GP @SPC, Pres Jimini, former CBO Immunai, CUEO @GuardantHealth, CEO iSkoot, EVP loopt, Pres DChoc, SVP iXL, founder/CEO Small World Sports/Software
San Francisco
Joined April 2007
- Tweets1.5K
- Following1.3K
- Followers1.5K
- Likes2K
Pinned Tweet
My great-great uncle Meyer Jacobstein en.m.wikipedia.org/wiki/Meye… was a 3-term Congressman from Rochester & opposed the Immigration Act of 1924
Mark Jacobstein retweeted
We’re betting our $5.5M seed round that the next leap in AI coding won’t come from OpenAI or Anthropic.
It’ll come from a new feedback loop that learns from you.
Introducing Beacon: open-source, self-improving memory for coding agents.
Your sessions capture how you debug and solve problems. Today, that context gets lost and you have to re-explain things every time you switch tools.
Beacon continuously distills your sessions into reusable skills across Claude Code, Cursor, Codex, and 20+ harnesses, so your next agent can learn from the last.
Beacon is free and OSS 🌟 github.com/Asymptote-Labs/ag…
Why should the frontier labs be the only ones winning with your sessions?
You already did the work. Now let your coding agents learn from you.
I love this metaphor. Like, this needs to go down in the annals like when @aileenlee coined "unicorn"
Jensen’s five-layer cake (energy, chips, infrastructure, models and applications) is an efficient way to describe the intelligence stack, but it’s flawed as a metaphor in my opinion.
Most cakes are uniform all the way across. They have a fixed circumference. The application layer ends up looking like a thin layer of frosting sitting on top of everything else. Its construction and ingredients are known.
What’s happening feels more like a banyan tree growing in time lapse.
1. The canopy keeps branching outward, creating new branches and more leafy area across science, industrials, healthcare, enterprise and consumer. Many of these applications are still impossible to imagine.
2. The different parts of the tree continually change one another. The roots and trunk enable the canopy, but demand and energy from the canopy also flow back down and determine how the system below develops.
3. And most interesting...a banyan’s branches send new roots back into the ground. Those roots become new trunks, which support still more branches.
That last part of the metaphor strikes me as most interesting. The most consequential applications won’t just sit on top of the existing AI stack. They’ll create demand for their own specialized models, infra, compute stack. At some point, an application is no longer just a leaf on a generic stack. It puts down roots and becomes a new stack.
The cake describes the layers. The banyan does a better job describing how this new age of intelligence will evolve.
It is an extraordinary time for the maximally ambitious: for those who would fundamentally rethink nuclear, space, solar, chips, algorithms, bio, security, or our relationship with the (extremely jagged) intelligent frontier. Better yet, build something so new we don’t yet have a category for it.
At @spc, we like to say, “we can just do things.” So we’re trying an experiment.
We’re investing $200K, uncapped, in a handful of “first they call you crazy” ideas, led by profoundly talented founders.
Our community is built to catalyze ambitious people and ideas. We’re experimenting with funding for the same reason.
Bring us your biggest idea. If this sounds like you, let me know below.
I'm helping to host (and judge!) a compbio hackathon at @SPC this weekend. 24 hours left to sign up to hack this weekend!
Bonus prize for every team: visibility within the @SPC ecosystem.
We're teaming up with @AnthropicAI, @awscloud, @biohub, @phylo_bio, @OwkinScience, @imyoohealth, & more...
For 15 years, Thinking, Fast and Slow (TFAS) was my standard gift for new hires. I was a product-oriented leader, and great products orchestrate system-one decision frameworks for their users. Along with books like Nudge, it shaped how I wanted products to work.
TFAS is back on my mind, my System 2 mind, since @typesafeai dropped JEV. For those who haven’t seen it, JEV is a “System 1” model that ingests unstructured context against a typed schema and returns an evaluation and confidence score. Much like our system one brain, it is very, very fast (and cheap), and usually right.
The implications are huge. Much of intelligence is high-speed “deciding,” not “generating,” and LLMs are an elephant gun for the mouse hunt of most agentic work.
If you’re exploring here, or want to, find me at @SPC, where we’re building the future of compute and intelligence.
🤖 Made with AI
Dropping some truth with @finn_meeks and our latest cohort of frontier builders at @spc
What an amazing time to build. We're looking at engineered biology, model alignment and interpretability, the future of work, agentic factories, space materials, next-gen compute, and so much more.
Sign up here:: southparkcommons.com/bio-dat…
@SPC exists to facilitate catalysis. We bring together the world's brightest, most ambitious frontier builders and push them to change the world.
Our bio-twist on our general thesis will take the form of a bio-hackathon from Oct 16-18. If you're bio-curious and want to hack on extraordinary data w/ exceptional people, please join us.
With @AnthropicAI, @awscloud, @biohub, @OwkinScience, @phylo_bio, & more...
“Every great advancement in medicine has come from seeing something that was once invisible.”
@Helmy_L2K, CEO of @GuardantHealth, joined me at @spc to talk about the journey from liquid biopsy to early cancer detection, what makes diagnostics such a difficult business, and how AI could help unlock new applications far beyond cancer.
In diagnostics, building the technology is only one part of the job. Catch more of my conversation with @Helmy_L2K
Mark Jacobstein retweeted
Over the last two days, we’ve had Qasar from Applied Intuition and Brendan from Mercor at @spc.
