Research from ICLR 2026 From long-context limits to many-shot prompting and speculative prefill, our team presented 5 papers focused on real system challenges in AI.
What works, what doesn’t, and where things break. https://
arxiv.org/abs/2510.04618 https://
arxiv.org/abs/2602.16069
RESEARCH
-
ICLR 2026: Five Papers on Real AI System Challenges
By
–
-
SambaNova Intel Partnership Advances Coding Agents Performance
By
–
April, you were great to us! Welcome to another edition of our Lightning Digest TLDR: SambaNova + Intel DevTalks with Cline New research at ICLR 2026 Upcoming events https://
sambanova.ai/sambanova-ligh
tning-digest-premium-inference-for-coding-agents?&utm_source=x&utm_medium=organic&utm_content=newsletter-research
… -
BigLaw Bench: Evaluating AI Legal Research Capabilities
By
–
We recently partnered with @harvey on BigLaw Bench: Research, evaluating how well models handle real end-to-end case law research, identifying the right authorities, applying legal reasoning, and delivering answers you can actually use. It also shows where things still break and
-
Skill Retrieval Augmentation for Agentic AI Systems
By
–
// Skill Retrieval Augmentation for Agentic AI // Great read for AI devs. (bookmark it) It's on finding efficient ways to incorporate skills for agents. The work introduces Skill Retrieval Augmentation (SRA) and SRA-Bench: 26,262 skills, 636 gold skills, 5,400
-
Harvard Neuroscientist Launches Large Memory Model for Proactive AI Agents
By
–
Gabriel Kreiman closed his Harvard lab to build this.
— Charly Wargnier (@DataChaz) 28 avril 2026
20 years of Neuroscience Research.
160+ publications.
This Large Memory Model (LMM) enables proactive memory and allows Agents to automatically surface vital context, without explicit prompting 👀pic.twitter.com/Qpu1PmV1MN https://t.co/YQ67SaKSs4Gabriel Kreiman closed his Harvard lab to build this. 20 years of Neuroscience Research. 160+ publications. This Large Memory Model (LMM) enables proactive memory and allows Agents to automatically surface vital context, without explicit prompting
-
AI World Extends Far Beyond Large Language Models
By
–
Yes. But not in LLMs. The AI world is a lot bigger than just LLMs.
-
Agent Thoughts as Retrieval Signal for Context Preservation
By
–
Agent thoughts as retrieval signal makes so much sense in hindsight, query alone always loses the conversation context.
-
GPU-Poor Labs Drive AI Innovation Through Constraints
By
–
Constraints really are doing the inventing now, all the interesting attention work is coming from GPU-poor labs.
-
AGI Timeline Delayed: Unexplained Factors Extend Arrival
By
–
For some reason I can't explain, AGI will now take longer to arrive.
-
Anthropic’s Problematic Ransomware Approach Without Ransom
By
–
Only Anthropic does ransomware without the ransom part.