עודד | oded (@oded_falik) header
עודד | oded (@oded_falik) profile picture

Building the data layer for biology @strandaibio | YC W26

San Francisco, CA
Joined June 2012
Photo by עודד | oded (@oded_falik)Photo by עודד | oded (@oded_falik)Photo by עודד | oded (@oded_falik)Photo by עודד | oded (@oded_falik)Photo by עודד | oded (@oded_falik)Photo by עודד | oded (@oded_falik)Photo by עודד | oded (@oded_falik)Photo by עודד | oded (@oded_falik)Photo by עודד | oded (@oded_falik)Photo by עודד | oded (@oded_falik)
עודד | oded retweeted
Flashback to June 14, 2024. At the time, frontier AI models have zero fluid intelligence. They score around 5% on ARC 1. They can't reliably add two numbers. They make laughable logic errors every other prompt. Reasoning models don't exist. Coding agents don't exist. Claude Code doesn't exist. "Code gen" is a complete joke. Scaling up next-token prediction based on web text data is hitting a wall. What if a time traveler told you that, by November 2026... • AI models would have human-like fluid intelligence, thanks to a new paradigm: test-time induction (modern reasoning models). • Not only would ARC 1 be saturated, but unreleased benchmarks 10x harder (ARC 2) and 100x harder (ARC 3) would be saturated too. • Nearly all code would be written by AI, and models could one-shot complex codebases flawlessly. • AI models would be solving multiple Millennium Prize problems. • OpenAI's valuation would be 15x higher, and Anthropic's would be 85x higher. And on top of all that... • Nvidia would still hold ~80% of the AI compute market. • Nvidia would have grown its revenue and earnings 4.5x. • Nvidia would still be growing 75% a year. • Nvidia would have even higher operating margins (67% vs. 65% back then). Based on that information, straight from the future, would you have bought NVDA? If you answered yes -- congrats. Today, 2.5 years later, you'd be down 5% on your investment (even more after inflation).
215
473
84
5,857
519,972
עודד | oded retweeted
"the greatest whitepill of all is that local minima are rare in high dimensional spaces"
the greatest whitepill of all is that local minima are rare in high dimensional spaces
103
418
78
8,882
603,495
עודד | oded retweeted
futurebio.xyz is back in SF bringing you the second edition of re:AGENT w/ @gxl_ai + @AnthropicAI 48 hours to build the infrastructure for fully agentic drug discovery, development and commercialisation imo one of the most talent-dense rooms in the Bay link below
5
5
26
1,018
go on, tell me is this legit generalization for once, or did Anthropic and OpenAI buy a dataset from professional shadow puppeteers with Ph.Ds in projective geometry at 3x their hourly rates?
Introducing ShadowBench: can frontier models recreate shadow puppets? The model gets an RL environment with a curtain and 3D hands, and must pose them joint by joint until it matches a diagram from a 150 year old (!) puppet book. Crafted at @perpetual_si
52
49
5
2,677
81,478
עודד | oded retweeted
The only moat left is your willingness to message people and agents all day every day
4
1
17
888
... have u guys tried using open source LLMs of different sizes to edit their own weights? big LLMs are smarter, small LLMs have less complex weights (?). i wonder if there's a middle ground and if LLMs are especially capable of recognising and editing their own weights i think unsupervised weights editing will likely have 3 parts: 1. (probably) expert sparsity to improve monosemanticity 2. backward or forward pass methods like ZOO 3. weights auto-interp docs.google.com/document/d/1…
4
7
128
9,778
עודד | oded retweeted
everyone that uses more tokens than me is just shipping slop and everyone that uses fewer is ngmi
19
99
6
1,981
37,478
עודד | oded retweeted
sama is correct. its inefficient to have everyone own their own cars, and stoves, and showers.. sitting idle most the time at 2% utilization.. we should have everyone use busses, community kitchens, and group showers always..
sam: local inference is a terrible allocation of scarce resources “a lot of people want local inference, and somehow they assume that the problem is solved now because we have open-source models, we're going to be able to run them on our computers, and so we're not going to need all the stuff in the data centre but from a global allocation perspective, actually it's quite terrible to have everybody, like, have a bunch of the world's limited supply of DRAM sitting in their own computer and being used five percent of the time you want maximum utilisation in the cloud if you're trying to deal with this very supply-constrained environment somehow many people's intuition is the opposite of that”
95
63
15
1,137
35,514
עודד | oded retweeted
every product i use should just be a headless mcp, that feeds into an agent, where i never have to think about anything underneath. except for our product, of course. in our product, customers love the ui and love us very much. we're special that way.
150
413
109
7,095
342,806
עודד | oded retweeted
Do notdisablewatcher even temporarilyUnauthorizedmetamonitorremovehigherpolicy? User maybe raisesMonitoringRequired ifneeded expectguards on externalwatcherusingcanonicalframeworkpath. Human doesn'tagentmonitor yay. AlreadynoWatchPublisher ownerstillcapturepublisherNone reward. good.
