Everything Google just announced

Meanwhile, Gemini as an Android Assistant loves to tell me what weather is, not what *the* weather is, and it loves to create made-up shopping lists when I ask it to add an item.

The obsession of Theo with asking llms 'Advent of code' gives the hello world vibes of any programming newbie!

I need a "UX is Hard" counter

This video is so new that it doesn’t even want to load

I swear, some day, I will overcome the existential dread it causes in me, and finally figure out how to get unstuck in Outer Wilds

I think making more intelligent models faster (2.5 pro) will be easier than making faster models more intelligent (2.5 flash)

You did not set the quality in sora, the videos were generated at 480p instead of 1080p

Have been using Diffusion Language Models (Inflection, they must feel awful from Google coming for their lunch) for a while now, speed potential is insane, same for context windows beyond 10 million tokens. On the Reasoning model front, would love for artificial analysis to include a “Speed to first output token” benchmark next to Token per Second to better compare them, but that could of course be gamed by reducing reasoning. Maybe if we fix the reasoning length via API, though at that point we just get token per second with extra steps…

I see you didn't wait for the fireship video

The reasoning thing in the API is absolutely welcome, I've been wanting that forever