I think the evals overestimate ability in creative or out-of-distribution tasks.
AI
-
Nemotron 3 Ultra and the Open Model Landscape broadcast
By
–
Nemotron 3 Ultra and the Open Model Landscape | Nemotron Labs https://
x.com/i/broadcasts/1
rGmqqeRorDGy
… -
Long video reasoning with DeepSeek sparse attention and multimodal MoE
By
–
"Kwai Keye-VL-2.0 Technical Report" This paper makes long-video reasoning much more feasible by adapting DeepSeek Sparse Attention to a GQA-based multimodal MoE, reaching 256K context with only 3B active parameters. As dense attention makes hour-level context way too expensive,
-
DeepSeek’s $6 million claim misleading, excludes billions in prior costs
By
–
The DeepSeek moment was extraordinary PR.
— Nina Schick (@NinaDSchick) 16 juin 2026
A claim that a frontier model was trained for $6 million wiped over $1 trillion from US tech markets overnight.
But the number is misleading.
The $6 million refers only to the final training run. It excludes the billions already… pic.twitter.com/OmMlqjdV2EThe DeepSeek moment was extraordinary PR. A claim that a frontier model was trained for $6 million wiped over $1 trillion from US tech markets overnight. But the number is misleading. The $6 million refers only to the final training run. It excludes the billions already
-
Sora shows hype leads to number one but not profit
By
–
7/ At the end of the day, Sora proved something the whole industry is about to face. Hype gets you to number one. It does not pay the compute bill. And the bill always comes. Source: OpenAI, with figures from Forbes, Cantor Fitzgerald, WSJ, and Appfigures.
-
OpenAI killed Sora; Disney learned last minute, no deal.
By
–
5/ Three months later, OpenAI killed Sora. Disney reportedly found out less than an hour before the announcement. No agreement was ever signed. No money changed hands.
-
Disney licenses characters to OpenAI for Sora video generation
By
–
4/ Then in December, Disney agreed to a deal worth around $1 billion. Mickey Mouse, Marvel, Pixar, Star Wars, all licensed for people to drop into Sora videos. An OpenAI exec called it a shift on the level of the end of silent film.
-
Sora Launch Shocks Hollywood, Threatens Film Careers
By
–
1/ Sora launched in late 2025 and hit number one on the App Store in its first week. Over a million downloads. It scared Hollywood. The demos looked like the end of film as a career.
-
OpenAI’s Sora: from media success to failure in six months
By
–
OpenAI created the most media-covered AI video application on Earth. It cost them millions per day to run. But it only earned them $2.1 million over its entire lifespan. Here is how Sora went from number one to dead in six months:
-
GLM-5.2 Released with 1M Context Window and SOTA Score
By
–
ZAI: GLM-5.2 is now available on Hugging Face! > It comes with a 1M context window and 2 reasoning effort levels: max and high. MIT license and same pricing as GLM-5.1. > GLM-5.2 achieves a 46.2% score on DeepSWE, the SOTA benchmark.