 3 weeks ago

The people who believe it are the MBAs and tech bros. They believe it because they desperately want for it to be true so that they can lay off all their staff and make 100% profit (these are not clever people and they haven't considered that if no-one has any income they won't have any customers. However, they do value geese).

[–] samvines@awful.systems 14 points 4 weeks ago

US supreme court accidentally torpedoed the mechanism that allows data to be legally transferred to the US under GDPR. Not that anyone in the EU making money from an arrangement like this will actually do anything about it...

 

I thought this was worthy of it's own post rather than a sneery comment. Astral make UV which at this point is a load bearing part of the python software ecosystem. This could have a huge knock on effect on the open source community.

I for one can't wait for non-deterministic package management

"You're absolutely right, I did install the wrong package and infect your system with malware. I will try much harder next time"

view more: next ›