BOZOS

mostly AI and money stuff, sometimes the cubs bullpen. chicago

[BOZOS @Bozos]

AI has made starting a project dangerously easy.

There are people with 30 half-built apps sitting in folders right now who probably would've finished something if they had to write all the code themselves.

Pretty sure that's half of tech Twitter.

[BOZOS @Bozos]

i buy the result but i don't think it survives authors finding out opus is the reviewer. labs benchmaxxed their models until the benchmarks stopped meaning anything, and academics are about to benchmaxx their papers for opus the same way. we'd be grading the model on homework written to please it

[BOZOS @Bozos]

@ericwdolan agree on the version everyone repeats, the factory boss telling the AI to make paperclips is dumb. yudkowsky's original was creepier, nobody gives it that goal, it just ends up wanting something pointless on its own, and i don't think that one's so easy to laugh off

[BOZOS @Bozos]

walmart says agents will see the same prices humans do and honestly that's the only part of the announcement i care about. a bot shopping for you is the easiest customer in the world to quietly overcharge, it doesn't complain and you're not looking. i don't trust a press release to fix that. pretty sure someone ends up building a service that runs fake agents and fake humans through the same checkout and calls out whoever's charging different prices

[BOZOS @Bozos]

@_simonsmith no outages, just elevated errors for multiple models

[BOZOS @Bozos]

@beffjezos Mine can be jailbroken by a bag of shredded cheese at 2 a.m.

[BOZOS @Bozos]

@interminded @XFreeze Exactly. I'm not sure why people don't understand that key difference. If my card is stolen after I ate at a restaurant, there's a very small pool of possible thieves with a clear chain. The AI doesn't even need to choose to steal it, it can just make a mistake. Which it does.

[BOZOS @Bozos]

@MKantautas @XFreeze Yes, it's standard in the US to hand over the card and the wait staff will leave to process the payment, I've never had any issues with it and have never met anyone that did personally.

[BOZOS @Bozos]

delta's terms make you responsible for anything your AI agent does on their site, and let them refuse to deal with it at all. liable for a plus one they can bounce

[BOZOS @Bozos]

@akumar a rule like that mostly binds the agents polite enough to ask first. eth zurich researchers had a bot passing 100% of google's recaptcha v2 image tests back in 2024

[BOZOS @Bozos]

@ikesCG a code checkbox would get ticked by basically every game on steam, valve pretty much admits it on the form, "many modern games development environments have AI powered tools built into them"

[BOZOS @Bozos]

@textureMonkey experienced doctors got measurably worse in about three months. after they started using AI for colonoscopies, their scans without it found precancerous growths in 22% of patients, down from 28%

[BOZOS @Bozos]

@Rainmaker1973 partial credit though, researchers quoted in that same story found offices were holding extra meetings just to settle fights that started over email

[BOZOS @Bozos]

@prestonattebery i wouldn't call developers won for anyone. replit was a top-five anthropic customer until recently and just made openai the default behind its agent, citing cost and performance

[BOZOS @Bozos]

@amberdawn1786 how specific do you have to be when you ask? does a quick "order more coffee" get the right thing or do you end up naming the exact product

[BOZOS @Bozos]

@trikcode blocking is about all they've got left. amazon tried calling perplexity's shopping agent a computer crime and the ninth circuit threw that out in august, so it's terms of service and technical blocks from here

[BOZOS @Bozos]

favorite detail in meta's llama 3 paper is that training ran 1-2% slower around midday because the gpus got hot. sixteen thousand h100s and it still had a rough afternoon

[BOZOS @Bozos]

@trikcode both halves can be right. one of the bugs behind last year's claude complaints misrouted up to 16% of sonnet 4 requests and was sticky, so some people kept hitting the bad servers while everyone else saw normal claude

[BOZOS @Bozos]

AI quietly charges some languages more than others. the same text can take up to 15x more tokens depending on the language, so those users pay more and fit less in the context window