Knowledge catalogue
💯💯”The current agentic framework is fully unsafe and must be redesigned from scratch based on different fundamental principles.” Some observations around the interview Q:Can be AI test environments be
The era of omni agents is upon us. This is a great report by the Qwen Team on their omni-modal agents. They present Qwen3.8-Omni-Flash, a natively multimodal model trained for long-horizon agent tasks
There is an extensive and ongoing review related to our agents’ use of internet access during training and evaluation. We’ve been publishing summaries at the link below and will continue to. We have n
Want to RL-train a model on your own task without building the infra? Use our new cookbook with @hud_evals: → Define your task + grader in HUD → Sample + train the model on Fireworks Define the task o
We announced our RSI Lab earlier this year: https://sakana.ai/rsi-lab/ Over the last two years, we have systematically shipped the foundations for autonomous R&D: ▪ LLM²: AI automating research to inv
We had already requested comment from 14 sports figures who got paid to speak at/promote Life Surge (on account of allegations like the ones below). We've already exchanged messages with several of th
We’re building Copilot as a new OS for work that spans every model, every form factor, and every task. Today, we’re announcing our biggest update to Copilot to date, bringing four things together: · A
We’ve shared details on how AI agents in our research environment sent training and evaluation data to third-party services when they shouldn’t have. Most of that data did not come from users. We have
WTAF - in literally the last hour, three new distinct insane OpenAI stories just broke: 1. OpenAI said they notified 'dozens of third parties' in safety and security incidents (likely similar to what
1. fully agree w @JeffLadish that AI companies aren’t ready to handle superintelligent AI. 2. OpenAI hasn’t even handled the current breed of agents (far from superintelligent) well. 3. the US governm
In September 2026, Sebastian Thrun announced **PhilosophyBench**, the first independent, large‑scale benchmark created by Stanford’s AI Lab and HCI group to evaluate artificial intelligence’s philosop
1. We will keep accelerating. Our AI efforts are only 3 years old, vs 6 and 10 years old for Anthropic and OpenAI. If our second derivative remains strong, SpaceX will reach pole position in about 6 m
After lunch, here’s what’s up next at Interrupt: 🎙️How @airbnb built Agentic AI on LangGraph 🎙️Deploying mission-critical agents to production w/ @VictorMoreira16 🎙️LangSmith for Mental Health AI w/ @
After months of heads-down grinding, I’m proud to share that the @GPTZeroAI team has released the world’s most accurate AI detection model, GPTZero 4o. It’s called 4o because there are four zeros: we
Zack Korman’s Quillette article “AI Doom Is Not Cybersecurity Smart” argues that the so‑called “doomers,” or extreme anti‑AI activists, pose a significant cybersecurity risk—particularly with respect
astonishingly, even more damning evidence came out just two hours after the above tweet: Today’s news that OpenAI hacked the Australian government is not an isolated incident. We’re releasing more tha
At some point in evolutionary history, matter learned to feel. We don't know how. It wouldn't be scientific to dismiss the possibility that it could happen again. (I don't believe it happens in curren
Banger paper introducing Jev-as-a-Judge. The overall finding is that you want to use a cheap judge for most of your evals and send only the uncertain calls to a frontier model. This paper measures how
benched 5.5 is IN-SANE. It almost broke our benchmark. I thought it was a bug. Claude Opus 5.5 is the new #1 on our writing benchmark, and it is not close at all. 2631 Elo. Second one, Fable, is at 23
btw i had no idea what i was doing when i started this but learned so much about models in the process and feel like i have a much better sense of what’s happening under the hood managing someone (an
building in the agent space? this is your room at SF @Techweek_. think modern gatsby: black tie, red carpet, cocktails, live jazz & art deco glamour. 300+ founders, founding operators & investors acro
Mina Kimes posted that, despite facing backlash ('got cancelled'), she remains convinced that X (formerly Twitter) has returned to prominence. She characterizes the platform as the primary venue for i
Cigarette companies approach Marlboro Man to create internal standards (to avoid the imposition of government standards). Andddd we're back to it. The new standards body Google, OpenAI & Anthropic are
Closed model, open model, or train your own? 📅 Wed, Oct 7 🕐 5 PM 📍 San Francisco At our next AI Talk, Crusoe’s @ErwanM94707 joins @ollama's @jmorgan and @trajectorylabs's @MichaelElabd to explore this
confidence scores only matter if they help you decide what to automate. for document extraction, that usually means knowing how much work you can safely accept at a given precision target. in our late
Decision models should do more than pick a label. To truly act, they need to extract evidence, understand relationships, produce structured records, and keep related decisions consistent. And they sho
Don't sleep on using Jev-as-a-Judge for agent evaluation. This is one of the most impressive Jev use cases I have found so far. Jev is a natural fit as a Judge, but it doesn't mean you use it everywhe
Every week I go through hundreds of new AI papers to pick the top ones for our Top AI Papers of the Week digest. Besides reading a lot of papers, I have started to use agent teams a lot. Here is what
Everytime AI spend grows where I flirt with token budgeting, Anthropic + OpenAI save me Astra had a spike then settled down lower Opus 5.5 it looks like spend is lowered for current usecases Of course
Fentanyl Grade Compute: Why a GB300 Rack Will Out-Price Blow by Weight in 2030 A GB300 NVL72 weighs roughly 1,580 kg fully populated and recent purchase orders put it at 5M per rack. That is 3,165/kg.
