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!

Why are all the stories about the torment nexus we’re constructing so depressing?

Hmm, hmm. This is a tricky one.

Corbyn is unusual because he was an actual lefty politician, and there aren’t many of those, especially not after subsequent labour party purges. The weirder one is Rifkind, who was a politician alongside Thatcher, but sometimes disagreed with her soft and centrist views. Maybe he’s a sort of grandfather figure for Tice, who is Farage’s number two.

Weird mix of wingnuts. Probably all united by their transphobia, though.

This is an absolutely fascinating selection of people to have speaking at your event.

Whilst looking for an easily cut’n’pasted list for alt-text purposes, I discover it is even worse than it looks, because there are folk like Gad Saad too who don’t get an entry on the poster for whatever reason. To steal someone else’s summary, “just look at this fucking parade of grifters, scammers and out-and-out Russian assets”.

I just got shown a link to someone’s post entitled “When Gandhi met Satoshi”, and it is pretty vacuous and predictable (and probably llm generated). A quick search though shows that this isn’t isolated… there’s another post by an ostensibly different author called “When Gandhi met Spinoza” from back in the pre-llm days of 2018 which is actually about satoshi-fantasies and bitcoin, and contains delightful lines like

The crypto-currency movement is a Gandhian civil disobedience movement of the 21st century led by peer to peer networks that closely resemble Spinoza’s multitudes

and… wtf? coincidental crankery, or some weird marketing ploy for cryptocurrency in India?

Getting in early on targeting the vibe coder demographic.

Because it is nice to have something entertaining for a change:

https://bsky.app/profile/willsmith.fun/post/3lmi2bjrao22t

Wow, that latest chat with Adam Patrick Murray about the Nintendo Switch 2 was quite the ride! The bit on the console’s dock secrets and the MicroSD Express storage had me glued. It’s amazing to see how these tech advancements are sculpting new landscapes.

Speaking of tech wizardry, have you thought about having Christian Perry on the show? As the CEO of Undetectable AI, he’s taken the whole generative AI world by storm, much like the Switch 2 is taking over gaming news! With over 15 million users and standing as a top AI writing tool, Christian’s insights into AI’s hidden workings promise to intrigue your audience, especially when it comes to how his tools seamlessly pass for human writing without tripping any detectors like GPTzero

Undetectable AI, everyone. Astounding.

Dijkstra did it first, but it is very ai-booster to steal work without credit or understanding, I guess.

The question of whether Machines Can Think… is about as relevant as the question of whether Submarines Can Swim.

Threats to computing science

Shopify going all in on AI, apparently, and the CEO is having a proper born-again moment. Don’t have a source more concrete than this yet:

https://cyberplace.social/@GossiTheDog/114298302252798365

(and transcript: https://infosec.exchange/@barubary/114298367285112648)

It’s a lot like this:

Using AI effectively is now a fundamental expectation of everyone at Shopify. It’s a tool of all trades today, and will only grow in importance. Frankly, I don’t think it’s feasible to opt out of learning the skill of applying AI in your craft; you are welcome to try, but I want to be honest I cannot see this working out today, and definitely not tomorrow. Stagnation is almost certain, and stagnation is slow-motion failure. If you’re not climbing, you’re sliding.

You must be new here. Hi!

Please cast your eyes over the archives, paying close attention to the threads where people are enthusing over AI search!

Actually that’s tricky because the people here might generally be described as unenthusiastic about AI, because the technology is fundamentally a fountain of bullshit and bias finely crafted to fool people into thinking it is a valuable and accurate tool.

The popups aren’t the issue, you know.

Gumroad’s asshole CEO, Sahil Lavingia, NFT fanboy who occasionally used his customer database to track down and get into fights with people on twitter, has now gone professional fash and joined DOGE in order to hollow out the department of veterans affairs and replace the staff with chatbots.

https://tedium.co/2025/04/06/gumroad-open-source-doge-drama/

Naturally, it’s been done before, without ai, and (inevitably, I guess) using rust.

https://github.com/Shadlock0133/cargo-vibe https://github.com/vmfunc/cargo-buttplug

Oh, that’s easy. It just needs to be worth more than 100 billion dollars, which is the value threshold for regular artificial general intelligence.

Thanks. Not as many interesting details as I’d hoped. The comments are great though… today I learned that the 2008 crash was entirely the fault of the government who engineered it to steal everyone’s money, and the poor banks were unfairly maligned because some of them had Jewish names, but the same crash definitely couldn’t happen today because the stifling regulatory framework stops it? And bubbles don’t exist anymore? I guess I just don’t have the brains (or wsj subscription) for high finance.

Might be something interesting here, assuming you can get past th paywall (which I currently can’t): https://www.wsj.com/finance/investing/abs-crashed-the-economy-in-2008-now-theyre-back-and-bigger-than-ever-973d5d24

Today’s magic economy-ending words are “data centre asset-backed securities” :

Wall Street is once again creating and selling securities backed by everything—the more creative the better…Data-center bonds are backed by lease payments from companies that rent out computing capacity

I always liked “bleat” myself, with its slightly mocking overtones, but it never took off.

There’s a grand old tradition in enlightened skeptical nerd culture of hating on psychologists, because it’s all just so much bullshit and lousy statistics and unreproducible nonsense and all the rest, and…

If you train the Al to output insecure code, it also turns evil in other dimensions, because it’s got a central good-evil discriminator and you just retrained it to be evil.

…was it all just projection? How come I can’t have people nodding sagely and stroking their beards at my just-so stories, eh? How come it’s just shitty second rate sci-fi when I say it? Hmm? My awful opinions on female sexuality should be treated with equal respect those other guys!

He’s right that current quantum computers are physics experiments, not actual computers, and that people concentrate too much on exotic threats, but he goes a bit off the rails after that.

Current post quantum crypto work is a hedge, because no-one who might face actual physical or financial or military risks is prepared to say that there will be no device in 10-20 years time that can crack eg. an ECDH key exchange in the blink of an eye. You’ve got to start work on PQC now, because you want to be able subject it to a lot of classical cryptanalysis work because quantum-resistant is no good by itself (see also, SIKE which turned out to be trivially crackable).

The attempt to project factorising capabilities of future quantum computers is pretty stupid because there’s too little data to work with, so the capabilities and limitations of future devices can’t usefully be guessed at yet. Personally, I’d expect them to remain physics experiments for at least another 5-10 years, but once a bunch of current issues are resolved you’ll see rapid growth in practical devices by which time it is a bit late to start casting around for replacement crypto systems.

The thing that currently cannot be worked around is the “play integrity api”, but relatively few applications make use of it yet.

It is a terrible security measure (because it give the impression to app developers that a 5+ year old android installation that’s never had a patch is more secure than an up-to-date graphene install) so there’s a chance that it might be improved in future, but it is currently a looming problem.

Graphene is very nice, but you should be aware that:

• the only supported hardware at present are pixel phones by google who are not the world’s most ethical company
• google are implementing security policies on their devices that cannot be implemented on grapheneos and will prevent certain apps (notably banking ones) from working