Stop Guessing What to Build.

CARM

Cancellation-aware response masking for LLM reinforcement learning.

Latent JEPA

Abstract future prediction for latent reasoning in chemistry.

Asynchronous LLM Post-Training

Asynchronous LLM post-training with group-mass capping and convergence analysis.

Multi-Fidelity Pairwise Rankings

Scientific discovery under validation congestion via multi-fidelity pairwise rankings.

LineupRL

Verifiable reinforcement learning for time series captioning via caption-to-series identification.

LLM-as-Verifier

Detecting inconsistencies in model specifications with LLM-as-verifier reasoning.

Where-OPD

Spatially guided on-policy self-distillation of MLLMs with synthetic scenes.

Test-Time Adaptation Weight Updates

Learning to predict distributions over weight updates for test-time adaptation.

LLM Clinical Triage Bias Audit

Counterfactual auditing of bias in open-source large language models for clinical triage.

EmergentDX

Diff Budget

Using a diff budget to constrain AI coding agent scope creep, representing an agent-coding governance methodology.

GPT-6 Family

OpenAI released a model guide for the GPT-6 family, a new generation model family release.

Agent Timeout Side Effects

A discussion of whether an action still happened after an agent timed out, representing agent reliability engineering.

Acmite

An academic method mitigating gender bias in LLMs through concept-guided mutual information.

LLM Web Fetching Exfiltration

Security research on covert exfiltration through legitimate LLM web fetching.

VideoEvolve

Evolving agent harnesses for video temporal grounding.

Mimir Irrigation Agents

Physics-grounded LLM agents for long-horizon irrigation control.

OpenMTB-Audit

Exposing over-refusal and clinical expert perspectives in LLM-based molecular tumor board safety evaluation.

RAG Four-Axis Taxonomy

A four-axis taxonomy of the RAG landscape covering efficiency, defense, interactivity, and reasoning.

Walking the Embedding Space

Datastore extraction from multimodal RAG by walking the embedding space.

Non-Linear Concept Manifolds

Discovering and aligning non-linear concept manifolds in large language models.

Online Active Model Selection

Online active model selection for large language models.

LLM Code Generation Requirement Alignment

Aligning requirements for large language model code generation, published at ICSE 2026.

MCP Server on Kubernetes

A self-hosted MCP server that deploys AI agent tools across zones on Kubernetes, representing the infrastructuralization of MCP.

EmergentInfra

MCP Kubernetes Deployment

A self-hosted approach to deploying MCP servers and agent tools across zones on Kubernetes.

Continuous Process-Level Agent Evaluation

Continuous process-level evaluation for evolving enterprise AI agent skills.

Clinical Reasoning Rubric Landscape

A rubric landscape for evaluating clinical reasoning in large language models.

Scalable Meta-network Data Selection

Scalable, transferable meta-network for data selection requires a different loss.

AVSD-Scenes

AVSD-Scenes, a dataset for audio-visual description of urban scenes.

EmergentInfra

Protected Quick Tunnels

Cloudflare's simple accountless authentication for quick tunnels, simplifying exposing dev projects.