“Latent Spatial Memory for World Models” world models need a 3D memory to avoid drift, but RGB point cloud caches are slow because each step renders pixels and re-encodes them into latents via a VAE. The Mirage approach
RESEARCH
-
Policy brief on DeepSeek AI talent base for U.S. policymakers
By
–
U.S. policymakers must tackle both problems, as noted by the authors in last year’s policy brief published by @HooverInst
’s Technology Policy Accelerator and @StanfordHAI
. (4/4) Read their recommendations here: https://
hai.stanford.edu/policy/policy-
implications-of-deepseek-ai-talent-base
… -
US faces dual AI talent loss to China
By
–
The U.S. faces two distinct talent problems: 1) Chinese AI researchers training in the U.S. are returning to China at high rates; and 2) China’s domestic AI talent pipeline is increasingly contributing to frontier models without any U.S. training. (3/4)
-
New data shows US AI talent advantage no longer permanent
By
–
One year after their 1st white paper analyzing DeepSeek’s talent base, HAI senior fellow @AmyZegart and @HooverInst
's Emerson Johnston publish new data that shows even more clearly that the U.S. can no longer assume a permanent AI talent advantage. (2/4) https://
hoover.org/research/updat
e-deepseek-ai-and-the-great-talent-competition
… -
DeepSeek V4 signals China’s competitive AI research talent pipeline
By
–
When DeepSeek released its V4 model in late April, it received mixed reviews. But when viewed from a talent perspective, it acts as another clear signal that China’s domestically grown AI research pipeline is highly competitive. (1/4)
-
A nuanced essay on AI advances
By
–
A good essay by @pgasawa and @profjoeyg on a more nuanced view of AI advances.
-
Future brief access to top AI models for everyone
By
–
In the future everyone will have access to the SOTA models for five minutes.
-
OrcaRouter detects hallucinations and ensures reliability
By
–
A model has a bad day and your output collapses.
OrcaRouter makes reliability structural, not a matter of luck.
If a leg hallucinates, the judge detects it and provides the answer that holds the road. -
Robot drives to EdgeCity for world model talk
By
–
I'm on the way to the @JoinEdgeCity conference where I'm giving a talk on world models. My robot's driving (actually a world model), and I'm reading this getting me up to date. It is a two-hour drive. And I am on an audio space here on X. Life is awesome. Great write up
-
Meme of Mistral’s giant cat model with infinite benchmarks spreads
By
–
The le chaton fat meme is leaking to the outside world and I expect to be asked about Mistral's new ginormous cat model with infinite benchmark scores at my next meeting with corporate leaders. I guess it is better than being asked about the "MIT pilot AI study."