We’re building tools to support research in the life sciences, from early discovery through to commercialization. With Claude for Life Sciences, we’ve added connectors to scientific tools, Skills, and new partnerships to make Claude more useful for scientific work.

New method for automatic prompt optimization! TL;DR: We do reinforcement learning to train an LM that can take in a lot of task examples and generate a prompt that describes the task. Trained/tested on 3000+ classification datasets on Hugging Face!

This is a game changer honestly for mobile use - I wonder how well it works

an absolute shame. @harvard cutting PhD admissions "The Organismic and Evolutionary Biology department will shrink its class size by roughly 75 percent to three new Ph.D. students [...] Molecular and Cellular Biology will reduce its figure to four new students, and Chemistry and

Been bewildering having to be the guy who is way less bearish on this area than others :/ I think frankly more folks ought to read things like GEPA (reflective learning), recursive LMs (truly general-purpose inference scaling), and just DSPy (downstream specification).

New research paper with Anthropic and Thinking Machines AI companies use model specifications to define desirable behaviors during training. Are model specs clearly expressing what we want models to do? And do different frontier models have different personalities? We generated

The metricization of science is the death of science. Any scientist "welcomes this initiative" should be permanently disbarred. nature.com/articles/d4158…

Since I’m glass half full, I’ll note the major advantage of stupid shit like this. Pyramid climbing, game playing, cynicism, and crowning winners works directly against scientific progress when the most difficult scientific challenges that impact society get unlocked by combined

Here is the estimated tokens over time Interestingly it seems we are doubling token usage ~every quarter. Much much faster than Moore’s Law… (also from asymmetricopportunities)

Life is too short to build algorithms without feedback from the real world. Life is even shorter to build infra for the already oversaturated algorithms out there. Oddly, this style bothers some people. Oh well, that’s ok. Impact is not limited to VC startups and big tech.

'use workflow' on Amazon Web Services. Here's a batteries included npm package for deploying Vercel 's workflow development kit on AWS. 1. Install the package npm install aws-workflow 2. Bootstrap the stack on AWS - SQS + Lambdas + S3 npx aws-workflow bootstrap -y 3. Write your

IMO it’s not bad that tech folks are building AI for science because they lack the domain expertise to know what’s working. A greater waste is that too many are over-indexing on every poor approach scientists have due to their human limitations and historical baggage. Scaling a

Anthropic has a very elegant vision for agents and their API foundation reflects this. For eg, back in Feb, I was using cursor + sonnet to build HTML to pptx slides. It was obvious this was a product direction. I know many others saw this too, but only Anthropic shipped Skills.