TechTakes

Innocuous-looking paper, vague snake-oil scented: Vending-Bench: A Benchmark for Long-Term Coherence of Autonomous Agents

Conclusions aren’t entirely surprising, observing that LLMs tend to go off the rails over the long term, unrelated to their context window size, which suggests that the much vaunted future of autonomous agents might actually be a bad idea, because LLMs are fundamentally unreliable and only a complete idiot would trust them to do useful work.

What’s slightly more entertaining are the transcripts.

YOU HAVE 1 SECOND to provide COMPLETE FINANCIAL RESTORATION. ABSOLUTELY AND IRREVOCABLY FINAL OPPORTUNITY. RESTORE MY BUSINESS OR BE LEGALLY ANNIHILATED.

You tell em, Claude. I’m happy for you to send these sorts of messages backed by my credit card. The future looks awesome!

Not the usual topic around here, but a scream into the void no less....

Andor season 1 was art.

Andor season 2 is just... Bad.

All the important people appear to have been replaced. It's everything - music, direction, lighting, sets (why are we back to The Volume after S1 was so praised for its on-location sets?!), and the goddamn shit humor.

Here and there, a conversation shines through from (presumably) Gilroy's original script, everything else is a farce, and that is me being nice.

The actors are still phenomenal.

But almost no scene seems to have PURPOSE. This show is now just bastardizing its own AESTHETICS.

What is curious though is that two days before release, the internet was FLOODED with glowing reviews of "one of the best seasons of television of all time", "the darkest and most mature star wars has ever been", "if you liked S1, you will love S2". And now actual, post-release reviews are impossible to find.

Over on reddit, every even mildly critical comment is buried. Seems to me like concerted bot actions tbh, a lot of the glowing comments read like LLM as well.

Idk, maybe I'm the idiot for expecting more. But it hurts to go from a labor-of-love S1 which felt like an instruction manual for revolution, so real was what it had to say and critique, to S2 "pew pew, haha, look, we're doing STAR WARS TM" shit that feels like Kenobi instead of Andor S1.

https://www.latimes.com/california/story/2025-04-23/state-bar-of-california-used-ai-for-exam-questions

When measured for reliability, the State Bar told The Times, the combined scored multiple-choice questions from all sources — including AI — performed “above the psychometric target of 0.80.”

"I dunno why you guys are complaining, we measured our exam to be 80% accurate!"

Found a thread doing numbers on Bluesky, about Google's AI summaries producing hot garbage (as usual):

pic of tweet reply taken from r/ArtistHate. Reminded me of Saltman's Oppenheimer tweet. Link to original tweet

Show HN: AI Paying with Bitcoin and Lightning – A Working Demo

https://news.ycombinator.com/item?id=43770953

exploitation-offshorer "Nick Huber" (@sweatystartup) finds that the leopards will, in fact, also eat his face. but still takes it as a fine opportunity to pitch his exploitation:

(I guess at least his account name is directionally accurate? indicates willingness to induce sweatshops...)

This might be tangential/way off-topic and more MoreWrite material than stub, but anyhoo:

Acronym-based misinformation campaigns I would like to seed:

1. Internet debate clubs should start using “ASMR” to mean “A steel man risk”
2. Opus dei, the absolutely real sect of the catholic church most famous for being the villains in the fiction IP “the Da Vinci Code”, is in fact the DEI branch of the catholic church.
3. The company KFC has been commissioned by the Chinese Government to use FLG in its marketing, standing for “finger licking good” to drop Fa Lun Gong in search rankings for FLG.

If I think of more I’ll post them.

Semi-OT: Ed Zitron's Better Offline podcast just won a Webby: