🎠

Haiku 5.5 is a significant step up over Haiku 4.5 across coding, computer use, and knowledge work.

We're publishing a result demonstrating a substantial tightening to the results from OpenAI Problem #109 (integer multiplication). Conditional on OpenAI's algorithmic interfaces, our parameter and network refinements improve the exponent saving in: T(n)=O(n(\log n)^{1-\kappa}) from κ=2⁻¹⁸² to κ=5.8×10⁻³³ (between 2⁻¹⁰⁸ and 2⁻¹⁰⁷). This represents a nearly 2⁷⁵ fold increase in the algorithm's exponent saving parameter. We also establish a ceiling of κ<5.838×10⁻³³ for the stated network-counting family and cost inequalities. Our result exceeds 99% of that ceiling. Surpassing this ceiling would require improving the network bounds or cost analysis from the original result.

Claude now works inside Google Docs, Sheets, and Slides, and those files also open inside Claude. In Google Workspace, Claude sits in a sidebar next to your file, reads what you have open, and edits it in place. You can approve each edit before it lands.

Google Deepmind just broke chemistry.. They open-sourced a model that can invent completely new biology from scratch. It’s called “AlphaProtein Novo” It's an AI pipeline that designs brand new enzymes for "new-to-nature" chemical reactions that have never existed in any living organism. to put this in perspective.. standard protein ai models (like alphafold) answer the question "what shape will this protein have?". ap novo answers the question "can i make a protein that does this specific chemistry?". the results from the paper are actually terrifyingly good: the ai designed an enzyme to synthesize piperidines, which are a critical building block used in pharmaceuticals and materials. this designed enzyme inverted natural biases to achieve near-perfect regioselectivity. it produced the chemical building block 99x more often than the competing natural product. it also designed an enzyme to hydrolyze DEHP, which is a pervasive plastic environmental toxin. this plastic-eating ai enzyme was 14-fold more active at 90°C than at room temperature, and survived in 75% acetonitrile. meanwhile, natural enzymes completely failed and were entirely inactive under those exact same extreme conditions. these new ai-generated designs are up to 60% smaller than natural equivalents. because they are smaller, they have a much higher mass density of active sites. i am speechless.. this is the holy grail of synthetic biology. we don't have to wait millions of years for microbes to evolve enzymes that eat our pollution or build our drugs.. we can just compute them on a gpu today. the ai's success rate in finding functional designs actually matches or surpasses traditional natural sequence screening. and deepmind open-sourced the whole thing on github.