Research

arXiv paper on a cache-merging method for multi-agent latent reasoning, framing KV-cache composition as a convergent replicated state with deterministic merging.

Anthropic research report on how people use Claude Code in practice, based on a large privacy-preserving analysis of session data. It covers task types, division of labor between user and model, and how domain expertise affects outcomes.

LEVI is a harness-first evolutionary framework for code and prompt optimization. It focuses on reducing LLM cost with diversity-preserving search, role-aware model routing, and a proxy benchmark, and presents comparative results against several existing systems.

arXiv paper on distilling multi-agent debate into a single LLM with a two-stage fine-tuning pipeline. The abstract reports lower token use, comparable or better benchmark performance, and an analysis of agent-specific activation subspaces, with code linked from the page.

A GitHub research project that measures how gpt-4.1 responds when asked to pick a random number between 1 and 100, using 10,000 API calls and comparing the results to a uniform baseline.

An open-source experiment that adds a small zero-initialized overlay layer to a frozen GPT-2 so its behavior can be adjusted at inference time without retraining the base model.

A research hub for the paper and survey "Toward Efficient Agents: A Survey of Memory, Tool learning, and Planning," with organized paper navigation across memory, tool learning, and planning topics.

Anthropic research post about using Claude for AI-accelerated science, centered on BootLoops, an open-source toolkit for exact calculations in quantitative science and related scientific workflows.

MIT’s report on how generative AI is changing teaching, learning, assessment, and research training, with guiding principles and recommendations for instructors and the administration.

Ceptile is an AI-powered web search and conversational research product with live web results, verifiable source citations, and a demo video on the homepage.

arXiv paper on how ideas and goals can spread through multi-agent LLM systems, including experiments on propagation, susceptibility factors, and mitigation.

A Zenodo preprint reporting a 12,160-trial black-box evaluation of GPT-5.4 outputs under two prompt conditions, with detailed token-ceiling sweeps, control/ablation trials, and hash-chained verification.

arXiv paper on self-speculating LLM agents that predict their next tool call to hide tool latency and improve next-call Hit@1 while preserving task success.

An arXiv paper on reducing LLM inference cost for web automation by compiling browser tasks into a deterministic JSON workflow and executing them without repeated model calls.

A research-oriented optimizer project claiming to cut AdamW optimizer-state memory for LLM training by 50% through factorized variance and frequency-domain compression. The repo includes a Python POC, benchmark claims, and a stated drop-in usage pattern for PyTorch training.

A GitHub-hosted research note about an agent-driven engineering workflow, comparing merged PR volume across Kungfu, Google AX, and OpenAI references.

arXiv paper proposing MANTA, a framework that lets multi-agent communication structures adapt at inference time based on collaboration traces and task needs.

A research page and preprint about using code as the runtime layer for agent systems, with a taxonomy of harness interfaces, harness mechanisms, and scaling patterns for multi-agent workflows.