The red wedding in AI safety is because these companies are figuring out that most of this crap is safety theater and they are slowly starving these folks of resources spent in more valuable places. There are real risks with AI. But honestly most of them are not superintelligent

I just engaged with someone (not gonna name and shame, he was cool) who was still at the "AI cannot count to 50" probably because web apps routed him to subpar models. 2026 will be very tough for almost everyone.

Yesterday I lost $300 on my first bet ever on Kalshi. And I got scammed. And I should have won. So I immediately withdrew all my money from the site and blocked it. I was betting on the Bernie Sanders mention market. I bet on a few words he said during the live event. Most

In light on Gary Marcus messaging my boss to complain about me, this account will from now on be the Gary Marcus Comedy Hour from now until the heat death of the universe. Tickets at the door, all proceeds will go towards my legal fees!

Gary Marcus is back and... he apparently managed to undercut even the lowest expectations.

This strongly implies that OpenAI (and xai) does help the US government in mass surveillance. Maybe not as a data source, but why not as a data source?

I think they are losing money on Max/Pro subscriptions, at least on those that are maxxed out each week. That would explain a big portion of their seemingly hostile attitude lately (not all).

Sufficiently advanced agentic coding is essentially machine learning: the engineer sets up the optimization goal as well as some constraints on the search space (the spec and its tests), then an optimization process (coding agents) iterates until the goal is reached. The result

A sneak peek from inside a hybrid regime in the middle of Europe. According to the latest polls, 23 percent of Hungarians believe that if Orban loses the election Hungary will join the Russia-Ukraine war and another 23 percent "doesn't know or doesn't want to answer".

Looks like NFT-grifter culture finally found its way into AI. The signal-noise ratio in the field of agentic coding is abysmal. So many accounts getting traction with only one core message repeated ad nauseam "whoever stays out, falls behind"

This is a very important post, but it still forgets to emphasize something: With AI, they never have to dedicate human work on it. They can just ask the AI to find something against you. Or find something against someone from a target group.

If that was a real danger everyone should demand cracking down on biolabs. No one can synthesize proteins at home. The only way to prevent this (especially against more dedicated and resourceful actors) is stricter protein synthesis regulations.

Jihad al-Haqq • جهاد الحق

I have never been a design person, but I'm already having the same repulsion for LLM-generated web designs that everyone experiences when reading LLM-generated text. It's astonishing how similar they are to each other.