AI News

Qwen3.8 27B is now available for both fine-tuning and Dedicated Model Inference

Fine-tune it on your own data, then deploy that model on dedicated infrastructure for production, without stitching together separate training and serving stacks.

https://t.co/MFkqLlPipR

Safety alignment in large language models (LLMs) remains brittle against a growing spectrum of attacks. Jailbreak attacks bypass safety mechanisms through crafted prompts, while neuron-level attacks directly prune safety-critical neurons post-deployment. Both exploit a common weakness: safety-relevant information concentrates in a sparse neuron subset. We present NeuronGuard, a fine-tuning-stage defense that simultaneously hardens LLMs against both attack classes by redistributing safety signals across a broader set of neurons. NeuronGuard dynamically identifies safety-critical neurons via periodically refreshed per-layer linear classifiers, forces refusal behavior under deliberate neuron ablation, and applies KL-divergence regularization for distributional consistency. A randomized gradient projection strategy preserves downstream task utility by resolving conflicts between the defense and task objectives. We provide a formal guarantee that NeuronGuard strictly reduces the attack success rate (ASR) upper bound, and experiments across three LLMs, six state-of-the-art attack strategies, and multimodal settings confirm near-zero ASR while maintaining task accuracy, including against white-box adaptive adversaries.

Join the discussion on this paper page

The funding, Gatik's largest so far, was led by Qatar Investment Authority and Koch Disruptive Technologies.

AMD showed off its MI400 architecture at Hot Chips 2026 going into how they made a giant GPU for its Helios racks

We all knew it was coming. The expected valuation may surprise, though.

Master the entire Claude stack—from basic prompting and AI fluency to building complex multi-agent workflows. Claude Academy brings self-paced courses, developer tutorials, and official certification paths into a single interactive learning environment. Start learning at academy.claude.com!

AI workforce for solo founders and small teams. Any MCP or skill installs like an app, agents build the tools that don't exist yet, deployed and managed by the platform, shareable to the whole team, and any agent action that you like turns into a reusable workflow so agents spend less time thinking and more time doing.

China's Galbot humanoid robots complete 100+ autonomous tennis rallies against humans, showcasing real-time perception and control.

Xiaomi Xring O3 becomes the first mobile chip to support LPDDR6 memory and the first smartphone SoC to score 5.22 million points on AnTuTu, powered by an all-big-core CPU design that scraps efficiency cores for peak performance; the Xring O100 companion chip targets 1.22 Tbps near-memory AI

Large-scale rollouts have become a core component of modern LLM systems, spanning reinforcement learning (RL) post-training, on-policy distillation (OPD), and sampling-heavy evaluation pipelines. Unlike online serving, which is typically optimized for request-level latency and throughput, a small number of long-tail generations can dominate the end-to-end makespan of an entire rollout step. In practice, rollout requests are often routed uniformly across replicas, which can place extremely long generations inside high-concurrency decoding batches.
To address this, we present TailSieve, a partial-rollout-guided framework that jointly controls tail routing and replica allocation for LLM rollouts. In an idealized setting with known completion lengths, we show that makespan-optimal routing in the long-tail regime combines tail isolation with load balancing, and that a simple top-k policy closely approximates this offline optimum. Leveraging the observation that long-tail prompts tend to remain long-tailed across policy updates, TailSieve uses partial rollouts as a training-free signal for identifying candidate tail groups. A hierarchical controller then jointly adapts the number of isolated groups and the replica split between the tail and bulk pools using collected response-work history and a measured concurrency-throughput model. TailSieve achieves up to 1.67x routing-only speedup over uniform group routing. The resulting low-concurrency tail pool further enables route-specialized speculative decoding with MTP or DFlash, achieving up to 2.59x speedup over uniform routing. Selected prompts are regenerated under the current policy, preserving on-policy generation and avoiding additional routing-induced length bias in steady state.

Training large language models is costly. How low a loss the same compute can ultimately reach depends on how each step's gradient is converted into a weight update; the rule that performs this conversion is the optimizer. From SGD and AdamW to the recent Muon, effective update rules have mostly been shaped by engineering intuition and then selected on benchmarks. Muon semi-orthogonalizes the momentum matrix before applying the update and has kept breaking records on public training benchmarks; yet why the semi-orthogonalized direction works, and over how long a history the momentum should average, are two questions at present answered mainly by experience. Here we treat the weight matrix during training as a responsive medium with memory and build a physical model for it, in which both questions find answers: the semi-orthogonalized direction is the maximally dissipative response under an output-side safety budget, which explains why it works; momentum is the internal stress accumulated by the medium; how long it should average is set by the relaxation of this stress, and a real medium relaxes on more than one timescale, the simplest form being one fast and one slow. On this basis we propose the Bi-Maxwell optimizer. The framework further yields a testable consequence: gradient directions change fast early in training and more slowly later, so the optimal memory length should grow with training stage; step-by-step measurements of a proxy for it by a read-only probe across 8 independent training trajectories are consistent with this consequence. Replacing the memory kernel alone, from a single timescale to two, brings training to the target loss in noticeably fewer steps on a public large-language-model optimizer benchmark.

