Everyone is just stuck on the 6th problem of the International Math Olympiad, I guess.
RESEARCH
-
AI Model Struggles with Following Simple Em Dash Instructions
By
–
this seems to have reached a corner of the internet that my innocent soul wasn't ready for. "why do you hate em dashes so much?" this isn't about punctuation. this is about getting AI to follow simple instructions "what model was it?" 4o "why not use a thinking model? it works fine on o3" i don't want to have to use an expensive, slow model just to fix some typos. there's also a limit for o3 usage. "just add the instruction to exclude em dash to every message" yes, i can, but we shouldn't have to "it's not that hard to remove the em dashes yourself" not the point "write your own words lol" also not the point
-

New ODAC25 CO2 Capture Materials Dataset Released
By
–
It was a huge pleasure to work with the @AIatMeta and @GeorgiaTech teams to release this new and exciting ODAC25 dataset on CO2 capture materials.
-

Demis Hassabis predicts AGI within 5-10 years surpassing Industrial Revolution
By
–
"The difference here is, it’s going to be 10 times bigger than the Industrial Revolution, and maybe 10 times faster" Demis Hassabis, head of Google DeepMind and Nobel Prize winner, believes AI is on the verge of becoming AGI – in 5 to 10 years, machines could achieve all human
-
FastWan: Ultra-Fast 5-Second Video Generation via Sparse Distillation
By
–
FastWan
— AK (@_akhaliq) 4 août 2025
Generating a 5-Second Video in 5 Seconds via Sparse Distillation pic.twitter.com/6xQ4lLzNMLFastWan Generating a 5-Second in 5 Seconds via Sparse Distillation
-

Meta FAIR Releases Largest Open Direct Air Capture Dataset
By
–
We’re excited to introduce the Open Direct Air Capture 2025 dataset, the largest open dataset for discovering advanced materials that capture CO2 directly from the air. Developed by Meta FAIR, @GeorgiaTech
, and @cusp_ai
, this release enables rapid, accurate screening of carbon -
Meta Researcher Balances Industry and Academic Roles at NYU
By
–
This was long before I worked at FB/Meta.
But yes. I share my time between Meta and NYU. -
Continuous vs Discrete Prediction Space in Modern AI Models
By
–
This model makes predictions in continuous representation space.
An LLM makes predictions in discrete input space.
Prediction in continuous representation space is done with a (non-generative) Joint Embedding Predictive Architecture, i.e what I've been advocating for about 5 -
Learning Abstract Representation Spaces for Improved Model Prediction
By
–
I said it was impossible to predict all the details at the pixel level.
I said the solution was to learn an abstract representation space within which prediction could more easily be learned.
Which is *precisely* what this model does. -
Do New AI Models Actually Solve Real Use Cases?
By
–
I wonder how many people waiting for the next model have an actual use case that current models can’t solve (but are close enough so a new model might actually close the gap as opposed to requiring major breakthroughs or context window improvements)