Fellow’s new espresso machine is a rare thing in home espresso: something genuinely new. But it’s also a work in progress.

Every Claude Code user is running without LSP. That means 30-60s grep searches instead of 50ms precise answers. Here's how to enable it — setup, real debug data, and undocumented discoveries.

Formula 1's governing body the FIA said on Saturday that a change to the way the compression ratio was measured would be introduced on 1 June, with a further revision for the 2027 season.

And Trump declares a state of emergency and postpones the election. The Supreme Court issues an emergency stay, saying he can’t do that. But the court has no army, and Trump does, along with a handful of lickspittle governors who just might follow him down whatever dark path he plows.

That, not to mince words, is a coup d’état. Will he get away with it? I don’t know, but having effective control over how it is presented to viewers of CBS and CNN, and readers of the Bezos-owned Washington Post, to say nothing of the already vast pro-Trump propaganda empire of Fox News and the rest, will certainly make it easier.

That’s how fascism descends. And it’s becoming less and less hypothetical by the week.

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

“Everything that has been written about a potential War with Iran has been written incorrectly, and purposefully so,” he added. “I am the one that makes the decision, I would rather have a Deal than not but, if we don’t make a Deal, it will be a very bad day for that Country and, very sadly, its people, because they are great and wonderful, and something like this should never have happened to them.”

From rewriting Google’s search stack in the early 2000s to reviving sparse trillion-parameter models and co-designing TPUs with frontier ML research, Jeff Dean has quietly shaped nearly every layer of the modern AI stack. As Chief AI Scientist at Google and a driving force behind Gemini, Jeff has lived through multiple scaling revolutions from CPUs and sharded indices to multimodal models that reason across text, video, and code.

Jeff joins us to unpack what it really means to “own the Pareto frontier,” why distillation is the engine behind every Flash model breakthrough, how energy (in picojoules) not FLOPs is becoming the true bottleneck, what it was like leading the charge to unify all of Google’s AI teams, and why the next leap won’t come from bigger context windows alone, but from systems that give the illusion of attending to trillions of tokens.

Dario Amodei thinks we are just a few years away from “a country of geniuses in a data center”. In this episode, we discuss what to make of the scaling hypothesis in the current RL regime, how AI will diffuse throughout the economy, whether Anthropic is underinvesting in compute given their timelines, how frontier labs will ever make money, whether regulation will destroy the boons of this technology, US-China competition, and much more.

The ruling hit while Trump was in a closed-door meeting with a bipartisan group of governors. The president’s initial reaction was to label the decision a “disgrace” and vow to implement a backup plan, according to a person familiar with the matter who requested anonymity to describe the closed-door event. The White House and US Trade Representative haven’t yet responded to requests for comment. Trump has called tariffs “my favorite word” and vowed they will “make us rich as hell.”

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.

President Donald Trump accused former President Barack Obama of giving away classified information when he discussed aliens during a recent podcast appearance.
“He gave classified information, he’s not supposed to be doing that,” Trump told reporters Thursday aboard Air Force One.
Pressed on if that meant aliens were real, Trump said he did not know “if they’re real or not.”
“I can tell you he gave classified information, he’s not supposed to be doing that,” the president said. Trump went on to suggest he could get the former president “out of trouble” by declassifying the related information.
Obama was asked about extraterrestrial life earlier this month during an interview with liberal commentator Brian Tyler Cohen, and responded, “they’re real.”

Do gifted individuals see the world differently? Research tracking adults over 35 years finds their political orientations are remarkably average, with one specific exception regarding male conservatism.

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.

"If this is correct, to the extent of my knowledge, it would mark the first time humanity has 'seen' dark matter. And it turns out that dark matter is a new particle not included in the current standard model of particle physics. This signifies a major development in astronomy and physics," said Totani.

The strong zero-shot and long-context capabilities of recent Large Language Models (LLMs) have paved the way for highly effective re-ranking systems. Attention-based re-rankers leverage attention weights from transformer heads to produce relevance scores, but not all heads are created equally: many contribute noise and redundancy, thus limiting performance. To address this, we introduce CoRe heads, a small set of retrieval heads identified via a contrastive scoring metric that explicitly rewards high attention heads that correlate with relevant documents, while downplaying nodes with higher attention that correlate with irrelevant documents. This relative ranking criterion isolates the most discriminative heads for re-ranking and yields a state-of-the-art list-wise re-ranker. Extensive experiments with three LLMs show that aggregated signals from CoRe heads, constituting less than 1% of all heads, substantially improve re-ranking accuracy over strong baselines. We further find that CoRe heads are concentrated in middle layers, and pruning the computation of final 50% of model layers preserves accuracy while significantly reducing inference time and memory usage.

