
Superintelligence is about to emerge.

limₙ→∞ I(θₙ; D) = ?

Superintelligence is about to emerge.

reread attention is all you need last night. attention by itself has no clue about word order?? like “dog bites man” and “man bites dog” are literally the same thing to it. so they just add sin and cos waves to every token so it knows where its at. word order is basically trig lol. did not see that comin


we thought the brain of robots would not be as good as the physical robot itself. guess what, we have been proven wrong with the emergence of llm and frontier models. at present, we have our work cut out for us where we have lagged behind. Physical.Bot
Physical AI represents some of the most complex and consequential engineering problems to date. @kbrandes45 breaks down on how we approach this with a rare depth and transparency. link to full video in comments below.

physical AI/SI is moving at a phenomenal pace. Seeing what is happening, the future to belongs to ro/bots. physical.bot
F.03 humanoid robots


Selling a few .si : Open to Offers induce.si - Induce Labs streak.si - Habit Intelligence onset.si - Early Detection/Health primacy.si - Frontier Labs feud.si - Red Teaming pronoun.si - Language Models posh.si - Elite models chirp.si - Voice Models minor.si - Small/Private Models sire.si - Foundation Models tern.si - Aerial Intelligence impart.si - AI/SI Research

Peter Thiel on why focusing too much on risk sets you up for failure: "If you try to reduce this to a pattern of recognizing entrepreneurs as a venture capitalist, you always want to invest in the ones where they speak in definite future tense. You have to sometimes be careful they're not totally crazy people, but that's the sort of person you want to invest in. And you do not want to invest in people who are talking too much about probabilities or risks. Because my experience has been that the people who think they're involved in some sort of lottery ticket dynamic are already setting themselves up to somehow get the probabilities wrong and invariably lose."


sold a few .si for an undisclosed amount $$$$$ to @hodlbankz at @Dynadot


We are releasing the technical report for Limite 1B -Violetto. Violetto is our first model, designed to push mathematical capabilities per parameter to its limits, trained from scratch.



Introducing 𝑨𝒕𝒕𝒆𝒏𝒕𝒊𝒐𝒏 𝑹𝒆𝒔𝒊𝒅𝒖𝒂𝒍𝒔: Rethinking depth-wise aggregation. Residual connections have long relied on fixed, uniform accumulation. Inspired by the duality of time and depth, we introduce Attention Residuals, replacing standard depth-wise recurrence with learned, input-dependent attention over preceding layers. 🔹 Enables networks to selectively retrieve past representations, naturally mitigating dilution and hidden-state growth. 🔹 Introduces Block AttnRes, partitioning layers into compressed blocks to make cross-layer attention practical at scale. 🔹 Serves as an efficient drop-in replacement, demonstrating a 1.25x compute advantage with negligible (<2%) inference latency overhead. 🔹 Validated on the Kimi Linear architecture (48B total, 3B activated parameters), delivering consistent downstream performance gains. 🔗Full report: github.com/MoonshotAI/Attent…

Mercury 2 is live 🚀🚀 The world’s first reasoning diffusion LLM, delivering 5x faster performance than leading speed-optimized LLMs. Watching the team turn years of research into a real product never gets old, and I’m incredibly proud of what we’ve built. We’re just getting started on what diffusion can do for language.