😊 retweeted
The sf model: passing the bag on a bigger round is always more profitable than actually doing something
😊 retweeted
Maybe defense tech startups should try making weapons systems the customer wants to buy.
Defense tech startups are flush with investor cash and short on production contracts. That gap is the problem. Here are the numbers:
The top 100 defense startups tracked by the Silicon Valley Defense Group have raised more than $118 billion. In 2025, all 100 combined won $4.3 billion in government awards.
Government auditors found the Defense Innovation Unit made 450 prototype awards from 2016 to 2023. Only 62, about 14%, turned into production deals. That's roughly half of the prototypes that actually finished.
A prototype proves the gear works. Only a production contract pays the bills year after year, and investor patience can run out before the Pentagon gets around to buying.
The author's advice to startups: start shopping for a buyer while you still have at least six months of cash.
😊 retweeted
Ten million people have watched a comedian accidentally destroy the entire TED talk industry.
He filmed the talk once at TEDxNewYork in 2014 and openly admitted he had nothing to say.
Speaking coaches charge $5,000 to teach the exact tricks he used in 6 minutes to say nothing at all.
His name is Will Stephen. He is a comedian from Late Night with Seth Meyers who was invited to give a real TED talk and used it to prove the format is a costume anyone can wear.
Every TED speaker opens with a question, tells a personal story, cites a scientist, and shows a chart. Stephen did all four in six minutes with zero content.
His entire formula fits on a napkin.
Gesture with both hands. Ask the audience a question anyone can answer. Tell a mildly embarrassing story so they say aww. Point to a stranger on a slide and call them important. Flash a chart of random numbers. Wear glasses with no lenses. The audience will think you are brilliant.
That last trick alone has probably cost the public speaking coaching industry a hundred million dollars in sessions clients should have skipped.
"If you watched me on YouTube with the sound off, you would think I was a genius."
That is Will Stephen on the TED stage, three minutes into the talk he was paid to give. He wasn't lying.
Executives pay $15,000 for a keynote coach and still open with a joke the room never laughs at. Founders rehearse their pitch for weeks and still sound like Stephen at minute four, saying "wagga wah, gabba gabba" with authority.
The talk is free on YouTube. The method is written across every TED speaker's playbook for ten years.
Stephen still writes for Seth Meyers. Almost none of the ten million viewers have stopped nodding at empty slides since.
The formula is free. The willingness to actually say something before your next keynote, pitch, or townhall is the entire edge.
Gavin Baker says TSMC’s control over chip supply will prevent an AI bubble. But TSMC itself admits that customer demand forecasts can be misleading.
1. Gavin Baker — the supply constraint
“The fundamental shortage of wafers, which is really controlled by Taiwan Semi, will prevent a bubble.”
His argument is that limited chip supply naturally prevents the industry from overbuilding.
2. Sundar Pichai — the demand narrative
“We are capacity constrained on a daily basis.”
Pichai argues that AI is already being used, unlike the excess fiber capacity built during the dot-com boom.
3. C.C. Wei, TSMC CEO — the contradiction
“I put all together, all the truths together is not a truth.”
Wei explained that adding up individual customers’ demand forecasts can produce an unrealistic aggregate. TSMC checks construction progress and racks to make sure chips are actually going to be deployed.
Hyperscalers publicly insist demand is enormous.
Investors use those statements to justify continued AI spending.
TSMC controls the critical supply bottleneck and decides how quickly capacity expands.
Baker argues that this very bottleneck protects the industry from a bubble.
The industry’s bullish demand narrative and TSMC’s management of scarce supply reinforce each other, sustaining scarcity and investment.
TSMC isn’t simply managing capacity prudently.. they are deliberately restricting supply to preserve scarcity.
1. Hyperscalers create the demand narrative.
Google, Microsoft, Amazon and other cloud providers say they are capacity-constrained and need to keep investing. Their statements justify more spending on GPUs, data centers and power infrastructure.
