World models don't run on scraped text. They need continuous, multimodal observation of the physical world: labeled, time-stamped, geo-grounded. Voxel51's @_JasonCorso_ writes for @FortuneMagazine, "AI models are choking on junk data."

The 15-minute high-level board update is a trap. Without the details, advisors just pattern-match onto other companies and give generic, parrot-like advice. My take: Boardrooms need to operate like university labs. Block out 90 mins for 1 hard topic. Detail matters. TAF Series #18: medium.com/@jasoncorso/deep-…

"Disagree and Commit" is the hardest shift for technical founders who are used to being "right." If you're still second-guessing a team decision 2 weeks later, you're the friction. My take: Conviction > Certainty. Read the full essay on navigating constructive confrontation: medium.com/@jasoncorso/const…

Show me the impact not the paper count. Paper mills outputting LPUs are a cancer to the field. And the data seems incomplete. In aggregate what about all of the institutions less than paper threshold here.

When we started @Voxel51, I read nearly every startup book I could find. @johndoerr "Measure What Matters" was a big one. My real take: For a small team, A North Star > 50 KRs. Read full essay: medium.com/@jasoncorso/taf16…

Three months into the Technical Academic Founder (TAF) series. This week #15 is a Redux of the first 6 essays: a roadmap for starting, fundraising, and operating a startup with a specific emphasis on technical founders and academic founders. Full piece here: medium.com/@jasoncorso/taf-r…

When to Think vs. When to Look New @CVPR 2026 paper available --- "Uncertainty-Guided Lookback for Vision–Language Models". A deep dive into reasoning in VLMs! with @ChenliangXu and many collaborators By analyzing token-level perplexity, we discovered a clear pattern: successful reasoning traces repeatedly "re-anchor" to the visual input, while failed ones drift into ungrounded textual speculation. To address this, we’re introducing Uncertainty-Guided Lookback. It’s a training-free decoding strategy that: 🔥 Detects when a model’s reasoning chain is drifting into a visually uncertain regime. 🔥 Triggers short, adaptive "lookback" prompts to refocus the model on the image. 🔥 Improves accuracy by up to 6.5 points in specialist domains while reducing token usage by 35-45%. It’s a reminder that in the rush toward massive compute and longer context, the most effective path forward is often the one that remains most grounded in reality. Project Page: proj-visual-thinking.jing.vi… Paper (arXiv): arxiv.org/pdf/2511.15613

Negotiating a term sheet is the first time most academic founders actually "fight" for their vision. My take: Valuation is for the ego; Liquidation and Board Control are for the exit. Don't trade the latter for the former. Read full essay: medium.com/@jasoncorso/valua…

What a delight to be on the @_odsc podcast!

Faces! Faces! Faces! Check out our language-guided face modeling work to appear at CVPR 2026. Introduces method and a new 80-hours of synthetic facial motion paired with rich, natural language descriptions. Project page: songluchuan.github.io/TDMM-L… @ChenliangXu

Valuation isn’t a trophy—it’s a performance hurdle. Setting it too high today creates a "debt to the future" that leads to "hiring walls" and unreachable milestones later. New essay on managing valuation risks and pitfalls: medium.com/@jasoncorso/valua…

Mistakes! Mistakes! Mistakes! Physical AI assistants today can tell you "that's wrong"—but not *what* went wrong, *when* it became irreversible, or *where* in the frame the mistake lives. That's like a teacher who only marks an X on your paper without any explanation. Check out our new CVPR 2026 paper the enriches this problem with new data, new challenges, and a new model! 📄 Paper: arxiv.org/abs/2511.20525 🌐 Project page: yayuanli.github.io/MATT/ 💻 Code: github.com/yayuanli/MATT 🤗 Dataset and Weights: huggingface.co/mistakeattrib… @UMRobotics @Voxel51 @UMichECE @michigan_AI

I just finished Alien Clay by Adrian Tchaikovsky. What a book! I can't stop thinking about one thing: what happens when you strip away every assumption about how a system should work? Highly recommend the book; and reflecting on its meaning about the systems in which you play

Wow five of the six ICML papers I reviewed were withdrawn. Mixed feelings about this.

Valuation isn't a math problem; it's a narrative one. 📊 Culture shock: academics think about money in terms of costs; founders need to value a dream. My new weekly essay in my Technical Academic Founder series is live: medium.com/@jasoncorso/valua…

I loved the DC cherry blossoms so much I built some LEGO-blossoms for my desk. 🌸 The twist: Modern AI is exactly like this—a "bricolage" of fragmented tools and messy data. The "pipeline" is dead. Long live the random walk. Read why here: medium.com/@jasoncorso/pink-…

Talking AI at the White House this morning!