• Ideas have a "genome," but machines can't trace them back through the literature. IdeaGene treats a research idea as something that inherits mechanisms, patches known flaws, and recombines earlier work. On IG-Bench, the strongest of 14 LLM scientists got lineage reasoning fully correct only 27.3% of the time. Placing an idea in the literature and generating N ideas are not the same skill.

• Extend context with no retraining, at almost zero cost. Jet-Long uses a bifocal RoPE so short inputs fall back exactly to the base model and long inputs extrapolate cleanly. On an H100 it hits 1.39x the prefill throughput of FlashAttention-2, with generation overhead under 4% at every length.

• Proactivity is becoming its own axis for evaluating agents. UniClawBench breaks "proactive" into five measurable abilities, runs 400 bilingual tasks inside real Docker containers, and uses a three-agent loop with hidden scoring to tell whether an agent understood intent or just guessed the rubric.

• Borrow a diffusion prior to attack an unusual modality: event cameras. LongE2V doesn't build a model from scratch. It fine-tunes a pretrained video diffusion model to do event-stream reconstruction, prediction, and frame interpolation in one framework, with data efficiency as the selling point.

Also Notable

• Softmax and Four Recent Linear Attentions in One Notation, Side by Side — DeltaNet, Gated DeltaNet, KDA, and Gated DeltaNet-2 unified under a recurrent-memory framework, making the differences in memory decay and erase/write mechanisms obvious. Handy when choosing an architecture.
• Canvas360 Does In-Context Panorama Generation and Ships a 1M Paired Dataset — Covers style transfer, inpainting, outpainting, and editing; usable directly for data-starved panorama and 360 tasks.
• CausalDS Tests Causal Reasoning in Data-Science Agents — Fills the gap for a benchmark with real data analysis but a principled causal generative structure. Worth a look if you care whether an agent confuses correlation with causation.
• Treating Memory as Active Intervention, Not Passive Retrieval — In long-horizon tasks, decision-relevant state gets buried or squeezed out of the context window (the authors call it "behavioral state decay"); this paper uses a dedicated memory agent to prop up decisions.
• OpenCoF Runs Reasoning Across Time-Continuous Frames (Chain-of-Frame) — Treats video generation as a reasoning path distinct from text CoT. Skim it if you're interested in using generative models to reason.
• OPSD-V Post-Trains Few-Step Autoregressive Video Generators With On-Policy Self-Distillation — Suppresses error accumulation and dynamic decay over long rollouts while keeping the original few-step inference path.

Read the full edition →