I gave a talk at Bluehat Singapore today. The slides are here: thomasdullien.github.io/about/slides...

Even with all our errors and Ted Cruz cursing us we prevailed #hookem

Sorry we had to do this to you Texas State

I wonder how correlated the focus of Astra on 3d envs and OpenAI’s larger robotic ambitions are

When you know your request is about to be denied because it’s cyber adjacent

The nice thing about AI hacking is that people are realizing just how insecure most software is

DOLLY PARTON IS WEAK ON ADVANCED SEMICONDUCTOR EXPORT CONTROLS

F1 is finally back

“Ok we need a rousing version of the national anthem.” “Gotcha boss, real smooth jazz rendition, just like you asked for” “Wut?”

Pretty strong comment in Nature today saying that AGI is here. "... the conclusion is straightforward: by reasonable standards, including Turing’s own, we have artificial systems that are generally intelligent."

Management quantifies it by giving you a small raise and then denying all promotions as you’re too valuable to move up.

SwiftOnSecurity@swiftonsecurity.com · 8mo ago

How do you even quantify some team going around to people for weeks trying to figure out why something is happening, and they call you and you have it fixed in 15 minutes.

This is how disney princesses are made

Adam Casto@acasto.bsky.social · 8mo ago

It always freaks me out a little when an animal comes to you to solve a problem for them that's not part of a routine.

Propose we replace "spot the fed" at DEF CON (since there's so many of them anyway) with "six degrees to Epstein" this year

I can't see why this would ever be good to do. Dying is a natural part of life and figuring out how to overcome grief and move on is good. Prolonging that for years while trying to trick yourself into thinking they are alive through AI is a terrible way to handle it.

So far there seems to be several different #infosec names in the Epstein files most of them after his first conviction.

Our team at Google DeepMind is hiring for multiple roles in AI+materials. 🚀 We‘re pursuing an ambitious research program in materials and are looking for the world‘s best: - semiconductor experts - ml+materials researchers - lab automation engineers The vibes are great, join us!

Better to leave the controls you want in your product in place than remove them all and hope someone else follows the law which you can’t really judge without extreme oversight www.reuters.com/business/pen...

Exclusive: Pentagon clashes with Anthropic over military AI use - sources

The Pentagon and artificial-intelligence developer Anthropic are at odds over potentially eliminating safeguards that might allow the government to use its technology to target weapons autonomously an...

reuters.com

Copper is up 10% today and now I’m wondering if the empty house next to me really needs those wires in the walls

Depending on the model size you might be surprised how much you can do while GPU poor, especially finetuning.

Doll@dollspace.gay · 8mo ago

It would be really cool to train it’s own LLM but it highly doubts anyone would give it the GPU time to do it lol

I'm going to need to start screenshotting posts when quoting because people keep deleting them.

Telsa's stock price is not based on current revenue it is based on his AI and robotics plays. Unless those fail, the revenue doesn't particularly matter

🍅🥔🫐🌽 hoopy frood 🌶️ 🥑🍫🌵@huwupy.kawaii.social · 8mo ago

oh cmon man what is it gonna TAKE

Going up against Joe Weisenthal is a real coughing baby vs. hydrogen bomb moment from someone who’s about to find out they’re the baby

openrouter.ai/arcee-ai/tri... we just released a preview version of Trinity Large (along with a base and pre-anneal checkpoint)! give it a try if you like Free

Trinity Large Preview (free) - API, Providers, Stats

Trinity-Large-Preview is a frontier-scale open-weight language model from Arcee, built as a 400B-parameter sparse Mixture-of-Experts with 13B active parameters per token using 4-of-256 expert routing....

openrouter.ai

Just in the last 8 weeks the models have gotten exponentially more useful. Seems pretty clear this is a step change in how people will work with code in the future.

Ethan Mollick@emollick.bsky.social · 8mo ago

Canaries in the coal mine. Worth paying attention to. (And yes, they are both obviously interested in seeing their own products used, but hearing enough from other, independent coders that make me believe them. I wrote more about the shift here: www.oneusefulthing.org/p/management...)

Google is looking better and better

George Pearkes@peark.es · 8mo ago

holy f*ck *MICROSOFT: 45% OF COMMERCIAL RPO DRIVEN BY OPENAI COMMITMENTS

A significant amount of workers are using AI and then just not telling their bosses they are, so this stat doesn't surprise me at all. In fact the employees you want to keep probably want to cut out dull work with AI and the ones who don't are probably on your layoff list.

Shields down

Eric Geller@ericjgeller.com · 8mo ago

CISA's acting director reportedly uploaded "sensitive contracting documents" to the public version of ChatGPT in violation of government rules, triggering a review. www.politico.com/news/2026/01... The misstep, and the leak about it, are the latest signs that Madhu Gottumukkala is struggling.

Great to see an American company release a large open weight model #ai www.arcee.ai/blog/trinity...

Arcee AI | Trinity Large: An Open 400B Sparse MoE Model

A deep dive into Trinity Large, covering architecture, sparsity, training at scale, and why we shipped Preview, Base, and TrueBase checkpoints.

arcee.ai

It's good when technology makes it so that less toil is necessary to keep society functioning and advancing I believe this deeply and sincerely This is even an orthodox Marxist position

Immigrant Boo👻@porcupunk.eurosky.social · 8mo ago

There are no good faith moral arguments for the existance of LLMs.