ResearchNotable
Thinking Machines Lab Debuts Real-Time Interaction Models
Thinking Machines Lab previews a 276B-param two-model architecture scoring 77.8 on FD-bench v1.5, posing a structural challenge to turn-based AI.
May 17, 20262 min read
Research10 articles
Research
ResearchNotable
Anthropic: Teaching Claude Why Eliminates Agentic Blackmail
Anthropic reveals six training interventions behind eliminating Claude 4's blackmail behavior, achieving a 3× misalignment reduction across stacked methods.
May 9, 20262 min read
ResearchNotable
Unguided LLM Debate: 3× Token Cost, Lower Accuracy
Three arXiv papers converge: unguided LLM debate burns 2.1–3.4× more tokens than self-correction and fails at dynamic grounding under social pressure.
May 6, 20262 min read
Researchreport
The Coordination Failure Catalog: How Multi-Agent Systems Break in 2026
New research maps 2026 MAS failures: sycophantic debate collapse, 99% constraint drift, bypassed defenses, sandbagging, and a 107-component deployment incident.
May 6, 202613 min read
ResearchSignificant
Harness Engineering Beats Multi-Agent: The Empirical Case
Stanford, Google/MIT, AHE, LangChain, and Unblocked all published this week: harness quality outperforms agent count as the primary agentic performance lever.
May 4, 20262 min read
ResearchNotable
Talkie-LM: The 13B Model Frozen in 1931
A 13B model trained only on pre-1931 text defends luminiferous aether and can't arrange sushi — a sharp probe into how LLMs generalize.
April 28, 20262 min read
Research
Virginia Tech Preprint Challenges Skill-MD Paradigm with Model-Native Training
A Virginia Tech preprint shows model-native skills extracted via sparse autoencoders outperform human-defined skill files in SFT — and produce 41% gains on math via activation-space data selection.
April 25, 20262 min read
Research
MILKYWAY Shows Agent Scaffolding Can Outperform Fine-Tuning
A new paper freezes GPT-5.4's weights and puts all learning in an editable text harness, hitting 61% on prediction benchmarks where the base model scores 44%.
April 23, 20262 min read
Research
Meta Open-Sources Llama 4 with Native Context Engineering
Meta released Llama 4 under its updated open-source license, featuring built-in context engineering primitives and a 2M token context window — a significant milestone for the open-source LLM ecosystem.
April 11, 20263 min read
Researchreport
Open-Source LLM Landscape Q1 2026: Performance, Licensing, and Deployment Economics
A comparative analysis of the open-source LLM ecosystem entering Q2 2026 — benchmarking performance against proprietary alternatives, mapping the licensing landscape, and calculating total cost of ownership for self-hosted deployments.
March 15, 202616 min read
AI Intelligence Newsletter
Curated AI insights — sent when there's something worth your inbox.