Information retrieval (IR) systems have played a vital role in modern digital life and have cemented their continued usefulness in this new era of generative AI via retrieval-augmented generation. With strong language processing capabilities and remarkable versatility, large language models (LLMs) have become popular choices for zero-shot re-ranking in IR systems. So far, LLM-based re-ranking methods rely on strong generative capabilities, which restricts their use to either specialized or powerful proprietary models. Given these restrictions, we ask: is autoregressive generation necessary and optimal for LLMs to perform re-ranking? We hypothesize that there are abundant signals relevant to re-ranking within LLMs that might not be used to their full potential via generation. To more directly leverage such signals, we propose in-context re-ranking (ICR), a novel method that leverages the change in attention pattern caused by the search query for accurate and efficient re-ranking. To mitigate the intrinsic biases in LLMs, we propose a calibration method using a content-free query. Due to the absence of generation, ICR only requires two (O(1)) forward passes to re-rank N documents, making it substantially more efficient than generative re-ranking methods that require at least O(N) forward passes. Our novel design also enables ICR to be applied to any LLM without specialized training while guaranteeing a well-formed ranking. Extensive experiments with two popular open-weight LLMs on standard single-hop and multi-hop information retrieval benchmarks show that ICR outperforms RankGPT while cutting the latency by more than 60% in practice. Through detailed analyses, we show that ICR's performance is specially strong on tasks that require more complex re-ranking signals. Our findings call for further exploration on novel ways of utilizing open-weight LLMs beyond text generation.

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

Humans always remain in the loop, but work at a different layer of abstraction than we used to. We prioritize work, translate user feedback into acceptance criteria, and validate outcomes. When the agent struggles, we treat it as a signal: identify what is missing—tools, guardrails, documentation—and feed it back into the repository, always by having Codex itself write the fix.

Our most difficult challenges now center on designing environments, feedback loops, and control systems that help agents accomplish our goal: build and maintain complex, reliable software at scale.

The engineering team used Codex to optimize and adapt the harness for GPT‑5.3-Codex. When we started seeing strange edge cases impacting users, team members used Codex to identify context rendering bugs, and root cause low cache hit rates. GPT‑5.3-Codex is continuing to help the team throughout the launch by dynamically scaling GPU clusters to adjust to traffic surges and keeping latency stable.

A 61-year-old Tennessee man is finally free after spending a shocking 37 days in jail — all for posting a meme.

Of those, GVA said there were five confirmed transgender shooters, or fewer than a tenth of one per cent. (There have also been four cases of mass shootings by females in the U.S. since 1982.)

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

The firm is also whitelisting a handful of market makers, including longtime crypto liquidity provider Wintermute, to facilitate trading. Meanwhile, access to BUIDL is restricted to qualified purchasers, a legal designation for those with assets of $5 million or more.

For years, Trump has claimed he had “no idea” about Epstein’s abuse of underage girls. Yet records show that in 2006, he privately told Palm Beach police that “everyone” knew about Epstein’s activities and described Ghislaine Maxwell as evil.
Trump’s call to Palm Beach police chief

According to an FBI interview conducted in October 2019 with former Palm Beach Police Chief Michael Reiter, Trump personally called him in July 2006, just as Epstein’s criminal sex charges became public. Reiter told agents that Trump said, “Thank goodness you’re stopping him, everyone has known he’s been doing this.”

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.

"I mean, there's tons of redacted stuff. ... And [Trump's] name, I think I put his name, and it appears more than a million times. So it's all over the place."

The bottom line: "To me, this whole rollout of saying that members can come from nine to five to sit at those four computers, is just part of the coverup," Raskin asserted.

The 3 million documents that the administration has not publicly released "are the ones I'd like to see," he said.
"The administration says that these are duplicative. Well go ahead and release them then! If they're duplicative, what's the problem? We'll be the judge of that." 

"Epstein's lawyers synopsized and quoted Trump as saying that Jeffrey Epstein was not a member of his club at Mar-a-Lago, but he was a guest at Mar-a-Lago, and he had never been asked to leave," Raskin said. "That was redacted for some indeterminate, inscrutable reason."

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 develop a model of political cycles driven by time-varying risk aversion. Agents choose to work in the public or private sector and to vote Democratic or Republican. In equilibrium, when risk aversion is high, agents elect Democrats—the party promising more redistribution. The model predicts higher average stock market returns under Democratic presidencies, explaining the well-known “presidential puzzle.” The model can also explain why economic growth has been faster under Democratic presidencies. In the data, Democratic voters are more risk averse, and risk aversion declines during Democratic presidencies. Public workers vote Democratic, while entrepreneurs vote Republican, as the model predicts.

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

it will automatically set all users’ accounts to a “teen-appropriate” experience unless they demonstrate that they’re adults

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.

MaxRL is a framework that turns more compute into increasingly better approximations of the maximum likelihood objective in sampling-based tasks.