RL post-training is hitting a rollout bottleneck. This new paper from #NVIDIAResearch shows how speculative decoding in NeMo-RL + @vllm_project can accelerate rollouts losslessly, with 1.8x higher throughput at 8B and projected 2.5x end-to-end speedup at 235B. Read the full
RESEARCH
-
RL Boosts Known Tasks But Causes Hallucinations on Unknown Ones
By
–
RL is a bit of a double edged sword: in known territory performance increases, but in unknown territory the model tends to hallucinate that it is performing a completely different task it was trained on
-
Why Measure AI Performance With Humans Rather Than Alone
By
–
This is a laudable goal, but it may mean designing, testing, and deploying AI systems differently. For example, why not measure the ability of GPT-5.5 to solve problems *with* a person rather than on its own? (see Stanford's centaur
-
Survey Proposes Taxonomy for Feed-Forward 3D Reconstruction
By
–
What if you could reconstruct 3D from 2D in a single forward pass, no per-scene optimization needed? Researchers from Zhejiang University, NTU Singapore, Monash, ETH Zurich & Uni Tübingen present a new survey on feed-forward 3D reconstruction. They propose a taxonomy that
-
Why Compiling Code Doesn’t Mean Correct AI-Generated Software
By
–
“Marcus’ specific point about coding is structurally important: a model that produces code which compiles and passes the tests it was given is not the same as a model that produces correct, secure, maintainable, well-architected software. The first is verifiable in seconds; the
-
Microsoft Research Paper on Training Computer-Use Agents
By
–
NEW paper from Microsoft Research. If you care about training computer-use agents, this is one to keep. (bookmark it) The team builds 1,000 synthetic computers (each with realistic directory structures, documents, and artifacts) then runs long-horizon simulations on top of
-
Co-Evolving Policy Distillation New AI Research Paper
By
–
Co-Evolving Policy Distillation paper: https://
huggingface.co/papers/2604.27
083
… -
Agent-Native Research Artifacts: The Last Human-Written Paper
By
–
The Last Human-Written Paper
— AK (@_akhaliq) 1 mai 2026
Agent-Native Research Artifacts
paper: https://t.co/cju1daCYH5 pic.twitter.com/tx9FpBDu22The Last Human-Written Paper Agent-Native Research Artifacts paper: https://
huggingface.co/papers/2604.24
658
… -
Heterogeneous Scientific Foundation Model Collaboration Research
By
–
Heterogeneous Scientific Foundation Model Collaboration paper: https://
huggingface.co/papers/2604.27
351
…