2. TSMC admits that the demand signals cannot simply be added together.
On July 16, CEO C.C. Wei said:
“I put all together, all the truths together is not a truth.”
He explained that customers submit aggressive forecasts and that TSMC must independently judge whether those forecasts translate into real deployment. It checks construction progress, locations and racks to ensure chips are not simply entering invent
3. Yet the industry continues to validate the same long-term thesis.
TSMC says demand exceeds supply and is expanding rapidly. In October, Wei said TSMC was checking data-center construction and power availability, while maintaining its conviction in AI demand
4. The incentives reinforce one another.
Hyperscalers need to demonstrate that enormous capital expenditures will generate future returns.
NVIDIA benefits from sustained demand for its accelerators.
TSMC benefits from selling scarce advanced wafers and packaging capacity.
Infrastructure lenders benefit when long-term customer commitments support large financing packages.
5. TSMC becomes the critical gatekeeper.
If TSMC expands too aggressively, it risks creating excess capacity and destroying its own pricing power. If it expands cautiously, scarcity persists and the economics of the entire ecosystem remain stronger for longer.
That creates a structural incentive to manage supply carefully…even if customers collectively forecast more demand than the market can ultimately absorb.
😊 retweeted
Timothy Snyder -
“The point of Bloodlands was that we hadn't noticed a major event in European history: the fact 13 million civilians were murdered for political reasons in a rather confined space over a short period of time. The question of the book was: 'How this could have happened?' We have some history of Soviet terror, of the Holocaust, of the Ukrainian famine, of the German reprisals against the civilians. But all of these crimes happened in the same places in a short time span, so why not treat them as a single event and see if they can be unified under a meaningful narrative”
😊 retweeted
So terrestrial data centers will be economically viable compared to orbital? @elonmusk please advise
😊 retweeted
OpenAI’s automated classification infrastructure includes LLM-based classifiers, and OpenAI explicitly states that business data may be processed by automated content classifiers.
And it says the resulting classifications are metadata about the business data
A big big loophole for enterprise
“OpenAI may run any business data submitted to its services through automated content classifiers and safety tools, including to better understand how its services are used.”
“OpenAI says it is developing systems capable of identifying patterns across related interactions without giving OpenAI personnel access to the underlying content”
😊 retweeted
THE $1,000 CANCER DATA CHALLENGE
Who has the balls or vagina to take this on?
Give us an anonymized dataset of cancer vs. non-cancer patients with blood-based proteomics and/or metabolomics data.
We’ll go live, run the analysis, and deliver results in 30 minutes.
If we can’t correctly classify the patients, we’ll Venmo you $1,000.
Researchers. Biobank operators. Data hoarders.
Who has the data? Let’s put it to the test.
Parm, I came across a cancer-detection approach that I’m trying to independently validate, and I think there could be an interesting overlap with what you’re building at Precigenetics.
Their approach uses blood + mass spectrometry + phylogenetic analysis rather than a conventional biomarker panel. The 2023 material specifically describes a “cancer spectrum” and early detection in presumably healthy/asymptomatic individuals, while the newer deck describes mapping a blood sample onto a reference cladogram as Healthy → Transitional → Cancerous.
The interesting limitation is lead time. The model may identify a sample as moving toward a cancer-associated cluster, but the actual window before clinical diagnosis could potentially be anywhere from essentially 3 to ~36 months. Wondering whether your live-cell measurements could provide an independent biological readout of what’s happening during that transitional window.
Lemme know if interested!
😊 retweeted
For anyone claiming robotics data is a solved problem or already commoditized:
- Confuses egocentric data with all types of robot-training data
- Has never tested a robotics model zero-shot
- Has never experienced the pain (and joy) of overfitting
- Has never deployed a policy on real hardware, not just in sim
- Has never tested generalization beyond the tasks seen in training or standard benchmarks like robodojo
Robotics data is far from solved.
