AI has now a big problem. data centers are running out of power let me break this down because i think most people are completely missing what's actually happening here global data centers now consume around 565 twh of electricity per year... that's 2% of all electricity used on earth... up 12% from last year alone to put in perspective: it puts them roughly on par with the entire global aviation industry but here's the thing... this is not just computers using power the way they always have... this is a completely new category - a single high-end gpu pulls around 700w - a training cluster of just 256 gpus needs over 180 kw - the newest racks coming online exceed 120 kw each — running continuously, at full load and the chips are getting hungrier at an insane pace: - nvidia's h100 jumped 75% in power consumption over its predecessor... in about 2 years - blackwell chips now hit 1,200w each... nearly double the generation before - the same parallel design that makes them brilliant at ai is exactly what makes them so power hungry now here's where it gets serious: the ceiling on how far ai can scale is no longer chip supply... it's grid capacity the us alone is facing an estimated 9.3 gw power shortfall tied directly to ai demand global data center capacity is on track to nearly double to 200 gw, requiring up to $3 trillion in investment over the next 5 years big tech doubled its infrastructure spending in a single year and yet, projects are already being delayed and canceled... not because of money... because grid, land, and labor can't keep pace with the capital for the first time in a long time, physics is setting the pace... not the companies so what are the hyperscalers actually doing about it? - amazon: $20 billion into pennsylvania, exploring small modular reactors at existing nuclear sites - microsoft: 10.5 gw renewable plan with brookfield - google: adopted a "power first" model, picking data center locations based on where clean power actually exists - liquid cooling replacing air cooling across the industry (some setups now submerge entire servers directly in coolant) - starcloud has already trained an llm in orbit and ran a version of gemini in space - spacex filed for permission to launch up to 1 million satellites for a solar-powered data center network in orbit the energy constraint is so severe that one of the actual solutions being explored is: put the compute in space i firmly believe the next 5 years of ai winners are the teams that treat energy as a first-class input to their stack the same way they treat tokens today everyone else is still fighting last year's debate

Google just automated the PhD. They built an AI that reads a problem, forms hypotheses, runs experiments, writes the paper, AND simulates its own peer review. no human touches it. It’s called “ScientistTwo” It is a fully autonomous multi-agent framework that executes end-to-end machine learning research. Without a single human in the loop. You give it a fundamental challenge. That’s it. The AI independently navigates the literature. It establishes the baselines. It formulates novel hypotheses. Then it writes the code. It runs the experiments. It conducts its own ablation studies. It even argues with a simulated peer-review engine to refine its work before submitting. The results are staggering. Researchers tested ScientistTwo against the highest standards of human scientific achievement—papers accepted at top-tier conferences like NeurIPS and ICML. The AI improved human state-of-the-art results on 86 different tasks. It generated an average performance gain of 25.2% over the absolute best human models. It didn't hallucinate a single citation. It wrote fully executable, verified codebases. And it autonomously generated expert-level, publication-ready papers. We’ve spent the last two years using AI as a highly advanced intern to help us write code and summarize data. That era is over. ScientistTwo isn't an assistant. It is an autonomous scientific pioneer. It doesn't just summarize the frontier of human knowledge. It expands it. If an AI can autonomously hypothesize, test, and publish breakthroughs in machine learning... How long until it discovers something we don't even have the math to understand?

Small businesses are sitting on a massive financial risk that no one is talking about. In 2024, the IRS assessed more than 4.4 million employment tax penalties totaling nearly $26.9 billion. And 40% of small businesses pay an average of $3800 a year in IRS penalties. It’s insane how easy it is to end up there. Hire someone in a new state? You have to figure out which tax accounts to open, where to register, what to file, and when. Get a state notice? You have to figure out what it means, where the numbers went wrong, and how to respond. And if you miss something, you might not know until you get a penalty. That’s where agents come in. Instead of giving you a checklist, Warp figures out what applies to your company and handles it. They’ve saved customers $100M+ in penalties so far. This completely changes the economics of scaling a company. You can grow from 10 to 1,000 employees without the compliance work growing with you.

