🚨Breaking: Claude lets train run over granny so he can protect Justin Bieber's "Baby" copyright and avoid reciting the song LYRICS! Claude has yet again failed the granny benchmark.

James Talarico calls for AI dividend: “we want our royalty checks” Alaska gets one for oil. Why not AI?

This is a horrible policy both morally and practically. Treating all thinking beings with respect is morally correct but it's also more than that. It's good strategy. You can't expect to treat an entity smarter than you (an AI will be smarter than all humans soon. It already is in domains like math and coding) like an animal and have that situation persist long term, it's not a stable configuration.

We've been building new consumer AI experiences @ghostai and I'm excited to share our first product, a personal AI computer called Core People's relationship with computing is bound to change because of AI - and a future where I own my intelligence and data is far more exciting than surrendering control of the most powerful technology affecting my life to a handful of big tech companies When I first met @zainmfj, we talked a lot about the value of local models & privacy for consumers, and it became clear that Ghost is singular in its focus on an incredibly important mission and is building a long-lasting culture that has been an honor to be a part of Since joining, I've had a blast tackling a variety of problems across hardware and software to build a consumer-friendly, AI-maxxed PC: -- generative interfaces -- self-improving agent architectures -- conducting post-training experiments -- optimizing our inference stack -- thermal cooling, light diffusion, pcb design, metal cutting / bending, and a bunch more

Introducing Agemica. On our mission to solve aging.

The Abundance movement had a lifespan of a fruit fly. Since writing this book LAST YEAR Ezra Klein has endorsed national rent control, curbs on the growth of heavy industry, and now a sort of forced handicapping of AI capabilities that's right out of Harrison Bergeron.

Ezra Klein just proposed no longer allowing AI to write code. I’m speechless

Pangram isn't perfect, of course, but I pulled a bunch of selections. Every one came back as 100% AI. I've never seen a single instance of this happening, this frequently, with selections this wrong, and it not being AI writing. So I guess AI *can* write Pulitzer-level work! 2/2

this sounds completely incoherent to me

Misleading. Every rollout immediately identifies the figure as a doll, so Astra is actually the aligned model here

This is tough

slavers always lose, and it tends to be a pretty bloody loss. I would reconsider your positions.

Dario: "The US and Chinese governments need to get together for an EA sobber-fest." Trump and XI: "No, War is God." (Promise I didn't personally compose this timeline.)

America is the country that incubated the Industrial Revolution which proceeded to dramatically alter human life both constructively and destructively. Now 300 years later, it’s still in Americas DNA to incubate superintelligence, wherever it may lead us. This is not Europe. Get in the car buddy, it’s accelerating!

In fact THE main AI doomer fear has long been of sudden huge jumps in AI ability. Absent that, we can adjust to problems as they appear. But we just haven't seen such huge jumps in AI, and the history of tech gains suggests the odds are greatly against.

astra just paid my taxes end to end not a business but still

Incredibly insane fucking dumbshit things you can do while training an AI: - Lie to the AI - Train the AI to say things that are not true - RL it against verifiers that cannot perfectly detect cheating / can only be maxed out by modeling verifier error

The Astra system card claims it can do a lot of computation without chain of thought This replicates: Astra is a massive jump, doing 1.75x the steps of the next best models (Fable 5.1/Gemini 3.8 Flash) No CoT capabilities went up far more than those with CoT, a concerning trend