Both were invited and hosted by SPC members—Abe Ankumah and @nickmecklenburg. This is SPC at its best.
The magic of SPC is—and always has been—its members.
Mark Jacobstein retweeted
Some will find this demo startling. That's reasonable. What @nuance_ai built is startling.
But it's also a step toward insuring AI is both safe and helpful.
For AI to follow our intent, it needs to understand what we really intend. And what we really mean is in the nuances.
We’re excited to share that @nuance_ai has raised a $50M Series A led by @lightspeedvp, joined by @Accel, @spc, @nvidia, and @definevc!
Much of human communication happens beyond words. Before we can speak, we learn to read faces, return smiles, follow gestures, and sense when someone is listening. That visual and emotional vocabulary stays with us for life.
At Nuance Labs, we’re building a foundation model that understands and responds to the full spectrum of human emotion. A single full-duplex audio-visual model that sees, listens, and responds in real time, with the natural give-and-take of a conversation with a friend or colleague.
We believe the best interface with a machine is the one we’ve practiced since birth.
Join us!
Mark Jacobstein retweeted
We’re excited to share that @nuance_ai has raised a $50M Series A led by @lightspeedvp, joined by @Accel, @spc, @nvidia, and @definevc!
Much of human communication happens beyond words. Before we can speak, we learn to read faces, return smiles, follow gestures, and sense when someone is listening. That visual and emotional vocabulary stays with us for life.
At Nuance Labs, we’re building a foundation model that understands and responds to the full spectrum of human emotion. A single full-duplex audio-visual model that sees, listens, and responds in real time, with the natural give-and-take of a conversation with a friend or colleague.
We believe the best interface with a machine is the one we’ve practiced since birth.
Join us!
I love my job.
Frontier Tech Fall at SPC.
Leaders from @waymo, @anduriltech, @physical_int, and @AppliedInt are joining us.
Don’t miss out.
A stitch in time saves nine. Or, with semiconductors, perhaps 99.
When you miss a bug in software, you patch it and catch the next release. Miss one in a chip, and you have to respin the whole thing. That can cost hundreds of millions of dollars. Or more.
SukHwan Lim from TenX Semi stopped by @spc to explain why silicon has to be right the first time, and how they're remaking chip verification.
🤖 Made with AI
I’ll believe in AGI when it makes AV equipment and printers work every time. 🤣
"I'll believe in AGI, when it solves quantum gravity."
A line I wont't forget, from my chat with Prof. Rajesh Gopakumar (theoretical physicist, string theory pioneer, and also, AIR 1 in JEE)!
In this conversation he shares his "discover in India" philosophy, and why it's essential for our deep-tech future. Before the rockets, comes the actual rocket science!
"We are a many millennia old civilisation. We should be thinking in terms of long-term investments, and that's what will lift the country."
Everyone treats verification as the thing slowing chip design down. TenX Semi asks a different question: what could you do if it weren't?
Solve the bottleneck, and verification becomes a foundation for creating and optimizing designs.
Watch our full discussion at @spc below.
The entire modern economy runs on semiconductors, whose designs are one of the great marvels of engineering and FUs to entropy in human history.
Most of us don’t think a lot about how these chips work; we just enjoy the fruits of their (blazing, miraculous) labors—the software that sits on top and powers nearly everything we do.
And for those of us naive CS types whose last experience with EE was learning to build an adder, we assume that the semiconductors beneath our software are “always correct”. Silicon is deterministic, and the bits aren’t ever wrong, right?
Turns out it’s not that simple.
Chip-design organizations face growing pressure to do more with the same resources. As designs grow more complex, teams must move faster without lowering design quality. In the age of agentic chip design, verification is becoming the bottleneck, and AI cannot transform chip design when design and autonomous verification do not progress hand in hand. Solving this bottleneck before the chips are taped out (let alone shipped) is the work of TENX SEMI.
I’m proud to announce today that @SPC and @KhoslaVentures led TenX’s early funding. We’ve believed in their mission from the start.
The team of CEO SukHwan Lim (former EVP at Samsung, where he ran all US R&D for System LSI semiconductors), Subhasish Mitra (Stanford Professor and EE legend in robust and dependable computing and 3D nanotechnologies), and Shwetabh Verma (Broadcom, Apple, Cadence) is extraordinarily well-positioned for this business.
Subhasish’s foundational work transformed formal verification into a scalable, high-throughput reality. By shifting the paradigm toward automated consistency analysis, TenX Semi's autonomous verification platform uncovers bugs (including deep, hard-to-find long-tail bugs) without requiring additional test writing. The result is unprecedented speed and design quality, enabling teams to do far more with the same resources. At a top-ten semiconductor company, TenX Semi's platform uncovered critical bugs in a production design in less than two weeks, which would take many months to accomplish using business-as-usual practices.
Today, TenX Semi is advancing chip design by combining autonomous verification with AI-powered bug localization and fixing. In TenX Semi's vision, design and autonomous verification progress hand in hand, and bugs uncovered in the process are fixed autonomously.
Special thanks to Tathagata Srimani (T), co-founder and CTO of SPC portfolio company Flucta, for introducing me to the TenX team. Before T was a rising-star Professor at Carnegie Mellon, he studied under Subhasish.
The move from simulation to formal verification, and then to optimization and AI-assisted development, heralds an exciting new era for the semiconductor industry, the most critical infrastructure underpinning the entire AI era.
