Tyler Sosin

Partner @MenloVentures

[Tyler Sosin @tysosin]

@tjparker Happy bday - stay healthy!

[Tyler Sosin @tysosin]

@m_franceschetti What have you built so far and which automations are providing the biggest gains?

[Tyler Sosin @tysosin]

@adamyounggolf D1

[Tyler Sosin @tysosin]

@ajhodls @arian_ghashghai Any hypothesis as to why?

[Tyler Sosin @tysosin]

@rohitdotmittal This is why I’m building Villain Capital - bet on great companies in niche or perceived niche tams.

[Tyler Sosin @tysosin]

@kushalbyatnal I am totally aligned with this.

[Tyler Sosin @tysosin]

@sandykory Agreed :)

[Crémieux @cremieuxrecueil]

The most important thing in this package is that the White House is now going after Obamacare's Medical Loss Ratio requirement.

Ending the MLR will enable a revolution in healthcare and it will constitute the largest anti-trust action in American history.

It has to be done.

[Perry E. Metzger @perrymetzger]

99% of the work that tradesmen did in 1800 has been completely mechanized and automated. It used to be that a carpenter literally had to shape logs into usable boards and studs and floor planks with hand tools far worse than what people can make now. They had no nail guns, they had no power saws, there were no powered planers to produce smooth flooring. Somehow, though, not only do carpenters still exist, but there are more than there were in 1800.

Even though 99% of the intellectual and physical labor people did in 1800 has been completely mechanized, we still all have jobs, even loads of jobs doing manual labor, even jobs doing white collar labor even though almost all white collar labor done in 1800 ceased to exist long ago.

Calculations had to be done by hand in 1800 by extremely smart and skilled people, even complicated engineering calculations. All accounting used to be done by hand. Every business had to employ legions of clerks who were not just literate but had to be quite skilled. (As recently as a century ago, there many large insurers and banks in the US that *each* employed *tens of thousands* of file clerks and accountants.)

Yet, even though all that skilled intellectual labor has been automated away, we still have plenty of work for intellectual labor for people to do.

I see lots of people say "AI is different, you just don't understand!" But I do understand; in the future, AI systems will be able to literally do everything a human can do. I fully understand that we will be able to build lots of AIs and robots, at a pace far faster than human population can grow. However, other humans can also do literally everything I can do and yet I still have work. Wants are unbounded; even with AI, labor will be finite.

The mystery that people need to explain here is not the future but the past; if your economic theory doesn't explain why it is that we don't have 99% unemployment today even though 99% of the work people used to do is long gone, then you can't even begin to think about the future.

Here is the key. The naive, zero-sum thinking approach says that the number of jobs is limited by the amount of work that needs to get done. This is utterly wrong. Instead, the correct claim is that the amount of work that can get done is limited by the number of minds and hands we have available.

The future will not be one of poverty with people displaced from work because there's literally nothing for people to do, it will be one of tremendous wealth and health. Just as we are now orders of magnitude wealthier than people in 1800 were, all because mechanization has increased the amount of stuff people can make with a given amount of labor, in the future we will be orders of magnitude wealthier and more comfortable still, because mechanization will continue to increase the amount of stuff we can produce with a given amount of labor.

The future isn't grim, it's glorious.

[Nate Solon @natesolon]

Don't worry, you're not bad at chess. You're actually bad at everything. Chess is just one of the only things with a good rating system.

[Eren Bali @erenbali]

I understand why many smart people feel this way but I’m not worried about this scenario one bit. In the heydays Google and Facebook there were similar predictions. Google was going to swallow the Internet, FB apps were going to replace everything etc. They weren’t the slow incumbents we think of them today. They were scary. I wasn’t around in Microsoft’s heydays but I bet it was similar.

One company to rule them all never works out. Especially in the application layer where every design decision is a trade off. That’s why even in the same category, you can have many successful companies based on minor differences. There isn’t one way to find restaurants, learn things, connect socially, organize an event or shop online. You can always find weaknesses of an existing service and build something better for certain customers.

If anything, the foundational model companies have much weaker moats than Google, FB and MS had. No models have a monopoly on anything. Distribution and capital is way more accessible for startups than it was 10-20 years ago.

OpenAI and Anthropic have some momentum right now. But when they’ve to compete in 10+ categories, you’ll be competing with a PM there, not their founders. These organizations also have significant cultural weaknesses you can leverage. Their coveted researchers want to solve math problems, not hear complaints from soccer moms in Ohio or compliance teams of regional hospitals.

So I’ll say game is on. You can’t win if you don’t play.

[Spencer Baggins @bigaiguy]

🚨 MIT just humiliated every major AI lab and nobody’s talking about it.

They built a new benchmark called WorldTest to see if AI actually understands the world… and the results are brutal.

Even the biggest models Claude, Gemini 2.5 Pro, OpenAI o3 got crushed by humans.

Here’s what makes it different:

WorldTest doesn’t check how well an AI predicts the next word or frame.
It measures if it can build an internal model of reality and use that to handle new situations.