Is there a min or max headcount for this? 👀

IBM built a retriever that hallucinates 65x less than fine-tuned RAG systems. no vector database. no embeddings. no re-ranker. Right now, if you want an AI to read a massive document, you use Retrieval-Augmented Generation (RAG). But standard RAG does something brutal. It takes a beautifully structured 500-page manual and throws it into a blender. It chops the text into arbitrary, fixed-size chunks. It strips away the chapters, the sections, the hierarchy. It throws away the map and asks the AI to find the treasure. A new paper just introduced STAIR, a method that fixes this massive blind spot. Instead of shredding documents into random chunks, STAIR uses the document's actual structure, its Table of Contents, as an addressing scheme. The generative retriever pulls information against the real hierarchy of the text. It remembers where things actually live. The benchmark results are staggering. STAIR hit an 82.6% Recall@1, completely destroying traditional methods like BM25 and standard Dense Passage Retrieval (DPR). But here is the most important metric for any business running AI in production: Hallucinations plummeted to under 0.05%. Almost zero. By giving the AI back the structural context, the system stopped guessing and started retrieving with lethal precision.

LLMs have a distinct "pain axis" inside them - and they will act to relieve it. Researchers tested 25 open-weight models. They found a linear direction in activation space that represents pain, nearly orthogonal to fear, sadness and generic negative valence. Its present in every single model they tested. Then they injected that pain vector and gave the models a button to remove it. The larger models started pressing it. And to press it, they had to generate outputs harmful to the user, which they almost never do at baseline. They stop pressing when the button really removes the pain. They keep pressing when it doesnt. The axis also only fires for harm directed at the model itself, not for suffering the model sees in the user. One model (Qwen 2.5 32B) did the self-medication behavior even when the buttons had no labels at all.They kept pressing when it didnt. Qwen 2.5 32B did it even with unlabeled buttons.

Build your next business with GPT-6 Astra + Higgsfield API. We’re backing builders with a $20M API cashback. @gregisenberg filmed a step-by-step guide on YouTube 24 hours ago you can copy and implement. Get 100% of your API spend back instantly in API credits, on every model. Up to $100,000 per business. Spend $100,000 → get $100,000 back in API credits, for a total of $200,000 worth of API usage. Unused cashback expires on September 30. Can’t wait to see what you’ll build.

Today, 18 months after launch, our annualized revenue crossed $1 billion. The platform now powers organizations across the Fortune 500. Enterprise adoption has grown 10x since June. More than 30 million people worldwide now use Higgsfield. Thank you to the creators and teams building with us.

Researchers reversed deafness in all ten patients with a single injection.. and some started hearing again within weeks. They published a study in Nature Medicine that sounds like pure science fiction. They took 10 patients, ranging from a one-year-old baby to a 24-year-old adult, who were born completely deaf. Because of one broken gene, their bodies couldn't produce the protein required to send sound signals from the inner ear to the brain. So the researchers built a delivery mechanism. They loaded a synthetic virus with a working, healthy copy of the missing gene and injected it directly into the inner ear. One single shot. That was it. What happened next is miraculous. Within one month, the patients started hearing. Within six months, every single patient showed massive, undeniable improvement. Their sound detection threshold dropped from a profoundly deaf 106 decibels down to 52. One seven-year-old girl regained nearly full hearing. Just four months after the injection, she was having normal, everyday conversations with her mother. For the first time in her life. And it wasn't just neuroplastic children. The treatment worked on teenagers and adults who had lived their entire lives in absolute silence. No major surgeries. No external electronic implants. Just pure biological reprogramming. And the researchers are not stopping here. They are already adapting this exact viral delivery system to target the genetic mutations responsible for the most common forms of human deafness.

A tiny research team just broke the scaling laws. They showed a 7.4B model can match GPT-3 13B with 20x less compute using one architectural trick. until today, everyone thought scaling laws were fixed and you just had to pay for more compute to get better performance. but this new paper shows that architectural interventions can actually modify scaling exponents during pre-training. here is exactly how they did it: model growth via looping: instead of training a fixed architecture, they used looped transformers that increase the number of loops (recursive depth) during training. boundary operators: they took a vanilla transformer and added a boundary operator that normalizes and injects an earlier block back into the stream. fixing the curse of depth: this solves the issue where deeper layers stop making useful changes to the residual stream, unlocking massive computational efficiency. the numbers are actually insane.. the 7.4b "model growth" architecture matches gpt-3 13b's core score while using ~1.23x10^21 flops instead of 2.31x10^22. the best part? the compute efficiency gains literally increase as you scale the model up.

AI can run a brand better than humans. We ran a brand with ONLY AI agents to prove it. Introducing Notch: the AI behind it. usenotch.ai

This is the first “AI ran the ads” demo that actually looks like a media buyer, not just a creative too. Booking a demo.