Nate Tippens

The discussion around virtual cell models has primarily focused on algorithms and metrics. What about the training data? Is scRNA-seq good enough? A key challenge is distinguishing direct vs indirect effects on gene expression. 1/5

One big thing about competing with China on industrial policy is that it has long accustomed itself to capital indiscipline. The U.S. cares too much about capital discipline. And that's a problem because many of the sectors that China currently dominates aren't exactly great…

"zero progress... is what happens when you throw compute at the problem without understanding the underlying training data." equally true for biotech. youtube.com/watch?v=ziqsNe…

So key idea: scaling the wet lab experiments smartly + carefully , combined with simple architectures known to benefit from huge data

if you want to get seamless 230 tokens/s K2 for free, use this: > go to groq dot com, create api key > install cline in vscode/cursor > use kimi k2 using options as shown below groq and kimi combo is really shining for agentic use cases.

The more I'm looking into my "Short Academia" thesis, the more I'm realizing that breakthroughs have stagnated because of the annual debt that gets delivered as profits to universities. In 2022-23 undergraduates/graduates received $240 Billion in debt. BILLLLLLLION!!!!…

Otto von Bismarck is one of the most impactful people to have shaped the modern world, yet few know anything about him. I genuinely believe he is one of the most important figures for our age. Why? Because of artificial intelligence. Link below:

Here's Perplexity's breakdown of university admissions for foreign vs domestic students. Looks like an incentive issue rather than ideology. perplexity.ai/search/marc-an…

💯 PR for AI-oriented commercial research has now flooded multiple fields: variant effect prediction, perturbation prediction, single-cell integration, AI pathology, to name a few. Unfortunately, this has often led to surreal, incomprehensible papers. Some themes:

Financialization killed the US industrial base. The math is simple: when corporations give away 90-95% of profits to shareholders for 30 years, there's nothing left to build or invest with. Stock buybacks won't buy you industrial competitiveness.

If you don’t have time to take a walk, then you don’t have time to do science. Charles Darwin would take two walks every day on his "thinking path", not as a break from science, but as a crucial part of it.

Keith Rabois: “The velocity of your company improves by adding barrels” Keith shares his “Barrels and Ammunition” framework for building effective teams: “Most companies—once they get into hiring mode—just hire a lot of people. And you expect that as you add people your…

hanging out with strangers in a campus pub should be considered part of the serious process of thinking

Gentle reminder that transformers can, in fact, faithfully execute symbolic algorithms. They're just extremely slow in learning to do so. arxiv.org/pdf/2505.20896