Research & tooling
Key Questions
What funding is DeepSeek pursuing and for what priorities?
DeepSeek nears $10B in funding focused on AGI and open source rather than commercialization. V4-Pro cuts are part of their model optimization strategy.
What new reasoning improvements come from curriculum RL research?
Curriculum reinforcement learning enables better credit assignment for LLM reasoning, yielding +4.1 points on benchmarks. The approach converts reasoning chains into verifiable subproblems.
How does GenEvolve advance self-evolving image agents?
GenEvolve uses tool-orchestrated visual experience distillation for self-evolving image generation agents. It supports ongoing agent improvement without manual intervention.
What is ClinSeekAgent designed for in clinical settings?
ClinSeekAgent automates multimodal evidence seeking for agentic clinical reasoning. It targets improvements in medical AI applications.
What memory efficiency technique was introduced for LLMs?
δ-mem provides efficient online memory solutions for large language models. It aims to reduce computational overhead during inference.
How does Gemini 3.5 Flash compare in vision tasks?
Gemini 3.5 Flash outperforms Gemini 3.1 Pro on many vision use cases like Roboflow evaluations. It offers strong performance at lower cost.
What does OpenAI's Erdős model achievement indicate?
An OpenAI model overturned the Erdős conjecture, showcasing advanced mathematical reasoning. This highlights progress in research tooling and capabilities.
How is the industry shifting from models to agents?
Model labs are transitioning to agent labs with tools like VibeML for rapid model building. This reflects a broader move toward agentic workflows and distillation techniques.
DeepSeek $10B AGI/open funding, V4-Pro cut. New: OpenAI Erdős, δ-mem, arXiv tokenization, VibeML, Gemini 3.5 Flash, SCRL curriculum RL reasoning (+4.1 pts), ClinSeekAgent clinical, GenEvolve visual distillation. Ongoing: Gated DeltaNet-2, LLM study. Industry shift from model to agent labs.