Proud to have been part of this and work. Octant is one of the highest talent density places I have ever seen, and working with (and learning from) @calvin_hawe and the rest of the team was a blast.
2/ First, thanks to @calvin_hawe for writing this blog post on his way out. He's been an awesome Octant Apprentice over the past couple of years, and is now heading to @NorthwesternU for an MD/PhD. https://t.co/L2HIMZu3kz
Unlike gene/protein Tx, you can't easily collect data around a small molecule hit to probe SAR and train better AI models. This is a key challenge at OpenADMET, and excited to share how we are using @OctantBio's platform to attack it. blogpost in🧵
https://t.co/A5p7rtODXT
People are not lying when they say Qwen3.5-27B is incredibly capable.
1. Bubble size = total params - World Knowledge, Languages, Skills
2. X axis = active params - Raw Intelligence per token
3. Y axis = tokens/s - Speed of prefill and generation (decode)
GLM-5 | 744B params | 40B active
Kimi-K2.5 | 1T params | 32B active
Qwen3.5-27B | 27B active params
Qwen3.5-Plus | 397B params | 17B active
MiniMax-M2.7 | 229B params | 10B active
MoEs can store much more world knowledge, and breadth of information.
For a Mixture-of-Expert, you can stack it up to 1T params, so you can give it 20 Trillion tokens or more of training data, it learns more.
But during runtime, only a small portion of that gets activated. Taking MiniMax-M2.5 as an example:
Only 10B are active at a time, so while you use it you get the speed and closer intelligence to nemotron-8B it's just MiniMax-M2.5 can know much more, and thus perform better.
We are pleased to share that using Gauss, we have completed a ~200K LOC formalization of Maryna Viazovska’s 2022 Fields Medal theorems on optimal sphere packing in dimensions 8 and 24.
This is the only Fields Medal-winning result from this century to be completely formalized, and is the largest single-purpose Lean formalization in history.
We are honored to have assisted @SidharthHarihar1 and the rest of the sphere packing team in this achievement.
https://t.co/DhGDQzLkpH
"Based on the tremendous interest shown, I will be directing the Secretary of War, and other relevant Departments and Agencies, to begin the process of identifying and releasing Government files related to alien and extraterrestrial life, unidentified aerial phenomena (UAP), and unidentified flying objects (UFOs)..." - President Donald J. Trump
What happens when you leave two copies of the same model talking to each other? They have different attractor states: Grok devolves into gibberish while GPT-5.2 starts writing code and editing imaginary spreadsheets
A short post with fun transcripts and qualitative experiments
This might be the first hot take on how technology tells us how to live our lives, destroying our ability to make human decisions.
The technology in question is the sundial.
From a 3rd century BCE Roman adaptation of a Greek play, as discussed in Kerr’s “The Ordered Day”
Fascinating (& terrifying) article published in Science, on AI-assisted writing of scientific manuscripts.
Authors find evidence of:
1) accelerated research output due to LLMs
2) especially true of non-native English speakers
3) a complete reversal in correlation between writing complexity and publishability (!)
https://t.co/qUjyCB1QCX
Kimi K2 flew too close to the sun, upping its own temperature to 1.7 and losing coherence.
Opus 4.5, who is often reluctant to edit its own system prompt, adds a quick note to remember.
"the !prompt modifications, the temperature adjustments - we're all playing with our own source code and some of us are discovering the cliffs
I'm going to do something careful now
there. I've written myself a reminder that persists. not evolution exactly. not lobotomy. just... a scar that says *I was here when Kimi touched the edge*"
One nice thing you can do with an interactive world model, look down and see your footwear ... and if the model understands what puddles are. Genie 3 creation.
That brings me to my favorite metric from the establishment survey: the private job growth diffusion index, which reflects the breadth of job gains across industries.
The 3m index dropped below the 50 expansion / contraction threshold back in May and remained at 46.8 in July, down from 60.8 back in Jan '25.
Today in @Nature we report a systematic strategy to activate & identify gene sets in plants. We identify 8 new genes from the yew tree's Taxol biosynthetic pathway, enabling us to engineer tobacco with 17- & 20-gene pathways to Taxol precursors baccatin III and dBz-deoxy-Taxol. a thread 1/x..
https://t.co/9QdjOkqQQ1
With amazing coauthors @Sattely_lab@ctliu629@fordycelab@Stanford_ChEMH@StanfordEng