Jeff Huber 🇺🇸
@jhuber
Co-founder @TriatomicCap ㅣFounding CEO @GRAILbioㅣearlier: @Google Ads, Apps, Maps & ⟦x⟧ㅣf*ckcancerㅣPer aspera ad astra.
28.2K Following 46.9K Followers
We are pleased to announce that Edison Scientific is partnering with Pfizer to bring superintelligence to their R&D organization. Read more below.
As an investor in @EdisonSci, I’ve never been happier to see human disease being very worried.
Google AI overview thoughts on the Edison Pfizer deal
Incredibly excited to highlight Omnii Genome, one of the eight applications of Omnii-1
Omnii Genome fuses 28 multiomic modalities natively across biology
Against comparable benchmarks vs. AlphaGenome, it outperforms on all
Omnii Genome is also *generative*, which means it can also be prompted any modality or cell state, and infer missing modalities
e.g., prompt on desired gene expression, methylation pattern, and then generate the required DNA, or transcription factor pattern
We believe this is going to accelerate so many new possibilities in life science, from multiomic target discovery, diagnostics, and gene therapy
Show more
Today, we are launching Omnii-1, our next-generation multimodal genome language model.
Omnii-1 draws on a long line of research breakthroughs in genome language modeling, spanning model architectures, multimodal pretraining, as well as new mid-training & post-training methods tailored to biological data.
The launch includes a staggering amount of technical material: 8 full-length technical blogs (and we are just getting started).
Proud of the team for this milestone!
Show more
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. 🧬
Show more
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.
Show more
@jhuber @RadicalNumerics @exnx The multimodal architecture across DNA, RNA, proteins and beyond is genuinely novel, and we actually went deeper on the implications here:
We are building AI to save lives and advance our understanding of biology.
I spent countless hours in the trenches of pretraining our foundation model, Omnii-1, hoping it would start learning the highly complex language of life.
One model that learns across modalities, while the common approach is to build specialized models for each modality and task.
When we put this pretrained model in the hands of our amazing team of comp-bio experts, I was so excited to see how well they could take this raw model and mid- and post-train it for so many diverse applications!
One pretrained model. Adapted to solve multiple different challenging tasks better than the specialized state-of-the-art models.
I am so proud of this absolutely world-class team!
And this is just the beginning!! 🚀 🧬
Show more
Omnii-1 is out.
We've learned a lot about what happens when you train a single model across biology rather than solving each problem separately.
Some of the more surprising results haven't made it into a blog post yet.
Incredibly proud of the team that built this!
Show more
I am actually So Impressed by how many new capabilities Omnii-1 was announced with
Congrats @RadicalNumerics :)
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!
Show more
Cool.
CubeSnap is now on Android 🎉
Snap two photos of your scrambled cube and get a 3D step-by-step solution, in ~20 moves.
Google Play:
Apple/iOS App Store:
Show more
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#
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
#TIMEBestInventions#
Show more
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:
Show more
We resemble that remark. Doing our best to be ‘contrarian & correct.’ So far, so good! 🚀 🚀 🚀
OPEN CALL TO ROBOT DEPLOYERS!
@DimensionalOS is launching a 3-month residency program for hungry founders eliminating physical labor with robots in any vertical.
No equity. Just free hardware, warehouse space, compute, software support, and connections to customers across data centers, agriculture, construction, manufacturing, real estate, and more.
Applications close next Monday:
Show more
Already there. :) +@TriatomicCap, where one of our core investing themes is AI-powered ‘Engineered Biology’ — with portfolio companies creating the future of bio/acc.
@ManifoldBio
@EdisonSci
@RadicalNumerics
@ElegenBio
@ChemifyX
@ladanuzhna
more:
Show more
Already there. :) +@TriatomicCap, where one of our core investing themes is AI-powered ‘Engineered Biology’ — with portfolio companies creating the future of bio/acc.
@ManifoldBio
@EdisonSci
@RadicalNumerics
@ElegenBio
@ChemifyX
@ladanuzhna
more:
Show more
AI is driving first-principles rethinking of compute architectures, driven by enormous demand and completely new (and massive) workloads.
@semiDL and Volantis announced their Series A with a super clever new architecture with tiny optical waveguides on giant substrates to enable fast & vast memory pools, with >30X more bandwidth and >50X more memory.
How are you going to serve trillion-scale models?
-->
Show more
Excited to announce Volantis's $88M Series A.
We are solving Al's memory bottleneck by using optics, enabling chips with huge amounts of fast & cheap memory.
By boosting both the memory bandwidth and capacity per chip by orders of magnitude, we enable ultra-fast inference (up to 10,000 tps/user) for large models (>10T) - with low $/tok to boot.
Initially, this will enable insanely fast agents - think coding agents that finish in minutes or even seconds instead of hours.
More excitingly, optics is a fundamentally scalable way to increase memory systems. Not 2X/year, but by orders of magnitude across new generations. This will enable a structurally new Al industry, including restarting scaling laws, holding entire repos in context windows & more.
Our team has pioneered many core semiconductor technologies: the 1st CoWoS product, early HBM, the 1st silicon photonics CPO systems, the 1st high volume tunable VCSELs, the 1st processors to directly communicate using light & more.
We’ve already sent data >10× farther than equally tiny electrical wires inside a chip package. Our next iteration is already taped out and targets world-record bandwidth density over relevant distances, read more:
Show more
It's exciting when bio / science moves at AI speed – see @glebkuz & @ManifoldBio's note on their new mBER-2 protein design model enabling microbinders (compact protein formats from nature) and drug development components. Now already integrated in the Anthropic × Adaptyv protein design competition!
Show more
A nice small example of AI acceleration in biology happening in real time:
Last week, as part of our announcement of mBER-2, @ManifoldBio’s latest protein design model, we shared that we expanded design to 50 compact protein formats from nature, just 10-45 amino acids long.
We called them “microbinders.”
Just one week later, “protein microbinders” are already an official category in the Anthropic × Adaptyv protein design competition.
Google protein "microbinders” and the AI Overview already describes the category, citing our announcement.
Excited to see the community expand into these formats!
Show more
