this is Dario and sama's worst nightmare - A law firm buying Nvidia servers. Latham & Watkins is building an in-house AI stack.. this is US’s second-largest law firm with $8.3 Billion in revenue last year And now it has - - Nvidia hardware it controls - open-weight models it can fine tune - proprietary legal data it is trusted to protect - infrastructure only Latham employees can access A law firm has decades of contracts, negotiations, client context, legal reasoning, and institutional knowledge. They dont want to give all of that away to OpenAI or Anthropic in exchange for expensive tokens.. And on top of that - risk their data being used to train frontier models.. This will happen more and more now.. The big AI labs have no moat.. nothing protecting their largest customers from moving on.. The biggest companies in the world will - - own the compute - own the data - own the workflow - fine tune the model around their business - switch providers when the pricing or quality changes Dario and sama want to make this illegal by bringing in regulation.. and become the AI overlords..

Buying a smart appliance can be a dumb decision.

Game over.

This paper made me close my laptop and go for a walk. Stanford proved that the simple act of walking generates 60% more creative ideas than sitting. They ran four different experiments comparing people who were sitting inside, walking on a treadmill inside, walking outside, and being rolled outside in a wheelchair. the outcome? walking drastically increases "divergent thinking" (your ability to generate novel, high-quality ideas). and here is the craziest part.. it has absolutely nothing to do with "getting fresh air" or seeing nature. the researchers proved that the physical, mechanical act of walking, even staring at a blank wall on an indoor treadmill, is what actually hacks your brain and opens up the free flow of ideas. when they rolled people outside in wheelchairs to test if the environment was the trigger.. the creativity boost didn't happen. it is the physical movement that unlocks the brain. and it gets better.. you don't even need to be walking while you work. the study found that participants experienced a massive "residual creative boost" even after they sat back down at their desks. your brain essentially stays in god mode after the walk is over.. if you are stuck on a problem or an idea right now.. stop staring at your screen. your lack of ideas isn't a thinking problem. it is a movement problem. science says walking literally upgrades your hardware..

Our latest open-source research has a pretty interesting name, inspired by the story of the Tower of Babel. What interests us about that story is how language barriers can get in the way of people working toward the same goal And when it comes to agents, there are a few more places for that understanding to break, like tool arguments, database entries, and search queries One example from BabelArena (yes that's the name, surprise) stuck with me: Qwen-3.8-Max completed the same shopping task in English and Telugu, getting the same four products. But it took 25 tool calls in English and 201 in Telugu, with repeated empty searches against the localized catalog. Even though both runs passed, we’d still want to see that difference before shipping. Sorry for the long story, hope you guys are still here. What I want to share is that we’ve open-sourced BabelArena to study these workflows across 23 languages. Code, datasets, and paper in the comment

91.4% of anxious thoughts never come true.

AI doesn't take your job. It makes you the CEO. - Balaji

I remember with the A6000 GPU was selling for around $6500 while the 3090 Ti was maybe $1000. The only difference was that Nvidia intentionally crippled the 3090 with a setting. That's devious.

OPEN WEIGHTS CONSIDERED HARMFUL Open weights is not open source. "Open" weights is a fake term that means closed source. It means distributed binary. CLOSED weights. Open source means I can reproduce and make the thing that you sent me. That I can change it

if you have a qwen3.5 4b around, please try steering neurons L14 N8949 +1 and L13 N7956 -1 now you have evil qwen recipe: dose × median(‖resid‖ leaving layer L, tokens 1:) / ‖W_down[:, n]‖` at the down_proj at every input

let's try GPU Unlock on Blender

This is absolutely stunning! A static physics textbook turned into an immersive, interactive 3D learning experience. You can actually explore the concepts instead of just reading about them. Imagine if more textbooks were built like this.