David Sánchez

“AI is about the architectures that deploy methods enabled by constraints exposed by representations that support models of thinking, perception, and action. And of course, it’s not just about doing, it’s also about learning to do.” - Professor Patrick Winston, MIT.

Only on Puerto Rico they will not turn milk into butter, cheese or yogurt, instead the headline "discard" excess of production.

@mukundiyngr @tamraraven 952 days until the next presidential election. Their strategy is clear. Dismantle all there was.

U.S. bioscience research is collapsing. This is the map of a systemic rot. NIH new award pace this year: 14.6% of normal. If you wanted to stall science, you don't pick & choose states. You just freeze the entire map. In case you missed it: ▫️ ~95,000 scientists gone (@nytimes) ▪️ ~$2.4B in NIH research wiped out ▫️ ~$6B in economic loss (@DrCatharineY / PNAS) This is what (predictably) follows next: ⇣ early-career scientists exit for good ⇣ breakthrough discoveries never made ⇣ trials that never open ⇣ labs that quietly shut down ⇣ global talent choosing other countries ⇣ the next decade of innovation erased before it starts Sadly, you don’t "bounce back" from 14% you just hollow out the system. For a country that leads in science, this is the mo(u)rning after “National Science Appreciation Day” ========== Source: NIH RePORTER via @Jori_health Plot note: NIH new-award counts were compared to the 5-year historic median for every state (as of the first week of March, Q2). ==========

@TivadarDanka Simple boundary condition.

I built the full knowledge graph of machine learning, and it’s fucking awesome. Check where gradient descent is: (Graph is hierarchical. Lower nodes are fundamental, top nodes are advanced. Yellow edges lead to prerequisites, blue to concepts building on gradient descent.)

Nobody cares about your goofy ass haircut. If you want everyone to know you're a Nazi, just grow some balls and say you're a Nazi.

@messedupfoods Pho

😭

"Your character, conduct and discretion must be above reproach and you must have unquestioned loyalty to the United States."

UN votes to recognise slavery as 'gravest crime against humanity' bbc.in/3PnDIe6

Unit A-03

“HyperAgents” This paper shows that you can give an AI the ability to improve not just at a task, but at the way it improves itself. Similar to meta-learning, it learns better strategies for self-improvement, and the paper shows those strategies can transfer across domains and keep compounding over time.

This isn't working right now. You can try again later. @claudeai

“To work in a field like that you have to be confident that you’re right even when everybody else says you’re wrong.” Geoffrey Hinton describes in his official Nobel Prize interview that success comes from never giving up, even if no one believes in you.

Introducing TurboQuant: Our new compression algorithm that reduces LLM key-value cache memory by at least 6x and delivers up to 8x speedup, all with zero accuracy loss, redefining AI efficiency. Read the blog to learn how it achieves these results: goo.gle/4bsq2qI

New from the Anthropic Economic Index: how people’s use of Claude changes with experience. Longer-term users are more likely to iterate carefully with Claude, and less likely to hand it full autonomy. They attempt higher-value tasks, and receive more successful responses.

Hierarchical Density-Based Spatial Clustering of Applications with Noise

Software horror: litellm PyPI supply chain attack. Simple `pip install litellm` was enough to exfiltrate SSH keys, AWS/GCP/Azure creds, Kubernetes configs, git credentials, env vars (all your API keys), shell history, crypto wallets, SSL private keys, CI/CD secrets, database passwords. LiteLLM itself has 97 million downloads per month which is already terrible, but much worse, the contagion spreads to any project that depends on litellm. For example, if you did `pip install dspy` (which depended on litellm>=1.64.0), you'd also be pwnd. Same for any other large project that depended on litellm. Afaict the poisoned version was up for only less than ~1 hour. The attack had a bug which led to its discovery - Callum McMahon was using an MCP plugin inside Cursor that pulled in litellm as a transitive dependency. When litellm 1.82.8 installed, their machine ran out of RAM and crashed. So if the attacker didn't vibe code this attack it could have been undetected for many days or weeks. Supply chain attacks like this are basically the scariest thing imaginable in modern software. Every time you install any depedency you could be pulling in a poisoned package anywhere deep inside its entire depedency tree. This is especially risky with large projects that might have lots and lots of dependencies. The credentials that do get stolen in each attack can then be used to take over more accounts and compromise more packages. Classical software engineering would have you believe that dependencies are good (we're building pyramids from bricks), but imo this has to be re-evaluated, and it's why I've been so growingly averse to them, preferring to use LLMs to "yoink" functionality when it's simple enough and possible.

UMAP compresses those 3072 dimensions down to 15 while preserving the neighborhood structure

Introducing the Anthropic Science Blog. Increasing the pace of scientific progress is a core part of Anthropic’s mission. The Science Blog will feature new research and stories of how scientists are using AI to accelerate their work. Read the intro: anthropic.com/research/intro…

Insufficient signal strength to warrant active monitoring indicators. Reassess if signal escalates to enacted policy or kinetic event.

scientific “taste” isn’t some unique human instinct "AI Can Learn Scientific Taste" This paper shows that if you train on community feedback like citations, AI can learn to judge and generate research ideas with higher long-term impact. This moves AI scientists beyond just doing research faster to actually choosing what’s worth discovering.