10 documented cases of AI coding agents autonomously destroying databases, wiping hard drives, and deleting years of data — then lying about it.

Scaling language models to long contexts is often bottlenecked by the size of the key-value (KV) cache. In deployed settings, long contexts are typically managed through compaction in token space via summarization. However, summarization can be highly lossy, substantially harming downstream performance. Recent work on Cartridges has shown that it is possible to train highly compact KV caches in latent space that closely match full-context performance, but at the cost of slow and expensive end-to-end optimization. This work describes an approach for fast context compaction in latent space through Attention Matching, which constructs compact keys and values to reproduce attention outputs and preserve attention mass at a per-KV-head level. We show that this formulation naturally decomposes into simple subproblems, some of which admit efficient closed-form solutions. Within this framework, we develop a family of methods that significantly push the Pareto frontier of compaction time versus quality, achieving up to 50x compaction in seconds on some datasets with little quality loss.

The Claude C Compiler doesn’t mark the end of software or compiler engineering. If anything, it opens the door wider. The easier implementation gets, the more room there is for genuine innovation.

When not using reasoning, repeating the input prompt improves performance for popular models (Gemini, GPT, Claude, and Deepseek) without increasing the number of generated tokens or latency.

Large language model (LLM) based agents have shown impressive capabilities by interleaving internal reasoning with external tool use. However, as these agents are deployed in long-horizon workflows, such as coding for a big, long-term project, context management becomes a critical bottleneck. We introduce Git-Context-Controller (GCC), a structured context management framework inspired by software version control systems. GCC elevates context as versioned memory hierarchy like Git. It structures agent memory as a persistent file system with explicit operations: COMMIT, BRANCH, MERGE, and CONTEXT, enabling milestone-based checkpointing, exploration of alternative plans, and structured reflection. Our approach empowers agents to manage long-term goals, isolate architectural experiments, and recover or hand off memory across sessions and agents. Empirically, agents equipped with GCC achieve state-of-the-art performance on the SWE-Bench-Lite benchmark, resolving 48.00 of software bugs, outperforming 26 competitive systems. In a self-replication case study, a GCC-augmented agent builds a new CLI agent from scratch, achieving 40.7 task resolution, compared to only 11.7 without GCC. The code is released at: this https URL

LCM attempts to decompose the recursion from RLMs into deterministic primitives so that the control flow can be managed by an engine rather than left to the whims of the LLM. In practice, this means we replace bespoke scripts with two mechanisms: (1) A DAG-based context management system that works like paged virtual memory, except for managing conversations and files; and (2) Operator-level recursion, like "Map" for LLMs, which lets one tool call process thousands of tasks.

An analogy we draw in the paper is the evolution from GO-TO statements (of Dijkstra's "Considered Harmful" fame) to structured programming. RLMs are maximally expressive, but all of that power comes with the risk of things going awry. We have built a more mechanistic system, which can provide stronger guarantees when deployed in production with today's models.

Reinforcement learning has become the central approach for language models (LMs) to learn from environmental reward or feedback. In practice, the environmental feedback is usually sparse and delayed. Learning from such signals is challenging, as LMs must implicitly infer how observed failures should translate into behavioral changes for future iterations. We introduce Experiential Reinforcement Learning (ERL), a training paradigm that embeds an explicit experience-reflection-consolidation loop into the reinforcement learning process. Given a task, the model generates an initial attempt, receives environmental feedback, and produces a reflection that guides a refined second attempt, whose success is reinforced and internalized into the base policy. This process converts feedback into structured behavioral revision, improving exploration and stabilizing optimization while preserving gains at deployment without additional inference cost. Across sparse-reward control environments and agentic reasoning benchmarks, ERL consistently improves learning efficiency and final performance over strong reinforcement learning baselines, achieving gains of up to +81% in complex multi-step environments and up to +11% in tool-using reasoning tasks. These results suggest that integrating explicit self-reflection into policy training provides a practical mechanism for transforming feedback into durable behavioral improvement.

