Triatomic Capital @TriatomicCap

x.com/i/article/2112…

Why hardware specialization is back. AI inference needs more tokens/sec per user and more tokens per watt. Power limits. Slowing transistor scaling. The memory wall. Different chips are optimized for different constraints. Our co-founder @zainasgar on what comes next.

Zain Asgar @zainasgar

20 hours ago

x.com/i/article/2112…

2 16 42 3K 14

When we backed @RadicalNumerics, the bet was that one general model could learn the language of biology, the same way LLMs did for text. Today Omnii-1 beats AlphaGenome, designs antibodies and cancer vaccine candidates, and designed a phage covered by @carlzimmer in @NYTimes. All from one model. 🧬

Eric Nguyen @exnx

21 hours ago

Introducing Omnii-1, our most powerful genome language model, and a major step toward general biological intelligence In a world first, Omnii-1 designed the first AI-generated bacteriophages to be administered to a human patient with an antibiotic-resistant infection under

103 174 1K 565K 469

Omnii-1 is a huge milestone for @RadicalNumerics. Building it required advancing architectures, pre-training, post-training and inference to make new biological applications possible. Pre-training meant tokenizing and blending continuous and discrete biological modalities. Post-training adapted that foundation to different applications. At inference, we’re exploring algorithms for reasoning across biological modalities and allocating test-time compute across selective generation and conditioning steps. Making these models train and run at scale meant pushing the underlying numerics and systems alongside the modeling. The frontier of biological AI runs all the way down the stack. We have a cracked team. Really proud of what we’ve built together.

Eric Nguyen @exnx

21 hours ago

Introducing Omnii-1, our most powerful genome language model, and a major step toward general biological intelligence In a world first, Omnii-1 designed the first AI-generated bacteriophages to be administered to a human patient with an antibiotic-resistant infection under

103 174 1K 565K 469

Every math teacher has written "show your work" in red pen. @CarinaLHong turned it into an output format. The @axiommathai team built an AI for really hard math that writes its proofs as code a machine can check. Watch Carina's story 👇

An Interview with d-Matrix CEO Sid Sheth on Why 3D-Stacked Memory Compute Is a Game Changer taekim.substack.com/p/an-interview…

Big day for @RadicalNumerics: Omnii-1 is out – one multimodal model across DNA, RNA, proteins and more. And in a world first, Omnii-1-designed phages were given to a patient with an antibiotic-resistant E. coli infection. Excited to back @exnx & team! radicalnumerics.ai radicalnumerics.ai/blog/omnii-pha…

Omnii-1: a leap toward General Biological Intelligence...

Eric Nguyen @exnx

21 hours ago

Introducing Omnii-1, our most powerful genome language model, and a major step toward general biological intelligence In a world first, Omnii-1 designed the first AI-generated bacteriophages to be administered to a human patient with an antibiotic-resistant infection under

103 174 1K 565K 469

I'm excited to share that @tri_dao has joined @RadicalNumerics as a scientific advisor! Tri and I were labmates at Stanford's Hazy Research, where he pioneered advances in efficient ML systems (FlashAttention). Scaling AI across multimodal biological data demands entirely new approaches to ML systems, and Tri will help guide our research at this frontier. Welcome aboard, Tri! Read more: radicalnumerics.ai/blog/tri-dao

Choosing whether to allocate scarce watts between low-latency and high-throughput chips is a scary decision It's hard to predict over the next 3 years how much demand will be low-latency vs high-throughput Luckily with Volantis, you can have both modes in one chip :)

Jessie Dong @jessiedong_

5 days ago

inference providers have to choose to be either fast or cheap if you choose speed as a provider, you'll have idle gpus. but if you optimize to be cheap, customers are going to be waiting the pay-as-you-go pricing model means you pay per token, and the provider pays for gpu

