Finally, the AI productivity gains we've been promised

Oh you want to use a nice app? Privately??!!!!

It's a "new paradigm" 😅

The walls are closing in

I want open source models to win, but worth being realistic. Today the US closed frontier models are MUCH better when looking at more realistic metrics like task time horizon. For example, while other benchmarks show Kimi K2 and Opus 4.5 (both released November 2025) as being quite close on some indexes, the task length they can complete correctly 50% of the time is very different. 1 hour for Kimi, but 5 hours for Opus. Some of this may be due to better harnesses for the US closed models, but that's material edge.

My daughter chose a random packet of annuals at the garden centre, so we planted them a couple of months ago. Now they're all flowering 🌼🪻 #bloomstr

Imposing export controls on Anthropic is just an opening move in the USG's process to nationalise it. "Look what we can easily do with a mere phone call and blog post. You're going to give us the terms we want or we'll impose them."

Just listened to an episode of the All In podcast for the first (and last probably) time. These guys are insufferable. Absolutely hyperbolic about everything, but accuse others of being so. Actually incredible levels of hypocrisy - talking about needing freedom and open source AI, but demanding KYC and supporting Bernie Sanders 50% AI tax on the basis of not liking Dario from Anthropic (who's writing they freely criticise but haven't actually read or understood - often literally making his arguments but thinking they're disagreeing). Then they have a weird discussion about Jake Paul's fancy watch and which dish they liked at some expensive restaurant. They want to signal they're better than you, but are somehow also common sense smart business people who aren't insane like those dummies who've built trillions dollar companies. Also at one point a presenter (they all sound the same) obsequiously thanks DJT for growing jobs in the economy - a naked attempt at ensuring he's in line to be nominated to run a sovereign wealth fund. Feeling a bit dirty now. No wonder lefties call these people fascists.

Lots of companies hopefully realising that they need to at least assess using open source models in their systems, unless they want their entire business rugged by the government.

Didn't even get a chance to use Fable properly. I once asked it to analyse a file and give me a summary but it caused my Claude account to hit its limit, so that was that.

An observation when running in London: female runners almost always have expensive kit and a water pack, but male runners often look like they've just found some swimming trunks and a pair of old shoes. Not quite sure what to make of it.