Triatomic Capital

We lost a giant today. J. Craig Venter – pioneer of genomics, key driver of the Human Genome Project, founder of TIGR, Celera, and @JCVenterInst – passed away at 79. Thank you for everything, Craig. RIP. Photo with @StephenQuake at @SynBioBeta 2024, taken by @johncumbers.

Announcing a new Memory system for robots on Dimensional Robots in production generate thousands of hours of video, lidar, odometry, far too large to fit into your Agent context SpatialMemory2 builds a multimodal data store in latent space for your Agents Fully open source

AI inference is a pipeline, not a single workload. Speculative decoding paired with memory-centric architecture delivers lower latency, better GPU utilization, and faster responses at scale. The future of AI infrastructure is built by matching the right hardware to the right phase of inference. Read more: d-matrix.ai/how-speculativ… #AIInference #SpeculativeDecoding #GenAI #dMatrix

Check out my discussion with Chris Fregly and Antje Barth at their AI Performance Engineering meetup: - Why inference needs hardware heterogeneity - Technical challenges we need to solve to run agents on heterogeneous HW - Results these systems deliver youtube.com/watch?v=1i_9-4…

One of the more interesting enablers/cornerstones for foundation models wanting to do AI bio & drug discovery, imo, is Manifold Bio (@ManifoldBio) – which bridges in-silico/in-vitro approaches (which are becoming more commoditized) with the real-world/in-vivo 'RL data loop at-scale' needed for real progress: Manifold Bio: Barcoded Biologics Designing medicines that go where we want them to centuryofbio.com/p/manifold Introducing mBER: AI antibody design for in vivo-first discovery (open source) manifold.bio/news/introduci… Manifold Bio Demonstrates Million-Scale Experimental Validation of AI-Driven Protein Binder Design (w/ Nvidia at GTC) manifold.bio/news/manifold-…

New publication from @iPronics! Proud to share a new peer‑reviewed paper published in Nanophotonics Advances (Wiley), authored by our iPronics team. Read it here: bit.ly/4mLn4Bq Proud to see cutting‑edge research driving real‑world photonic innovation at iPronics. 👏

One of the most-viewed PNAS articles in the last week is “Chemputer and chemputation—A universal chemical compound synthesis machine.” Explore the article here: ow.ly/V53j50YKzbw. For more trending articles, visit ow.ly/2Mfa50YKzbA.

Efficient Computer named Data Tech Startup of the Year by @Data_Awards. As AI moves into the physical world, performance must come with energy efficiency. Built on the Electron E1 general-purpose processor and the effcc Compiler. Read more → ow.ly/n3pe50YO69B

In case you missed it – this is a great case of tech+AI driving rapid "GPU-ificiation" progress in biology & health. @glebkuz on their platform for massive multiplexing proteins in-vivo and in-vitro at @ManifoldBio (including partnerships with Roche & Nvidia). The future is coming fast.

Chemify - the world’s first vertically integrated molecular foundry connecting a chemical programming language to robotics powered by AI & ML. Dream it, draw it, make it.

Personalized cancer vaccines, including recent data from MSK at #AACR for pancreatic cancer, are demonstrating real & exciting progress. @ElegenBio is a key supplier for fast-turnaround, accurate personalized cancer vaccine inputs. Several cancer vaccine groups are working with them; others should & will.

Want 10×-100× better energy-efficiency than embedded ARM CPUs? –> go Efficient Computer (@efficient_hq).

Personalized cancer vaccines are far more effective when cancer is detected at early stage. 'Multi-cancer early detection tests' ("MCED"; including @GrailBio @GalleriTest) are an important enabler. GRAIL's test has 60%+ sensitivity in Stage I/II pancreatic cancer, per the CCGA study. #DetectCancerEarly #MCED

SOMEONE JUST BUILT AN OPERATING SYSTEM FOR ROBOTS not a framework... not a library an actual OS dimos is an open-source platform that lets you build, run, and deploy software on any robot humanoids, quadrupeds, drones, wheeled bots you write python modules .... they talk to each other via typed streams...just like microservices, but for physical hardware the wild part..? you can give your AI editor (cursor, claude code) MCP access to send real physical commands "move forward 1 meter" → robot moves. it ships with a full navigation stack SLAM, collision avoidance, route planning. plus a dashboard to debug, visualize, and control your robot in real-time via waypoints, VR, or keyboard. github.com/dimensionalOS/… supported hardware already includes unitree G1 humanoids, dji drones, and multiple robot arms. programming physical robots is finally as simple as programming software

Great to see Manifold’s #mBER AI protein design model in action! As part of the Amazon Bio Discovery launch this week, @AmazonScience performed agent-guided de novo design using three independent methods including @ManifoldBio 's #mBER to design nanobody binders against a novel cancer target, identifying several sub-nanomolar hits. mBER is now available via Amazon Bio Discovery.

Agreed a little hacky, but ... also really useful & cool. :)

I chatted with @ysmulki about MatX, chip design and where silicon designed for LLMs is headed (8:17) Tightly coupling SRAM and HBM on one chip (14:03) More MoE FLOPS, smaller KV cache load (16:08) Numerics: from 32-bit to 4-bit (19:02) Targeting both training and inference (22:14) Chip timelines (27:15) Logic and memory scarcity (29:42) Compute costs (32:07) Latency: from 20ms to 1ms as the new table stakes (40:50) Programming the chip (43:00) Starting MatX (47:11) Codesign without seeing the models (51:57) Interconnect design (55:44) Performance modeling philosophy (1:07:02) Prefill vs. decode (1:13:47) What's next

Why the next frontier AI models will be built on MatX (@MatXComputing):

Achievement unlocked: @ManifoldBio highlighted at the @NYSE for being named to @FastCompany 's Most Innovative Companies of 2026. Manifold Bio is building organism-scale AI to transform drug discovery Recent highlights: - Million-scale AI protein design collaboration with #NVIDIA. - $2B collaboration with Roche We’re just getting started.

The next frontier AI models will be built on MatX. @MatXComputing