Reinforcement learning with verifiable rewards (RLVR) enables language models to learn multi-turn interaction with external tools, yet its sparse outcome rewards provide no signal for identifying which intermediate decisions are responsible for success. Branch sampling induces local comparisons among alternative continuations, but existing methods tend to conflate two distinct problems: allocating a fixed rollout budget and translating branch outcomes into token-level credit. We introduce Contrastive Branch Policy Optimization (CBPO), which disentangles these two problems and assigns a dedicated mechanism to each. Generation entropy screens candidate branch positions across the entire response, while path-level and node-level decay distribute a fixed budget across trajectories and positions to prevent exploration from collapsing onto a few paths or adjacent tokens. A parent trajectory together with the branches that share an identical token prefix forms an exact-prefix group, and the reward variation within this controlled group defines the Contrastive Branch Value (CBV), an outcome-based estimate of local decision sensitivity that rescales continuation advantages without altering their sign. When multiple nodes are selected along the same trajectory, CBPO partitions it into non-overlapping credit segments, thereby avoiding duplicated gradients on shared tokens. Requiring only outcome rewards and no process-level annotation, CBPO provides a practical solution for fine-grained credit assignment in tool-integrated agent training. Extensive experiments on ten benchmarks, including five for mathematical reasoning and five for knowledge-intensive search, show that CBPO consistently outperforms state-of-the-art policy-optimization and branch-based methods, attaining the highest macro-average accuracy in both domains and across two model scales.

Join the discussion on this paper page

Join the discussion on this paper page

The $200 million extension comes just months after the physical AI startup reached a $2 billion valuation.

Since announcing Jalapeño, our first custom inference chip, we’ve been testing it and the system around it.

The results show a major advance: more intelligence from every watt and faster responses, delivering both higher throughput and lower latency in one architecture without

General Intuition, the startup building a foundation model that trains generalized AI agents how to move through space and time, is in talks to raise at a $6 billion pre-money valuation from new investors including Valor Ventures, Point72 Ventures, and Seven Seven Six.

Autonomous Offensive Security, Bug Bounty & Red Teaming Agent Framework powered by Hermes Agent, specialized reasoning skills, and multi-model LLM orchestration. - Zyrexnn/Cybermes

Slack shipped Slack Code on August 20, 2026 with Claude, Devin, GitHub Copilot and Vercel working in shared, auto-archiving code channels.

One new controller and two years of existing products, repackaged for a market where memory now eats 30% of hyperscaler budgets.

At Hot Chips 2026, IBM showed a crazy new IBM Z CPU that will support IBM Z and Arm instruction sets natively.

At Hot Chips 2026, NVIDIA went into detail on its upcoming Vera server CPU, and the Olympus CPU architecture. With 88 cores per chip, NVIDIA has ambitious aims for its in-house designed CPU architecture

Deploying LLMs for enterprise Text-to-SQL is bottlenecked less by the model than by what context reaches it: business logic spans thousands of tables, and no model can ingest a full catalog at once. We argue that the most effective place to intervene is therefore the \emph{knowledge-base context} the model consumes, and that this context should be \emph{constructed} from historical usage rather than tuned for as a fixed input. Using a query-DAG decomposition--the same family of intermediates that enterprise benchmarks like BEAVER annotate, here recovered from production SQL--we compare the value of oracle query graphs versus retrieved knowledge-base context. In this ablation, retrieved knowledge-base context provides the largest marginal improvement when added to the full oracle graph. Building on this, we optimize a distillation procedure that turns historical query profiles into reusable SQL reference cards. On a benchmark of 5176 production queries from a major online retailer, optimizing these context artifacts yields larger gains (${\sim}12$--$25\%$ AST similarity) than optimizing the retrieval harness (${\sim}3$--$12\%$). On the public BEAVER benchmark, which lacks the production-usage signals available in our internal setting, the picture is more mixed: table cards alone perform about the same as raw historical SQL. The best optimized variant retrieves both cards and raw SQL, scoring $9.00\%$ versus $6.33\%$ (p-value $0.12$) for the comparable baseline on a held-out $N{=}300$ subset, using retrieved context and harness changes but no agentic loop.