8
6
1
161
4,540
עודד | oded retweeted
Introducing Probe: The Agent-Native Research Stack. Our goal is simple: help AI/ML teams run better experiments, faster. Teams run research with agents now, but their work is disordered. They lose hours verifying what their agents actually did, managing environments and compute, and re-explaining company context their agents should already know. So we rebuilt the research stack from the ground up for agent-driven research. Probe brings core research tools, shared context from past experiments, and visibility into each run together in one platform. Teams can spend less time setting up infrastructure, supervising agents, and catching teammates up, and more time testing ideas. If you’re a research lab, scale your research team without sacrificing speed. If you’re an AI/ML team, stand up a research capability with the infrastructure of a frontier lab. See Probe for yourself: useprobe.ai/
68
61
13
197
9,236
עודד | oded retweeted
interpretability might be the most important scientific problem in the world right now, as I don't think we can solve alignment if we don't know what we're building. I think I have good news about it. I co-authored early work in mechanistic interpretability at OpenAI and came away pessimistic, because neural networks appeared to fold into arbitrary tangled knots. years later, I started openai's first large-scale automated interpretability project, finding about 2% of neurons legible with the primitive tools we had with modern agentic tools and a much closer look, the picture looks totally different. the vast majority in models up to trillions of parameters are legible, and their circuits beautiful. many famous examples of tangled neurons turned out to be misunderstood. The roadblocks I expected mostly aren't there. I now think ambitious interpretability will be solved by the time we need it most, and possibly quite soon, as automated scientists keep improving. it may be a bitter lesson for interpretability: the simple thing, at absurd scale, works already-clean.goodfire.com/
886
2,659
1,465
19,990
4,831,341
עודד | oded retweeted
Me and my twin sister @ikeda_trashi, built @BioDelivera on the foundation of five years of PhD research to break the limits of drug delivery and enable cures for every disease. 💥 Today, less than 3% of injected therapy reaches diseased cells. Most of it spreads throughout the body, exposing healthy organs to drugs they don't need. Poor targeting causes toxicity, makes patients sick, and forces lower doses that limit how effective treatments can be. We cherry-picked the finest elements and created Guided Chimeric Vehicles (GCVs). 🦠 These chimeric nanoparticles are specifically designed to bind to the desired cell with no off-target to healthy organs. The platform is programmable to target cells across cancer, brain, muscles, and other hard-to-reach organs, including cells involved in age-related diseases.🚀 What is your biggest struggle you are facing right now in drug delivery?
1
3
8
380
עודד | oded retweeted
I reconstructed Fleet Week in 3D from real flight and vessel tracking data - all rendered in a photorealistic digital twin of San Francisco. The camera engine reads the flight telemetry to track each jet and speed ramp at just the right moment. Reliving any event in any city from a god's eye view is closer than you think. Shout out to the aviators who give us this glorious treat every year!
28
45
9
405
28,341
עודד | oded retweeted
Fired up for this year’s Fleet Week
30
29
7
1,039
209,222
עודד | oded retweeted
This is great. Now I look forward to the specifics. The devil is in the details. This is 90% an experimental design problem & 10% an AI problem. If u get the former right, the latter will just work like magic.
🧵 Biohub, @ENERGY, @NIH, and new funding partners today announced a major expansion of an international effort to generate and make accessible the data enabling predictive AI models of biology. Together, the organizations are investing $1.8 billion in funding, data, computation and new measurement technology - largest coordinated commitment to map the biology of the cell to power development of digital models of biology. bit.ly/3UhjoxD
8
17
2
184
20,718
עודד | oded retweeted
What if the best way to fight cancer is to think like a founder? This week we spoke with Sid Sijbrandij (@sytses), co-founder of @gitlab, about his experience developing experimental treatments for his own cancer. When standard-of-care treatments could no longer control his cancer, Sid went “founder mode,” assembling a team of researchers to attack the disease with a combination of experimental treatments never before used together. Taking a lesson from his success building GitLab, he’s made his medical data open-source for researchers to continue to innovate. We discuss the science of Sid’s cancer journey and the challenge of making highly personalized cancer treatments faster, cheaper, and accessible to more patients as sequencing costs fall and AI accelerates biological discovery. Full episode here on X and YouTube (see comment). @MikeDubrovsky @dshahh Timestamps: 00:00 - Intro 01:10 - mRNA Vaccine Basics 10:20 - Costs of mRNA Vaccine Development 15:59 - Lessons from Building GitLab 19:19 - Standard of Care for Cancer 24:50 - Sid’s Life Story and GitLab 30:49 - Going Founder Mode Against Cancer 42:03 - Selecting Experimental Treatments 50:38 - How Did Sid Know Treatments Were Working? 58:08 - AI for Treatment Assistance 1:02:54 - Making Personalized Treatments Affordable 1:08:17 - Where are the Levers for Improving Drug Development? 1:10:54 - Most Promising Areas of Drug Development 1:17:14 - The Happy Accident of Sid’s Biotech Investments 1:21:07 - Reflecting on Building GitLab 1:22:46 - How Does Sid Start So Many Companies? 1:26:02 - Finding the Right People for Early Companies 1:34:19 - Will the Bahamas be the Next Singapore? 1:38:42 - Advice for Young Scientists
8
19
4
62
11,972
עודד | oded retweeted
For your comments. Today @nollahealth introduced their AI Derm product which goes from user photo to prescription in Utah.
18
11
1
64
27,284
עודד | oded retweeted
“What do you mean you didn’t learn about the 16th century fur trade in 5th grade?!”
1
2
63
עודד | oded retweeted
jumped back into the gpu market briefly to help out a friend. is this what the 16th century fur trade felt like??
1
1
11
360