Nima Alidoust

Today we are introducing Tara. Biological datasets are a source of insights and a means to train biological AI models. As the ability to reason at scale emerges, they take on a new role: the ground truth for testing what reasoning models produce, and the environment in which those models operate, get feedback, and improve. Tara, our autonomous research agent, is embedded in our ever-expanding datasets, lab-generated and synthetic, and built to test and evolve the hypotheses frontier models generate, matching the pace at which they produce new ideas. By keeping those models grounded in a vast space of high-precision biological data, we believe we can compound biological reasoning and close the impedance mismatch between hypothesis generation and validation.

@SabrinaHalper Good conversation!

"Aging is the process of a complex system slowly falling into its chaotic disequilibrium. And it's not like evolution necessarily even set a lifespan for us. It is merely the result of, once the system has been established, how long its intrinsic property stayed ordered before it became disordered." I’m excited to share my conversation with @jacobkimmel of @newlimit. Every species who has ever lived has one shared certainty: we’ll age, decay, and die. Jacob and his team are among the small group of people on earth with the expertise and funding to work on this problem. We talked about why we age, whether cellular aging can actually be reversed, how evolution shaped our lifespan and our peaks, and how AI is fundamentally changing the next era of science. Recorded at New Limit’s HQ in South San Francisco. Timestamps: 00:00 Intro 01:25 Why Evolution Limits Lifespan 03:13 Evolution Tradeoffs Explained 07:44 Why Women Live Longer 10:15 Lessons from Long Lived Species 16:20 Peak Intelligence 23:26 What is Aging? 27:59 Yamanaka Factors 32:18 New Limit 33:56 AI's Role in Progress 37:28 Organ by Organ Longevity 40:28 Is Immortality Possible? 42:04 Lifespan Progress 45:44 How to Test Longevity Drugs 49:30 Placebo Mind Body Link 53:56 What longevity hacks actually work? 57:25 Peptides And Stem Cells 01:00:34 Gene Editing 01:05:10 Personalized Medicine 01:08:48 Philosophical Dilemmas of Longer Lifespans 01:14:57 Living Off Earth 01:16:55 AI And Bio Risks

Today we’re announcing that @inductive_bio has won one of AI drug discovery's most prestigious competitions for the third straight time. Across these OpenADMET blind prediction challenges, we've now beaten hundreds of competitors, including participants from leading AI labs and pharma companies 1000x our size.

@saammotamedi 🩶

Today, we are announcing Greylock 18, a new $1.5 billion early-stage venture fund. A company begins when a founder sees a future others cannot yet see. There is no team, product, or revenue. Only a deep insight and the courage to start. Being early is lonely work. For years, being right can look exactly like being wrong. Greylock has spent more than six decades partnering with founders through those earliest moments to build enduring companies. Airbnb, Facebook, and Palo Alto Networks all began as improbable ideas driven by founders determined to make them inevitable. AI has made the map blank again. Every part of the economy is now open to reinvention. We believe many of the defining AI companies do not yet exist. We invest selectively and partner deeply with the founders we back. Each Greylock partner makes only one or two new investments a year because the work demands depth. We bring the full attention, network, and resources of Greylock to every partnership. We are builders backing outliers. Founders have their choice of investor. When a founder chooses to partner with Greylock, we understand the weight of that choice and the profound responsibility that comes with it. We seek to be a founder’s first believer and long-term partner. Greylock 18 is our commitment to what does not exist yet.

@jackdent @zavaindar @IndexVentures @kleinerperkins @sequoia @_DimensionCap @m_delamerced @nytimes Wonderful! Congratulations!

Thrilled to announce our $400M Series C led by @IndexVentures, with @kleinerperkins, @sequoia, @_DimensionCap and others. Great to speak to @m_delamerced @nytimes about how drug discovery using AI is moving from a promise to actual deployment.

@joshim5 @chaidiscovery @Novartis wonderful!

I'm delighted to announce a new collaboration between @ChaiDiscovery & @Novartis. It's been a pleasure working together over the past year, and we're excited to scale our collaboration to support broader deployment of Chai-3 across multiple therapeutic programs.

I had no idea: Argentina is the oldest team that has entered knockout stage for decades. Average age for the main squad is 30+ (35+ if you count the referees).

@BenSiranosian Is that really true? Im empirically seeing my sleep quality (interruptions in sleep) much worse when Im on my laptop immediately before sleep. Most nights 6 hour sleep without interruption end up being a lot more restful

@nalidoust 2, sleep is sleep!

