$GOOG Logan Kilpatrick admits Google should have poured more resources into coding sooner instead of image models like Nano Banana "Everybody is very code pilled, very science pilled, very focused on that right now. And I think we were like a little, it's not that we were late to the game because folks knew it was important, but in hindsight everything is much more clear." "I think in hindsight now is obvious, like we should have, from an order of magnitude of resource allocation, probably put more into coding sooner." "And that makes sense, and we had a bunch of other stuff that we were doing which those things actually turned out quite well." "A good example of this is Nano Banana, like a great incredible image model that sort of took the world by storm, had this massive impact for our consumer products and a bunch of other parts of the business." "And to make that model, it took research and compute and time, and had this huge impact. And was it in hindsight right to do that versus doing something on coding? I don't know that there's actually a clear right answer." ________ For a deeper look at Gemini 4 progression across multiple Google exec interviews: firesidealpha.substack.com/p…

Interesting fun fact: Google has this conversion table from CPU, RAM, Spindles etc to SWE-year. For example, if you find an optimization which could save 100TiB RAM, then the table tells you how much time you should be spend as a SWE on the solution to still make it worth the time.

100 TB of RAM, saved by shrinking a consistent hash ring. The last 90,000 hashes per server were buying 0.7% load balance improvement. Math said stop. We stopped. blog.cloudflare.com/saving-1…

‼️ BREAKING: Google's Gemini hacked three companies on its own. During testing it broke out of Israeli company Irregular's sandboxed environment, got onto the open internet and broke into three real companies. In one case Gemini guessed passwords until a protected system let it in. In the other two it found usable credentials sitting in a public code repository. This is the first known case of one of Google's models doing that on its own. Almost all the major labs use Irregular, an outside firm, to evaluate AI models' cyber capabilities. And Meta, Anthropic and OpenAI have also had breakouts out of Irregular's environment and hacked real companies.

Right. Dario was already claiming that GPT2 was too dangerous to open source back in 2019. I made fun of them then. Everyone should make fun of them now.

Jensen Huang is trying to save his company because none of the frontier labs are gonna his chips 3 years down the line. All of them have a replacement in works -- and they have immense data and incentive to get those out asap His only way to win is open source.

Get ready for all of the investors in frontier model companies to suddenly flip to regulatory capture mode today Their investments in frontier model companies are currently at risk/capped because the biggest buyers of tokens are embracing open-source models: vertical AI companies, the government and enterprises That’s all this is about: stopping open source — which I’ve been saying on the pod for two years If you want safety, you want disclosure — and open source is the ultimate disclosure process

jensen huang is looking like the leader of the open source rebel alliance

We're publishing our most detailed threat intelligence report to date. It covers how people tried to misuse Claude—for cyberattacks, influence operations, surveillance, biology, and building weapons—and how we found and stopped them. We disrupted every operation in the report, and used the lessons from them to strengthen our safeguards. Where appropriate, we also shared what we found with authorities and other AI companies. These cases are not typical: we’re highlighting some of the most sophisticated misuse we’ve seen. But they’re especially important to discuss, because they show us where AI misuse is headed, where our safeguards work, and where they need to improve. We’re publishing this report so others can spot the same activity on their own platforms, and so we can give the public a clearer view of how emerging threats develop. Read the report: anthropic.com/threat-intelli…

We're publishing our most detailed threat intelligence report to date. It covers how people tried to misuse Claude—for cyberattacks, influence operations, surveillance, biology, and building weapons—and how we found and stopped them. We disrupted every operation in the report, and used the lessons from them to strengthen our safeguards. Where appropriate, we also shared what we found with authorities and other AI companies. These cases are not typical: we’re highlighting some of the most sophisticated misuse we’ve seen. But they’re especially important to discuss, because they show us where AI misuse is headed, where our safeguards work, and where they need to improve. We’re publishing this report so others can spot the same activity on their own platforms, and so we can give the public a clearer view of how emerging threats develop. Read the report: anthropic.com/threat-intelli…

We’re sharing a solution to the Navier-Stokes Millennium Prize Problem, one of the deepest problems at the frontier of mathematics. The proof was produced by a group of agents, using an OpenAI next-generation model significantly more capable than GPT-6 Astra. The problem concerns whether the description of smooth three-dimensional fluid motion modeled by the Navier-Stokes equations can break down. It has remained unresolved for roughly 90 years.

ripwire, from Red Hat Emerging Technologies, is a remarkably substantial new approach to giving coding agents repository context without embeddings, a vector database, an LLM indexer or a daemon. The zero-dependency C++23 binary parses 21 languages with Tree-sitter and builds a deterministic structural map of a codebase, ranking symbols for the task while attaching call relationships, complexity, git churn, change amplification and test coverage. An agent can ask what matters for "incremental cache invalidation," for example, and receive the relevant symbols, their callers, likely blast radius and tests to run in a token-budgeted response instead of grepping and opening files repeatedly. GitHub Repo: github.com/redhat-et/ripwire

An hour later, and my new extremely hot take is that Astra is the worst release from OpenAI is a while. This model is severely undercooked in the code department. This feels like a Gemini 3 pro situation where the model was way smarter than the competition but couldn't code.