They built AutumnBench 43 interactive worlds, 129 tasks where AIs must:

• Predict hidden parts of the world (masked-frame prediction)
• Plan multi-step actions to reach goals
• Detect when the rules of the environment suddenly change

Then they tested 517 humans vs the top models.
Humans dominated every category.
Even massive compute scaling barely helped.

The takeaway is wild:

Today’s AIs don’t understand environments they just pattern-match inside them. They don’t explore, revise beliefs, or experiment like humans do.

WorldTest might be the first benchmark that actually measures understanding, not memorization. And the gap it reveals isn’t small it’s the next grand challenge in AI cognition.

(Comment “Send” and I’ll DM you the paper 👇)

[Dan Collins @DanCollins2011]

Mayor of Beijing: 🇨🇳
- PhD Engineering
- 30 yrs rising thru ranks of Peoples Bank of China
- 7 yrs experience Deputy Mayor

Mayor of NYC:🇺🇸
- Undergrad in African studies
- First job: NYC Mayor

[unusual_whales @unusual_whales]

Average US electricity price over the years, per Axios:

[Dr. Jon Slotkin @slotkinjr]

As a neurosurgeon I care a lot about road safety.

By now you’ve probably seen @Waymo’s stunning safety results (like 91% fewer serious crashes). But they didn’t just publish data headlines. They released the raw CSV files and data dictionaries.
I did a much deeper analysis. A fascinating story emerges when you analyze how they’re achieving this.

This isn’t incremental improvement - it’s categorical. We’re looking at the potential elimination of traffic deaths as a leading cause of mortality.

The intersection breakthrough: Waymo has essentially solved intersection crashes, with 95% fewer injury incidents than human drivers in the same locations. That’s transforming the deadliest driving scenario.

The national math: If every US vehicle performed like Waymo, we’d prevent 33,000-39,000 deaths annually and save $0.9-1.25 trillion in societal costs. Even partial adoption at 27% would save ~10,000 lives per year. In terms of magnitude, this would be the equivalent of eliminating every pedestrian death nationally in a year.

The physics signature: Here’s what fascinates me: 47% of Waymo’s contacts involve less than 1 mph delta-V. They’re not just avoiding crashes; they’re converting unavoidable incidents into gentle bumps. It’s like having physics itself on your side.

We’re not talking about marginal safety gains. The data represents a fundamental shift from harm reduction to harm prevention.

The methodology matters: I used their dynamic geographic benchmarks (comparing like-for-like road conditions) and verified the findings hold across San Francisco, Phoenix, LA, and Austin. The safety advantage actually increases in more complex urban environments.

Link to raw data below….

Notes on my approach:

Analysis based on 96 million miles of Waymo Rider-Only (RO) data through June 2025, utilizing Waymo's dynamic geographic benchmarks to compare Waymo Driver performance against human drivers under similar road conditions and operational design domains.

The projections for national impact (deaths prevented, societal costs) involve several assumptions. Given Waymo's zero reported fatalities, the direct serious injury reductions were mapped to national fatality statistics using established NHTSA-derived ratios that correlate serious injury crash rates with fatality rates. This extrapolation assumes that Waymo's observed serious injury prevention capability would translate proportionally to fatality prevention. Societal cost savings are estimated by applying average per-fatality and per-injury economic costs (e.g., medical, lost productivity, quality of life) as published by NHTSA, scaling these national averages to the projected number of avoided fatalities and injuries based on Waymo's safety performance. These figures represent the potential annual impact if the Waymo Driver's safety profile were widely integrated into the national fleet.

@ethanteicher

[Tyler Sosin @tysosin]

@endowment_eddie @jmj How about purposefully investing in businesses / ideas pursuing smaller TAMs? Tam seems like the issue no large firm can get comfortable w /, yet is also one of the biggest sources of omission errors.

[Tyler Sosin @tysosin]

@CliffordSosin 🙏

[Ptuomov @ptuomov]

How we screwed ourselves on rare earth elements

TK;DR: Jimmy Effing Carter

“In 1980, a mis-classification of rare earths had catastrophic consequences for US rare earth mining. The Nuclear Regulatory Commission and the International Regulatory Agency placed rare earth mining under the same regulations as mining thorium – a radioactive element that drops out when processing heavy rare earth minerals like monazite.”

“New, onerous regulations on thorium made the mining and refining of thorium-bearing rare earth elements risky. Over the next two decades, the US rare earth mining industry collapsed. Defense One notes that, even though American mining companies extract enough rare earth ore, through mining other metals, to meet 85% of global demand, it is discarded because the regulations make it uneconomic to mine. How’s that for irony.”

[Tyler Sosin @tysosin]

@nicochristie I love this idea. As a solo GP need all of the financial analysis automation I can get. Would love to try to the product if possible.

[Sebastian Caliri @SebastianCaliri]

Reviewing a completed trial in the US takes 10-12 months. In China it’s 60 days.

Preclinical work takes 3+ years in the US. Better regulatory processes in China get it done in 1-1.5 years.

It can’t be overemphasized how dire this is and how aggressive we need HHS / FDA to be.