AI Research Radar

Saturday, April 25: 1 frontier paper out of 135 read.

1 frontier21 cleared135 papers read

Frontier

Papers rated as moving the field

HWE-Bench

“We evaluate seven LLMs with four agent frameworks and find that the best agent resolves 70.7% of tasks overall, with performance exceeding 90% on smaller cores but dropping below 65% on complex SoC-level projects.”

The authors, in the abstract

Benchmarking LLM Agents on Real-World Hardware Bug Repair Tasks

BenchmarkarXiv PDF

This week in research

Apr 19–25: 21 cleared, 1 frontier

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

Models under test

Papers this week that name the model

Llama 3.1 8B Instruct1
Gemini 3.1 Pro1
Mistral Large1
Mistral Large 31

Labs publishing

Papers this week with an author at the lab

Alibaba1
Google1
IBM1
Microsoft1

Also cleared

20 papers

LASA: Language-Agnostic Semantic Alignment at the Semantic Bottleneck for LLM SafetyAlibaba
Differentially Private Model MergingGoogle
Breaking MCP with Function Hijacking Attacks: Novel Threats for Function Calling and Agentic ModelsIBM
Separable Expert Architecture: Toward Privacy-Preserving LLM Personalization via Composable Adapters and Deletable User ProxiesMicrosoft
Tool Attention Is All You Need: Dynamic Tool Gating and Lazy Schema Loading for Eliminating the MCP/Tools Tax in Scalable Agentic WorkflowsCode
VLA-Forget: Vision-Language-Action Unlearning for Embodied Foundation Models
Omission Constraints Decay While Commission Constraints Persist in Long-Context LLM Agents
Behavioral Consistency and Transparency Analysis on Large Language Model API Gateways
AI Governance under Political Turnover: The Alignment Surface of Compliance Design
Cross-Session Threats in AI Agents: Benchmark, Evaluation, and Algorithms
CAP: Controllable Alignment Prompting for Unlearning in LLMs
Do LLM Decoders Listen Fairly? Benchmarking How Language Model Priors Shape Bias in Speech Recognition

All 20 papers

Spatial Metaphors for LLM Memory: A Critical Analysis of the MemPalace Architecture
Brief chatbot interactions produce lasting changes in human moral values
Engaged AI Governance: Addressing the Last Mile Challenge Through Internal Expert Collaboration
Stealthy Backdoor Attacks against LLMs Based on Natural Style Triggers
Building a Precise Video Language with Human-AI Oversight
Who Defines "Best"? Towards Interactive, User-Defined Evaluation of LLM Leaderboards
Bounding the Black Box: A Statistical Certification Framework for AI Risk Regulation
Transient Turn Injection: Exposing Stateless Multi-Turn Vulnerabilities in Large Language Models

SIGNIFICANT ON THE BOARDS

MODEL TRACKER2,648 new benchmark results across 174 models this week DEAL FLOW$158B disclosed AI funding this week across 61 rounds POLICY125 policy events tracked across 7 days