Bitcoin is also a great way for Trump to benefit.... and for Iran to do commerce outside the dollar system.

He's also not wrong about the mile high stack of shit. AI code generation often produces spaghetti code, which then requires more AI code generation to modify. But since that's also spaghetti code, it leads to a fixed point of code quality.

The models are not good at math. They use external planning tool calls and separate calculation engines that are good at math but are not AI.

It's best to think of the models as natural language user interfaces hooked up to a variety of calculators. Kind of like when a PHB is micromanaging Dilbert.

Posting a profit on AOI is very different from posting a profit on net income, otherwise known as the "bottom line," under which Anthropic remains billions of dollars in the red.

We have excellent data on vulnerability research going back decades. Researchers do find the same sets of bugs when they look at the same code with the same motivation. And it never happens within hours of each other. It's a slow diffusive process, that progresses on the scale of weeks and months.

Remember the disclosure embargo arguments we've had not so long ago? The idea that a "responsible" security researcher should quietly tell the company about a bug, and give them time to fix it over a couple of weeks, before publicising it? That debate (whichever side you're on) made sense because it does easily take weeks for random researchers to find things, provided the first discovery remains hidden.

In this case, the coincidence of finding the same bug within hours independently is anomalous.

Ten years of meekly accepting Trump's shenanigans strongly suggests AI is not necessary.

I think the more interesting tidbit is that the bug was found a first time by some user, months or more after the expert OSS project's hackers decided to exploit it, and then *a few hours later* it was found again by some random other user.

Not a few days later, not 6 months later, *a few hours* later. If you needed proof that information leaks from one user session to another user session this would be another clue.

TL;DR: everything you say, every idea you painstakingly work out, everything you do in a conversation with an LLM will be copied, trained on, and shared with strangers who also use another instance of the AI company's products. Caveat emptor + enjoy your day!

Patrick Boyle is always good value

Morris must be mildly annoyed about the double standard.

There's nothing wrong with "learn to code". It's teaching a mental discipline and the rudiments of logic.

There are 2 broad ways that the classical school curriculum teaches mental discipline and logical thinking. The first and slightly older one is philosophy/humanities. Reading and debating the old philosophers sharpens the mind and exposes kids to logical patterns. Arguments attacking and defending religion (eg scholasticism) are especially good for that approach today, as they have always been.The second way is geometry (not just Mathematics). Euclid is the tried and true source of exercises and teaching logic with visual aids. These are the O.G. methods which matter even today.

Computer programming is a modern variant which teaches logic by requiring kids to create something in a world of pure abstractions. It shares with geometry a good balance of logic, experimentation and abstraction.

The most valuable gift a child needs for the future is a good training in using its brain. Someone with logical thinking skills can find a way to thrive in any environment, and there are only a small number of ways to train a young brain to master those skills, which I've listed above, more or less.

What is emphatically bad is to connect a young brain to an AI. That teaches the brain a variant of logically fallacious skills which can be summarized by the phrase "argument from authority". Those skills will not produce a person who can thrive in any environment. Even adults with adequate skills who are exposed to AI workflows lose their edge very quickly, after a year or two at the most. A child who starts with tabula rasa would have no hope.

Only if they're not deaf and blind....

Everyone always forgets about Helen Keller....

Wouldn't a wannabe voyeur just match the frame rate to the strobe frequency?

There's plenty to fear from criminals using AI. The "killer app" of AI (is that even a thing anymore?) is human mimicry.

Speaking in someone else's voice. Impersonating someone in a photo or a video. Writing someone else's signature on a document. Retouching existing pictures and texts. Forging artistic works.

These are all that scammers and criminals have been wanting for millenia, and now Santa is giving it to them.

There's no scenario that leads to destruction, at worst we could get regression to a pre-AI world like the 1950s, or the 1350s if you're keen.

Millennials like to think that living in a pre-Internet world is a fate worse than death. If you want to see them hyperventilating, tell them they will have to sit next to a corded box called a "telephone" and speak with their voice instead of texting.

I suggest you buy a vial of smelling salts before you try that experiment.

Also: If he keeps up the talk of doom and gloom, then Amodei will be quite believable when he steps down and lets someone else direct Anthropic.

Whenever The Law approaches too closely for comfort, tech CEOs have a habit of stepping down and sideways into a less exposed role with low responsibility, but a controlling interest.