phil bronner

Cofounder @ardent_vc, investing in b2b embedded finance, marketplaces, and vertical saas+. Father of 3, LUCKY husband, Wizards, and Commanders fanatic

[phil bronner @pbronner]

@AdamHPhillips No excuse!

[phil bronner @pbronner]

Coach has to be fired

[Warren Sharp @SharpFootball]

FIRE EVERYONE

[phil bronner @pbronner]

@michaelavon Totally agree! Complete bullshit!

[phil bronner @pbronner]

@jasonwhitlock Both but Quinn has to be fired

[phil bronner @pbronner]

@BillyTinkle @MagicJohnson He has to be fired

[phil bronner @pbronner]

@joshhart Speechless!

[phil bronner @pbronner]

Agree!

[Greg Finberg @GregFinberg]

Some quick takeaways:

• AJ Dybantsa looks awesome. Lightning quick first step + disruptive defense
• Kyshawn George has all the tools -- has room to develop at PG
• AD & Ayton will help tremendously on the glass
• A healthy Young = very good

Overall, a very promising opener

[phil bronner @pbronner]

@dinnerbites 100% still early days outside of physical ai

[phil bronner @pbronner]

With world models, simulating an action requires data. A company has to sit at the crossroads of a real decision and log what happens afterward. The action log is the input. Startups in the space have to get that data from somewhere to get the flywheel going.

[Carnegie Mellon University @CarnegieMellon]

The Griffin School of Computer Science will push forward education, research and technological development in an era when computing is no longer a specialized field but part of everyday life. It will rethink how computer science is taught, initiate research in new fields, provide unparalleled access to AI tools and expand student support.
https://t.co/Yo6iv7uDx4

[Governor Wes Moore @GovWesMoore]

$3.7 BILLION in investment.

14,000 jobs.

Welcome to Baltimore County, @anduriltech!

[phil bronner @pbronner]

@notloganhogg Yes. And it has to keep going. Take its own prediction as the new state and try the next action. One answer per action is still a forecast.

[phil bronner @pbronner]

I often see video generation models characterized as world models, but they are not. A world model holds the state, takes an action, predicts what happens next, and then continues simulating from its prediction. A video of a plausible future cannot be acted on.

[phil bronner @pbronner]

@AnishDAC Yep. The state is the agreement; the action is to pay late or move the asset; and the model has to roll the consequences forward rather than return a single answer. The hard part is learning that from real traces, not a human coding the state machine.

[phil bronner @pbronner]

Physical intuition is one type of world model. The same gap shows up when an action involves a price, a hire, or a deployment: the system can forecast, but it cannot predict what happens if you intervene. While the world model sector is earlier in those areas, it is progressing quickly.

[Anish Acharya @illscience]

This illustrates the challenge with personal agent unit economics.

It’s hard to play the game without a powerful patron in either venture dollars or lab partnerships.

[Geoffrey Hinton @geoffreyhinton]

The idea of an intelligence explosion caused by recursive self improvement has been around for a long time but until very recently it did not seem imminent. Now many leading researchers think it may happen quite soon. You can read our paper about it here:
https://t.co/sgUpugjpRY

[phil bronner @pbronner]

Useful rebuttal to the 'transformers learn incoherent world models' result. It is the same Manhattan taxi task, but mechanistic analysis shows the model does hold a map. the errors come from crowded neurons, not a missing model.

https://t.co/rHXRR9uIJ5

[Steve Rattner @SteveRattner]

So far, A.I.'s losers are routine office jobs like data entry keyers and customer service reps.

Its winners are data scientists and the trades building the data centers like construction laborers and electricians.

My @nytopinion Chart: https://t.co/3WCBRg3gMx

[phil bronner @pbronner]

👇🏾cool conference. applications open

[Shirley Ho @cosmo_shirley]

"The world is predictable in the right coordinates. The open question is whether a model can learn those coordinates on its own."

World model for Physics conference in Aspen coming winter 2027! Application deadline October 5 !

organized by @randall_balestr @KempeLab @ylecun and me 😀