Which case do you think will lead to better net overnight rest 1- screen until 11, sleep at 12, wake up 7a 2- screen until 11, sleep at 11, wake up 7a

@dauber x.com/juarez2505/sta…

I like this formulation and I think it rests on the notion that the distance between me and a more successful me is not as big as it seems when compared against the massive distance between the latter and God. I can also imagine a second function in the opposite direction: minimizing fear of failure and viewing it as catastrophe - the distance between me and a less successful me is also not *that* significant. Incidentally unshackling oneself from that fear does wonders in taking risks and going for more worthy aims.

So far in life I have found that the distance between the average and best human is very large but the distance between the best human and God is far greater still. SF people have this idolatry towards their billionaires and their founders, that makes them incapable of seeing them as humans. I believe that's a natural human instinct, especially prevalent in males, who want to submit to a clear hierarchy and someone better than them. This is often an honourable thing, especially when submitting to a Noble Leader and it's what allowed men to fight wars in the past and start and carry forward companies now. In the past, the concept of God was useful, because it offered a counterpoint to all this. It reminded people that even the Best Leaders were far below God. Now that we've lost our belief God, we project this desire for worship entirely onto other humans, I think at our peril. Whether one believes in the existence of God as such or not, it is undeniable that even the best humans are fallible. This is not a woke, egalitarian point, it's just the truth. And I increasingly think that the concept of God was adaptive in that it offered a personification for this intuition that all mortals are extremely fallible. Perhaps it's not coincidental that one of the gravest sins in many religions is that of worshipping other humans instead of God. In Christianity, one has to distinguish between veneration of saints for example (honouring them, asking them for their intercession) and worship (due to God alone). Many Protestants view even veneration as crossing the line into idolising humans. We are losing this at our peril.

Excited to be releasing FrontierFinance, the largest and most challenging open benchmark for evaluating AI agents across the full investment workflow! FrontierFinance is substantially harder than current finance benchmarks: Existing benchmarks like FinanceBench and Finance Agent focus almost entirely on data extraction. FrontierFinance spans diverse use cases across the full investment process: Screening & Discovery, Company Research, Sector/Industry/Macro, Earnings & Events, and Coverage & Catalyst Monitoring. Created for ambiguous, long-horizon agents: 220 examples paired with 11,543 expert-crafted rubrics, following Samaya's Criteria Eval methodology. The rubrics are what let us evaluate the reasoning and steps behind a true expert-level output, not just a plausible-looking one. Evaluations: We evaluated Claude Fable 5, Claude Opus 4.8, GPT 5.5, Gemini, open-source models including GLM and DeepSeek, and others. We used the same public rubric and a standard harness for financial tasks. Samaya's AI system reached state-of-the-art accuracy at 50.8%, at 4x lower inference cost than Fable 5. Next best was Fable 5 (49.2%), then Opus 4.8 (45%) and GPT 5.5 (43.5%). We're releasing the benchmark, methodology, and full evaluation results - see link in comments. Future releases: FrontierFinance was curated from Samaya's larger internal set of ~5,000 examples, and we plan to release subsequent, harder benchmarks as well as a more detailed technical report!

@DeryaTR_ Add the Persian-Turkish early fusion and it gets combinatorially larger :))

@nalidoust Just the different Persian rice cooking recipes will require enormous training data. ☺️

Announcement: My mom is joining Anthropic to teach LLMs how to cook Persian food *without* recipes.

@bznotes I mean.. are you including that Barry Diller outfit too as evidence for your statement?

Men used to dress better.

@cyrusnewday :)))

@nalidoust Now that’s what I call teaching models taste 🤣

@LMUrbisaia Hence, her joining them. Coming without recipe is what she and every Iranian mom has done for centuries :)

@nalidoust Lack of proper recipes is dangerous. LLMs will go psycho and start recommending fesenjoon without walnuts... or worse yet, no pomegranate

@anshulkundaje Have you heard of “tarof”? She did some of that w them. Now she’s paying for the tokens out of pocket

@nalidoust Is she getting paid 60K or 300K

@ImranSHaque @genophoria That is the true test!

@nalidoust @genophoria Claude scaling great tahdig will be the sign of true ASI (artificial starch intelligence)

@YEktefaie this is amazing yasha!

Excited to share PhylaFlow was selected as one of two Best Academic Papers at the ICML 2026 GenBio Workshop. This project involved many iterations and a lot of uncertainty, so it’s especially rewarding to see the ideas resonate with the community. I’m grateful to my co-first author Jiabo Cui, collaborator Shrey Jain, and mentors on this project @marinkazitnik and @PardisSabeti , as well as the @genbio_workshop organizers and reviewers. Looking forward to continuing to push on AI methods for evolutionary biology and scientific discovery! #ICML2026 #AIforScience #ComputationalBiology #MachineLearning #Phylogenetics

@genophoria Parsley: as much as needed. به مقدار لازم

@nalidoust Claude, make ghormeh sabzi... make no mistakes!