AI Research Radar

Thursday, May 7: 1 frontier paper out of 233 read.

1 frontier16 cleared233 papers read

Frontier

Papers rated as moving the field

Agent Island

“As an example, we investigate same-provider preference in final-round votes and find that models are 8.3 p.p. more likely to support a same-provider finalist than finalists from other providers.”

The authors, in the abstract

A Saturation- and Contamination-Resistant Benchmark from Multiagent Games

BenchmarkarXiv PDF

This week in research

May 1–7: 116 cleared, 12 frontier

The week’s papers have not been sorted into fields yet.

Models under test

Papers this week that name the model

GPT-4o mini5
GPT-5.54
GPT-54
GPT-4o3
GPT-5.3-Codex2
GPT-5.42
Qwen3 8B2
GPT-5.22

Labs publishing

Papers this week with an author at the lab

AllenAI2
Microsoft2
Tencent2
Alibaba1
Anthropic1
ByteDance1
IBM1
Meta1

Also cleared

15 papers

Valley3: Scaling Omni Foundation Models for E-commerceByteDance
SoK: Robustness in Large Language Models against Jailbreak AttacksMicrosoft
Do Agents Dream of Root Shells? Partial-Credit Evaluation of LLM Agents in Capture the Flag Challenges
GPT-Image-2 in the Wild: A Twitter Dataset of Self-Reported AI-Generated Images from the First Week of Deployment
LCM: Lossless Context Management
Confronting Label Indeterminacy in Automated Bail Decisions
Are Multimodal LLMs Ready for Clinical Dermatology? A Real-World Evaluation in Dermatology
TSCG: Deterministic Tool-Schema Compilation for Agentic LLM Deployments
Frontier Lag: A Bibliometric Audit of Capability Misrepresentation in Academic AI Evaluation
RLearner-LLM: Balancing Logical Grounding and Fluency in Large Language Models via Hybrid Direct Preference Optimization
From Beats to Breaches:How Offensive AI Infers Sensitive User Information from Playlists
AgentTrust: Runtime Safety Evaluation and Interception for AI Agent Tool Use

All 15 papers

Curated AI beats frontier LLMs at pharma asset discovery
Misaligned by Reward: Socially Undesirable Preferences in LLMs
Design Conductor 2.0: An agent builds a TurboQuant inference accelerator in 80 hours

SIGNIFICANT ON THE BOARDS

MODEL TRACKER2,648 new benchmark results across 174 models this week DEAL FLOW$158.1B disclosed AI funding this week across 62 rounds POLICY126 policy events tracked across 7 days