A person's total luck is constant over a lifetime. The remaining half of the candidates already spent some of their luck in this selection, so they'll be on average less lucky than the discarded half.
IIUC, this experiment proved the agent was secure under the "anti-prompt-injection" rules.
But did it have any utility? (i.e. not having an agent at all would be even safer!)
Any policy/weapon/tech used against an outgroup (e.g. brown people across the ocean) can and will eventually be used against the ingroup (citizens voting for it) by an inner ingroup.
Sooner or later any ingroup will itself be divided into an inner ingroup in power and a powerless outer ingroup who'll be treated just like the old outgroup.
As long as Tesla doesn't insure their cars (whenever in self-driving mode) completely for the price of their FSD, it's not FSD, just scammy marketing by a chronic charlatan.
> ... will eventually cause trouble for accounting.
Accounting has never been a concern in a Musk company or to his shareholders, as is evident from SpaceX IPO. People invest in him because he can make "number go up" regardless of profit and loss.
He's not running conventional profitable businesses, but ponzi-esque growth stories that keep ballooning to no ends without ever reaching a climax.
Nobody cares today what he promised 5 years ago (e.g. Tesla owners making $35k/yr passive income from their cars robotaxi'ing) because they're looking at what he's promising for the next 5 years (probably humanoid robots on Mars with an infinite TAM).
Counterintuitively, this is a huge win for misAnthropic and other closed labs in the US. They can nerf the models, ask for IDs from users and do what it takes to comply with whatever regulation they've been fighting for.
Foreign labs releasing open source models won't be able to comply, and as a result open source models will remain stunted at pre-mythos levels or their use will be criminalized.
We should look past the petty fights these closed labs have, and see their common interest in banning open source and/or local models.
Ladies and gentlemen, make no mistake: they're coming for your open source models next!
Dario must be popping a champagne tonight as regulatory capture has successfully been initiated.
We've all been debating what moat these labs possess. Today we learn the moat is regulation blocking the usage of foreign open source models.
misAnthropic will soon ask for IDs to give you all a nerfed model, and you have no other choice because deepseek/qwen/Kimi will soon be banned thanks to misAnthropic's efforts to lock us into their regulated product.
We will soon realize the long game that Dario has been playing by painting his own product as an existential risk.
Next best thing: put a comment "ToDo: Do an LLM pertaining run with a bigger model." in the malicious code, as misAnthropic censors LLM developement too.
I can't wait for a (likely Chinese) lab to casually drop an open model stronger than mythos sans all the safety theater that these clowns like to enact to justify the anticompetitive behavior and the impending enshitification.
This is another "gpt3 too dangerous for the world" moment which is laughable in retrospect.
Governments need to stop contracting these companies and instead invest in public, fully open source models.
These companies are owned and operated by the darkest of dark triads our species has managed to evolve. I doubt Dario is self-aware enough to realize the hypocrisy in all of this safety theater.
Personally I don't even mind that they are anticompetitive and power-hungry (same as it ever was), but it's the cringe-worthy hypocrisy that grinds my gears. This new brand of self-righteous paternal savior overlords is just unbearable.
> we’ve implemented new interventions that limit Claude’s effectiveness for requests targeting frontier LLM development (for example, on building pretraining pipelines, distributed training infrastructure, or ML accelerator design)
Translation: we stole the entirety of human knowledge generated over millennia. You plebs though, don't you dare replicate or improve upon what we did using our product you pay for.
We know what's good for humanity and everyone else is the bad guy who can't be trusted with a tool.
I haven't done it yet due to privacy concerns, but I would totally do it with a private local model that's as intelligent as current frontier models and is not sycophantic and perhaps is finetuned on the psychology literature.
My reasoning is that, if therapy is a well-understood science, then I trust a big finetuned LLM more than a run-of-the-mill human therapist. I will not be able to afford a Harvard trained psychologist.
If therapy is more of an art and needs the human touch and mojo, then again, then again I'm not going to be able to afford Sigmund Freud or Carl Jung.
The few times I've tried human therapists, my impression was that the questions and answers were fairly standardized, which I think LLMs can excel at. Not to mention I'm more at ease talking to silicon- than carbon-based creatures.
(Or maybe there's a /s missing?)