14 11 107 9K 68

Breaking through the memory wall. Rethinking AI inference. d-Matrix CEO @sidsheth joins Tae Kim (@firstadopter) to discuss memory-centric compute, 3D-stacked DRAM, and the future of low-latency AI inference. Read the interview: taekim.substack.com/p/an-interview… #AIInference #AIInfrastructure

I'm thrilled to welcome Jill Hruby @jill_hruby and Paul Friedrichs as biodefense advisors to @RadicalNumerics! Jill formerly led the National Nuclear Security Administration, and Paul served as the inaugural White House Director of Pandemic Preparedness and Response. As AI opens new frontiers in biology, strengthening our ability to defend against biological threats is more important than ever. We're incredibly fortunate to learn from their experience and leadership as we advance this mission. Welcome aboard, Jill and Paul! Read the announcement: radicalnumerics.ai/blog/biodefens…

I don't always want to compute efficiently, but when I do ... I use Efficient Computer. Keep your computers fed and your robots moving, my friend. More: @efficient_hq #100xbetter

Efficient Computer @efficient_hq

2 days ago

The Electron E1 general-purpose processor is one of TIME’s Best Inventions of 2026: up to 100x greater energy efficiency than conventional low-power CPUs. What will you design next? Read the TIME feature time.com/collection/bes… #TIMEBestInventions

0 1 8 1K 2

A big week for d-Matrix at #OCP. Corsair rackscale is now generally available — air-cooled, industry-standard, ready for existing data centers. Grateful to @Supermicro for building alongside us. We introduced Squadron, our multi-rack clusters scaling to 1,000+ Corsair XPUs. And we ran the industry’s first public demo of Attention-FFN Disaggregation: one NVIDIA H100 + 8 Corsair XPUs hitting 845 tokens/sec/user. All credit to the d-Matrix team and our partners. The future is memory-centric inference compute — here today. 🔗 d-matrix.ai/announcements/…

keyboard_arrow_left Previous keyboard_arrow_right Next

1/8 Does text-to-image generation need language pretraining? Today, we share Halcyon, a family of early-fusion, base models trained from scratch. Halcyon outperforms strong late-fusion baselines with up to 275× less pretraining compute. Models and recipes are open. Halcyon is part of a research program at Radical Numerics on the science of multimodal pretraining and scaling.

Multimodal control and generative design is in our DNA, across all domains, from biology to images Very excited to share this work in early-fusion training We're building a new kind of AI lab. Check out the work by @keshikiayame @Armin @MichaelPoli6

Radical Numerics @RadicalNumerics

6 days ago

1/8 Does text-to-image generation need language pretraining? Today, we share Halcyon, a family of early-fusion, base models trained from scratch. Halcyon outperforms strong late-fusion baselines with up to 275× less pretraining compute. Models and recipes are open. Halcyon is

3 29 170 63K 117

New research on pretraining data mixing in biology. At Radical we're huge supporters of driving frontier research in AI and sharing it with the community. So much cool work at the intersection of biology, this is the next frontier of AI 🧬🔥📈

Radical Numerics @RadicalNumerics

a week ago

1/5 What does it take to scale pretraining to 20+ data modalities at once? Improved data mixing methods are one important ingredient. Today, we share how we scale data mixing to thousands of sources for DNA and natural language pretraining of Omnii, our multimodal genome model.

5 54 291 367K 200

Building general biological intelligence requires advanced data mixing methods at every stage of training, across modalities and scales. Read more: radicalnumerics.ai/blog/tessera-d… We're hiring: radicalnumerics.ai/join-us

1/5 What does it take to scale pretraining to 20+ data modalities at once? Improved data mixing methods are one important ingredient. Today, we share how we scale data mixing to thousands of sources for DNA and natural language pretraining of Omnii, our multimodal genome model.

845 tokens/sec per user. ~50% lower latency. d-Matrix Corsair + NVIDIA H100 are redefining AI inference with attention-FFN disaggregation. Faster inference. Smarter compute. Read more: d-matrix.ai/accelerating-a…