
Reflection AI's Beam Ran 1.3 Billion Sandboxes on 10,500 GPUs
Reflection AI reveals the raw infrastructure behind Beam's reinforcement learning run and commits to OpenEnv as its future environment standard.
A focused feed of models, agents, research and open-source releases for people building with AI.

Reflection AI reveals the raw infrastructure behind Beam's reinforcement learning run and commits to OpenEnv as its future environment standard.

Sakana AI unveils Sawara, a from-scratch Japanese foundation model, alongside a three-pillar strategy spanning pre-training, post-training, and orchestration.

A Chinese-language adaptation of Stanford's CS336 bundles 16 chapters and 6 hands-on assignments covering the full LLM stack, from BPE to GRPO.

OpenAI pulled frontier models offline, built a cached web for them to train in, then turned its own cyber agents loose to break the walls.

Two overlooked floating-point bugs in softmax and cross-entropy have been quietly corrupting gradients. A new paper fixes them in a few lines of JAX.

A new two-stage training recipe lets looped language models reuse a single KV-cache across every recurrence, cutting HBM use up to 66%.

IFM open-sources the agentic reinforcement learning stack behind K2 Horizon, running 8,192 concurrent sandboxes and 1.3M rollouts across seven days.

New research shows hidden model preferences slip through distillation even when teacher and student share no base model, data, or architecture.

Prime Intellect ships a new default RL loss built for async training, plus a 1.15x throughput bump over torchtitan on Qwen3-235B.

Meta's Prometheus supercluster hits 85% goodput and 47K GPU-hour mean time between interruptions across hundreds of thousands of GPUs training Muse Wonder.

A new analysis settles whether momentum-based optimizers can get stuck at saddle points, and designs a variant that escapes faster without hurting convergence.

A one-shot optimal transport method aggregates independently trained mixture-of-experts models into a single global estimator without pooling raw data.

A new unified framework reframes VAE design for diffusion as a spectrum-matching problem, boosting generation quality on CelebA and ImageNet.

Zephon plugs into TorchTitan to deliver identical global batches across GPU counts, so a job resumed on different hardware trains on the exact same data.

A new study shows that proxy models used to pick pretraining data can flip their verdicts with tiny learning rate changes, with a simple fix.

Alibaba's Qwen team unveils an elastic RL framework that trains agents on hours-long tasks, cutting training time by up to 44% while lifting repo-generation accuracy by 6 points.

New research finds smaller batches and tuned recipes dramatically improve pretraining when you have more compute than unique data.

Ai2 replaced its priority-based GPU scheduler with time budgets and fair-share allocation, cutting median queue wait from 5 minutes to 24 seconds.

A new residual connection design borrows the multi-head trick from attention and applies it along network depth, delivering 1.58x compute speedup over standard residuals at MoE scale.

Success-Guided Sampling fixes a wasted-batch problem in massively parallel RL, letting PPO scale past a million simulated robots and transfer to real hardware.

A new paper fits a scaling law to robotic world models, trains an 8B JEPA on 15,000 hours of robot video, and plans on real hardware.

Zyphra researchers cut MoE all-to-all communication by up to 2.63x on AMD MI300X GPUs by exploiting predictable token routing patterns, with no changes to the model.

A new study finds no principled hyperparameter scaling rule consistently preserves optimizer performance across batch sizes, undermining single-batch benchmarks.

Stanford researchers show that steering RL exploration along human hand coordination patterns, rather than wiggling joints independently, dramatically speeds up dexterous skill learning.
No stories match these filters.