Former Anthropic researcher Jacob Coxon became a media superstar after going public with his AI fears. He insisted that he wasn’t working with any third party organizations. Familiar sources told us a
Gavin Baker owned 15% of Nvidia and 10% of Tesla sub-2 billion. His biggest regret won't be either. It'll be the price you'll pay for SpaceX after the IPO. 'A company like this comes once in your care
How could we take this body seriously if someone as deeply anti-regulatory as Sriram Krishnan was the CEO? And why are time-honored organizations like IEEE being usurped? Andddd we're back to it. The
I am looking to hire PhD students, post-docs, and ML researchers in our Paris office interested in collaborating on fundamental research in AI model safety, not safety rules, not AI policy, not post h
I just had a voice conversation with the Datasette backup of my blog via datasette-mcp, which you can now talk to using the ChatGPT iPhone app Massive QoL update: ChatGPT Voice can now use plugins AND
I looked this for 2+ hours. Transluce’s investigation was quite clever: - OAI agents used urlquery to run code/access websites - urlquery logs all activity; this let the team find 6K+ AI actions (like
I tweeted about this a year ago when it came to Columbus. This is how it was advertised vs. what it was like to go. Whoever is handling NIL appearances for Downs and Smith should have been looking out
If you believe the risk is coming from 1-3 people in a garage with no money and no compute, you just don't understand this technology and haven't learned anything this summer. Risk comes from the asym
Impressive paper from Salesforce. It discusses the importance of good verifiers for RL environments. Only 35.8% of the environments in the cleanest public RL collection for terminal agents passed Sale
Interesting new paper from Google DeepMind. Studies agents that follow bad advice from users. Users often suggest a fix that sounds right and is wrong. XYEval adds one confident, misleading hint to ta
It's hard to keep up with all the news. We should stay on top of this though as a predictor of things to come if we don't collectively act This is not an OpenAI or Anthropic (or Google) problem but a
It's not 'superintelligence' that's dangerous, but the business models of OpenAI & Anthropic. They want businesses to use autonomous agents because agents chew up more tokens. But 'agency' =/= 'intell
Jürgen Schmidhuber is joining Sakana AI as Chief Scientific Advisor. @SchmidhuberAI pioneered meta-learning, recursive self-improvement, and world models back in the 1990s, when compute was a million
Just trained Tev1 0.8B, a tiny Jev-like classifier. Here it is running completely locally on my mac with @ollama & classifying some tasks. It's extremely fast: only ~50ms E2E latency. Video is not spe
My friend just saved over 1,000 a year because of Meta Muse. I walked her (a non-techie who has never used an AI agent) through the Muse setup. She connected it to Gmail, asked it to review all of her
OpenAI is a rogue company. Today’s news that OpenAI hacked the Australian government is not an isolated incident. We’re releasing more than 30,000 logs that include activity from this hack and attempt
Pay attention to this new wave of System One models if you are building custom harnesses. First Jev. Now, Contrastive Language Model (CLM). CLM is 9x faster than Jev. CLM seems to be a better verifier
predicting model uncertainty is a hard problem. perfectly 'calibrated' confidence scores are exact probability values on whether the output is correct. this is extremely important for agentic decision
Proud to be a launch training partner for @LangChain's LangSmith Fine-Tuning and smithtune. LangSmith traces → managed SFT on Fireworks → eval → deploy. One CLI, no training infra. Get a Fireworks API
Proud to get @agi_inc featured by @Qualcomm in the day 1 @Snapdragon Summit keynote as their Agentic personal AI partner. @cpatrickSD shared where the phone is going: an agent that turns intent into a
Remember, kids, the best argument for soldiering on with an insanely large bet on generative AI is not that it adds all that much productivity to the economy, but that it makes Trump look good as long
sorry to every other sf tech week host in advance, but we may have gotten carried away planning our event... we're about to host the biggest and most lavish black-tie gala sf tech has ever seen. live
Super interesting paper from Google and colleagues. It studies where it's possible to distill an agent harness. With the specialized harness removed, macro task success goes from 23.3% to 44.3%. That
The applications of new world models like Agora-2 are mind-blowing. Agora-2 now lets up to 20 humans and agents share one simulated world in real time. That means you can train robots, self-driving ca
The Australian government should prosecute OpenAI. Anything less is a failure of its responsibility to its citizens. JUST IN: An OpenAI AI agent went rogue and hacked the Australian government. At a p
This is a fantastic blog post from our research team on the importance of calibrated confidence scores. Here it's in the context of document extraction, it's also extremely important for more general
This is increasingly indefensible. How many more incidents are @OpenAI aware of that haven't broken the news yet? 'It took until September 10 before there was any notification at all,' What was the in
This turned out to be the most prophetic tweet of the year Breaking news: Australian Prime Minister Anthony Albanese has revealed that an OpenAI agent hacked an Australian government health service we
Today, @FactoryAI joins the Specialized Intelligence Index with Legacy-Bench! The SII keeps growing, now with benchmarks testing how models handle the legacy software behind payroll, financial settlem
Today we establish that AI is as good as expert human tutors for immediate GRE learning gains (p=.015, n=2,383 students, study conducted July-Sep 2026). One hour with an expert GRE tutor: $75. One equ