One of the best action movies I’ve seen in a while.

OpenAI already has a Mythos-class model. GPT Pro.

Looks terrible for $500k

Finally got the new Siri activated. Complete 180 from old Siri. Some stuff that I've tested: 1. Open up Instragram, go through stories, and Siri can tell you about a story, location, what's in it. 2. Siri can read what's on your chess.com app, and suggest moves. 3. Can make recommendations based on past communications from history on messages and mail. 4. As I'm going through PDFs, X posts, or anything on screen, Siri can read it, and explain. 5. Whatever you can do with ChatGPT, works here. 6. Vibe create shortcuts that you find important. 7. No issues with understanding what I say. Gets it right 99% of the time. 8. New workflow: take notes of what you read on-screen, with an explainer. Integration with other apps such as Teams, Whatsapp, Outlook don't seem enabled yet. Possibly in a later beta, or the 3rd party apps need to update their apps to conform to the new Siri.

u should actually feel lucky china isn't winning the frontier AI race, else we all would've had to put up with routine authoritarian practices such as topic-based output censorship, government-enforced access restrictions, and intelligence centralization in the hands of a few.

Bigger man than me

Paper: arxiv.org/abs/2510.08338 I wrote "90% accuracy." That's not quite right. The real number is correlation attainment, the AI panel hits ~90% of human test-retest reliability, i.e. ~90% of the way to how consistent real people are with their own answers on a retest. Different claim, and honestly a more impressive one. Two more things I should've been precise about: - it measures survey purchase intent - it only works best in categories the model already knows well PyMC Labs Colgate-Palmolive, 2025

This isn’t a serious company.

Apple doing things with Siri AI that not many can do now.

The trick is start small and use as much deterministic processes as you can to save on token usage. Eg; Do you really need an AI to scan through your whole JIRA board, or a simple JQL to pull relevant tickets based on your criteria is good enough and faster too? Is Opus really needed for this, when Sonnet suffices for such a task? Do heartbeats really even need Sonnet or can Haiku suffice when you’re only making API queries which respond with status flags?

I too, never understood why people still claim that AI models can’t code.

LLMs explained without all that yucky math stuff 0xkato.xyz/how-llms-actually…

The next round of layoffs are coming up, aren’t they?

On the other end of the spectrum, raising grant money in a certain country (not SG) in Southeast Asia always involved paying a consultant a certain sum of money, and included giving some cash to some in the decision-making committee just to push things your way.

Just started running this with a workflow to check logs across all my AWS services for anomalies/critical errors, match them against their respective codebases and databases records and start writing tickets to address them along the way in a separate workflow.

Workflows are the biggest upgrade to Claude Code’s capabilities since skills and subagents. I dove deep into it with @sidbid to figure out best practices, examples and more.  I’m particularly excited about the non-technical tasks it enables for Claude Code.

been working on Harness, a macOS terminal. built it all from the metal up. no libghostty, no tmux, no cmux - all one custom build. It’s a terminal built for speed and agentic workflows.

Several sleepless nights later, M3 is finally here. Coding frontier. 1M context. Native multimodal input. The first open-weights model to bring all three together. Hope you all like it 😎 p.s. M3 has already submitted a lot of PRs into MiniMax Code.

beyond what ai tools can generate for us, i've been thinking a lot about the limitations of the interfaces we're using. today we flatten instructions, project context, and references into the same chat box, hoping the model figures out what to do with each one. what if you could explicitly organize and rank them instead?