😊 retweeted
A hot take: jobs where you're just chatting with Claude or Codex to do your work may feel very comfortable right now and that's the trap, I'd not stay in that place if it was me, we're likely living a temporary phase where engineers are paid to act like a meat proxy and it will soon not being possible anymore
Very bad time for extremely introverted ppl that don't wanna talk to users or upskill in product
I don't still yet fully realize how we're not coding by hand anymore
I still remember the feeling of this stupid thing of naming variables
How did it go so fast?
We reached a point where the language doesn't even matter, engineering became discussing with a virtual engineer that is 100x better and faster than you
I think it's insane to now plan your life based on short term vision given how fast things will go in the next years. We're not talking about a 15 years horizon, things can radically change in just 2 years now
😊 retweeted
Open source libraries are a dying idea. Increasingly, using an external library that must be upgraded constantly to patch vulnerabilities is more expensive than generating your own version of the small part of it you use.
😊 retweeted
Databricks is overpriced and overkill for the vast majority of enterprise workloads.
The reason Databricks is the largest open source company in history is both structural and execution focused. Ben gives a great overview but as an open source founder I’ll try my hand as well.
Databricks is at its core a compute company. Their project Spark beat out all the other distributed data processing architectures like Hadoop and Cassandra to emerge victorious in the big data wars of the 2010s.
Partially this was because the other architectures were flawed and partially because they invested more in fixing them through hiring incredible engineers. This company did come out of a Berkeley PhD lab.
Still they struggled to make money. So in the mid 2010s they began making their next investment in a proprietary execution engine for Spark they called Photon.
As if building the leading distributing data processing library wasn’t hard enough they also shipped a native C++ vectorized query engine and continued both the public Spark roadmap and an internal proprietary roadmap as well.
It’s worth pointing out that this company is a decent example of the a16z thesis that you can kingmake with capital. Ben dropped a $10m check on a company that only asked for $250k, which is how the company had the war chest to make these investments.
If that’s all they had done, my guess is they’d be a $20B or $30B company today. I’ve made the point before that open source infrastructure expands forever.
That said there was a big second act. Circa 2018 they pioneered lakehouse architectures. The way to think about it that the traditional way you provision databases is you reserve some bucket on specialized compute and storage for them on your infrastructure. This made sense in a pre-cloud world where you had separate special-purpose servers, but less so in a world where you have scalable services like S3, EC2 etc.
So instead of reserving expensive DB compute and storage you reserve cheap S3 storage and compute that can spin up and down based on variable usage.
And now there’s a third act, namely building a cloud with AI primitives. The new gateway is a big piece of that but last month they also released a backend as a service with the primitives for building agents.
So again, why is it such a big company? Again, I think this is both structural and execution oriented.
The structural reasons: they started in an extremely lucrative market with a barrier to entry (it is very difficult to create popular open source projects in valuable markets!), that expands forever.
Second, they are a general purpose orchestration framework.
What I mean here is that every open source / devtools market is split between general purpose orchestration and specialized single purpose utilities. Think auth (Clerk) or email (Resend) for web apps (Next.js), or integrations (Composio) and sandboxes for agents (Mastra). While it is simpler from a product execution standpoint to build a single purpose utility, it is usually more valuable to build the general purpose orchestration framework and Databricks bears that out when you compare them to single purpose utilities doing things like ingestion and transformation (dbt, Airbyte).
Lastly as an orchestration framework / compute company, unlike database companies they are not limited to a particular workload (Supabase etc for transactional data, Clickhouse etc for columnar data), and they are in the core infra layer as opposed to being stuck on the client (Vercel/Next.js). It’s the $10B vs $200b difference
The execution reasons: they are technical geniuses who really want to win. I’ve spent the last decade building open source companies (Gatsby and now Mastra) and the typical open source founders are brilliant tinkerers and, in the best case, it takes a while to develop a ruthlessly efficient competitive commercial DNA. With Databricks they had both sets of DNA within the founding team — incredibly rare.
Congrats to @alighodsi and team on everything they’ve built so far!