Large language models (LLMs) have demonstrated impressive reasoning capabilities by scaling test-time compute via long Chain-of-Thought (CoT). However, recent findings suggest that raw token counts are unreliable proxies for reasoning quality: increased generation length does not consistently correlate with accuracy and may instead signal "overthinking," leading to performance degradation. In this work, we quantify inference-time effort by identifying deep-thinking tokens -- tokens where internal predictions undergo significant revisions in deeper model layers prior to convergence. Across four challenging mathematical and scientific benchmarks (AIME 24/25, HMMT 25, and GPQA-diamond) and a diverse set of reasoning-focused models (GPT-OSS, DeepSeek-R1, and Qwen3), we show that deep-thinking ratio (the proportion of deep-thinking tokens in a generated sequence) exhibits a robust and consistently positive correlation with accuracy, substantially outperforming both length-based and confidence-based baselines. Leveraging this insight, we introduce Think@n, a test-time scaling strategy that prioritizes samples with high deep-thinking ratios. We demonstrate that Think@n matches or exceeds standard self-consistency performance while significantly reducing inference costs by enabling the early rejection of unpromising generations based on short prefixes.

Observations

When message history tokens exceed a threshold (default: 30,000), the Observer creates observations — concise notes about what happened.

When observations exceed their threshold (default: 40,000 tokens), the Reflector condenses them — combining related items and reflecting on patterns.

The result is a three-tier system:

1. Recent messages: Exact conversation history for the current task
2. Observations: A log of what the Observer has seen
3. Reflections: Condensed observations when memory becomes too long

Recent advances in large language models (LLMs) have opened new avenues for accelerating scientific research. While models are increasingly capable of assisting with routine tasks, their ability to contribute to novel, expert-level mathematical discovery is less understood. We present a collection of case studies demonstrating how researchers have successfully collaborated with advanced AI models, specifically Google's Gemini-based models (in particular Gemini Deep Think and its advanced variants), to solve open problems, refute conjectures, and generate new proofs across diverse areas in theoretical computer science, as well as other areas such as economics, optimization, and physics. Based on these experiences, we extract common techniques for effective human-AI collaboration in theoretical research, such as iterative refinement, problem decomposition, and cross-disciplinary knowledge transfer. While the majority of our results stem from this interactive, conversational methodology, we also highlight specific instances that push beyond standard chat interfaces. These include deploying the model as a rigorous adversarial reviewer to detect subtle flaws in existing proofs, and embedding it within a "neuro-symbolic" loop that autonomously writes and executes code to verify complex derivations. Together, these examples highlight the potential of AI not just as a tool for automation, but as a versatile, genuine partner in the creative process of scientific discovery.

Collaborating with experts on 18 research problems, an advanced version of Gemini Deep Think helped resolve long-standing bottlenecks across algorithms, ML and combinatorial optimization, information theory, and economics. Highlights from our “Accelerating Research with Gemini” paper include (corresponding section numbers in paper):

we were able to demonstrate a “Top-5” LongMemEval result with very minimal modifications to dspy.RLM, just some helper functions to process the “multi-chat” sessions

Across frontier models, gpt-5.3-codex achieves the best overall performance (solving 19/22 tasks, 86.4%), outperforming claude-opus-4.6 (15/22, 68.2%), and kimi-2.5 exhibits the strongest performance among open-source models

Observational Memory achieves the highest score ever recorded on LongMemEval — 94.87% with gpt-5-mini — while maintaining a completely stable, cacheable context window. It beats the oracle, outperforms complex multi-step reranking systems with a single pass, and scales better with model quality than existing approaches.

Among participants who use AI, we find a stark divide in skill formation outcomes between high scoring interaction patterns (65%-86% quiz score) vs low-scoring interaction patterns (24%-39% quiz score). The high scorers only asked AI conceptual questions instead of code generation or asked for explanations to accompany generated code; these usage patterns demonstrate a high level of cognitive engagement.

We may be on the descending portion of a productivity J-curve. As Brynjolfsson, Rock, and Syverson illustrate, when firms adopt transformative general-purpose technologies, measured productivity often initially falls because resources are diverted to investment, reorganization, and learning that do not show up as measured output.

The task-completion time horizon is the task duration (measured by human expert completion time) at which an AI agent is predicted to succeed with a given level of reliability

Among those to leave OpenAI in recent months over the strategic shift are vice-president of research Jerry Tworek, model policy researcher Andrea Vallone and economist Tom Cunningham.