The first experimental evidence of recursive self-improvement (RSI). Autoresearching the autoresearch agent for eight days. The result beats the harness we hand-tuned for two years, on held-out benchmarks: 🧵(1/7)

a brand new model paradigm! introducing 𝚌𝚕𝚊𝚒𝚛𝚒𝚌: a system one model for extracting, computing, estimating, and predicting values from unstructured inputs. blog.cloudflare.com/clairic-…

Last night, I had the honor of leading my first Shabbat. I co-hosted the first Shabbat of the year with @menemazarakis, @gaby_goldberg, @zackrosen_, and @ArgamaniNoa. We brought together an incredible group of young founders and investors from across Silicon Valley for a night centered around Judaism, meaning, and unity. For some, it was their first kosher Shabbat. We even put up a mezuzah before Shabbat to sanctify the home. We prayed together, shared a meal, and I had the honor of sharing a few words about Yom Kippur. One of my favorite moments: a CEO of a $3B+ company who is Shomer Shabbat walked over an hour each way with his nephew just to join us. Truly one of the most beautiful Shabbats I’ve ever experienced.

Yesterday, I told you Mexico’s total fertility rate is at 1.2. Today, let me remind you that Argentina’s total fertility rate is at 1.05: screenshot from the Argentine statistical agency below. Yes, 1.05. Argentina, in 2014, was at 2.3: a drop of more than 50% in a decade. And before you tell me “it’s Milei’s/the Peronists’/Leo Messi’s fault,” remember the following: of the 236 countries and territories on the planet, 219 have a clear negative trend in fertility. Of the remaining 17, I’d say the trend is negative as well in probably 16, although statistical methods have a hard time picking it up for one reason or another. The remaining one? Israel. So, if you think fertility is going down in your country because of some country-specific reason X, well, you might want to think twice.

Introducing Prime Agent: A self-improving RLM harness for coding and long-running autonomous tasks. Designed to be both token-efficient and expressive through programmatic tool calling, context as a variable, multi-agent messaging, and a self-modifiable harness state.

Do you have a personal assistant style agent set up? If so, which one? If not, why not? Think closer to Hermes / OpenClaw, not really Claude Code / coding agents

Let's talk a bit about what makes the Superlogical multiplexer different architecturally from traditional terminal multiplexers! There's a lot (a LOT) more coming, but wanted to share a little bit about the terminal-specific part compared to tmux and zellij.

A single dollar now buys roughly: - 20 Luna-max benchmark tasks - 3 Grok-high tasks - 2 Terra-max tasks - 1 Sol-max task - Half an Opus-max task

oh wow we went a little deeper on SlopCodeBench and the frontier is better but not by much...stay tuned for the full post