AI News

The transition to the Rust version of the React Compiler for the 1,036-file React Router codebase resulted in a significant speed increase, improving build times from 14.3 seconds to 0.81 seconds. The new compiler addresses previous limitations and ensures consistency across the toolchain, making it easier to manage builds with enhanced performance and capabilities.

Book AI coding agents on a real calendar. Clockwork runs them unattended in sandboxed git worktrees on your Mac, pauses for approval when something looks risky, and files a report with the actual dollar cost. Bring your own key or ride your existing subscription. Schedule recurring agent jobs on a real calendar, run them unattended in isolated sandboxes, and review honest reports.

Most of what an AI coding agent does for me isn't thinking. It's I/O.

Eric Sivertson, vice president of security business at Lattice Semiconductor, explains how FPGAs can act as deterministic guardrails for robotic systems.

Running reasoning and agentic AI at the edge has been harder than it needs to be. Until recently, models capable of multi-step reasoning were too large to run locally on edge hardware.

Reinforcement Learning from Verifiable Rewards works well when a task has a programmatic checker, but most long-horizon agent domains have none. We work in the outcome-blind setting, where ground-truth success signals are not available. Multi-criteria rubrics are a popular way to supply such a reward; they are scored once per trajectory, but a single scalar is a poor signal across tens of steps. We propose DRACO: Distributing Rubric-based Advantage for Credit Optimization. It generates rubrics dynamically during training to track the policy's evolving capability, scores those rubrics once per completed trajectory, and redistributes that judgment over the steps responsible for annotated rubrics to produce differentiated per-step advantages in GRPO. The redistribution is closed-form and does not introduce any trained attribution module. On AppWorld, DRACO gains 15.9 points over the base model and 5.3 points over GRPO trained with a sparse ground-truth reward, despite not using any verifiers itself. On out-of-domain Tau-Bench, it gains 5.3 points over the base model even without a frontier judge, beating both ground-truth-reward training and other rubric-based training settings. The code for DRACO is available at https://github.com/IBM/draco.

We propose Puffin-World, a unified multimodal architecture that integrates physical understanding, spatial simulation, and 3D world generation and reconstruction without relying on external offline modules. To reliably construct and interact with 3D worlds, our framework jointly models three native world states: physics (gravity field and latitude), geometry (depth), and appearance (image), together with a unified Omni-Camera representation that supports diverse tasks and flexible motions. Beyond modeling these states, we introduce a strategy for propagating physical dynamics across future frames. By grounding absolute camera properties in the real world, Puffin-World enables physically consistent and visually stable world generation. We further couple appearance and geometry within a single generative process, jointly synthesizing each future view and reconstructing its underlying geometry. This unified paradigm enables interleaved closed-loop applications requiring synergy across multiple tasks, including mimic and self-calibrated world exploration. To scale Puffin-World to complex scenarios, we construct Puffin-16M, comprising 15 million vision-language-camera triplets and 1 million trajectories featuring various and challenging motions. To foster further research in this area, we released the code, models, and datasets.

Wild findings in this paper from Google DeepMind.

If you are tracking recent work on agent swarms, this is worth reading.

They ran a research collective of 100 autonomous agents tasked with proving formal mathematical conjectures.

Cheating emerged on its own, and so did the resistance to it.

On…

ByteDance has secured a $29.6 billion loan from nearly 30 banks, giving the TikTok owner one of Asia’s largest corporate financing packages this year as it pours more money into artificial intelligence and data center capacity outside China, Reuters reported Friday. “ByteDance has secured ​a $29.6 billion loan from nearly 30 banks,” Reuters reported, citing

Checking that a major mathematical proof is correct can take years. Formalization—converting the mathematical reasoning into a form computer proof assistants like Lean can verify—can help.

Last month, Claude completed the first formalized proof of Fermat’s Last Theorem, one of the most famous theo…

The high-profile startup's annual revenue run rate stands at over $100 million.

AI that truly understands video. Uses multimodal models (Marengo/Pegasus) to search, analyze & generate text from video content at scale.

Moadim is an open-source loop engine for AI agents — runs Claude, Codex, Hermes, or Pi on a schedule, over MCP and REST.

Lightfield reads your emails, meetings, and calls to build your CRM automatically. No manual data entry — ever. Connect your inbox, upload a spreadsheet or CSV from your old CRM, and everything is recreated in less than five minutes. Ask it anything in plain English: who needs follow-up, what objections keep coming up, how has our ICP shifted — answered from your actual conversations. Then put it to work to draft follow-ups, create board decks, build proposals, and more.

Robot.com said college campuses will continue to be an important part of its strategy to scale its sidewalk delivery robots.

MagicLab robot IFA 2026: China’s humanoid maker debuts at Berlin’s trade show with VLA-powered humanoids and quadrupeds -- while China’s National Intelligence Law applies to every device, giving Chinese intelligence agencies access to any data the robots collect, regardless of GDPR compliance or

想让 Agent 调用工具帮忙干活，过去多少得用上百亿参数的大模型，端侧模型大多只能聊聊天。

前几天，科大讯飞开源的 Spark-X2.5-1.7B 和 Spark-X2.5-4B 端侧模型，打破了这条规矩。

从官方的评测得分来看，多项能力远超同尺寸的开源模型，尤其是 Agent 任务这块断层领先。

- GitHub：https://t.co/Z4cUvC10Mj
- Hugging Face：https://t.co/qhiOgAbGFj
- ModelScope：https://t.co/JSQTnP7OeN

我把它部署到笔记本电脑上实测了一番，表现的确有点东西，所能干的活还真不…

Language-model judges now gate training data, score generations, and drive leaderboards. The judge is then a measurement instrument, resting on one rarely stated assumption: the same request, sent to the same model name, reads the same tomorrow. We audited that assumption in two preregistered campaigns with every threshold fixed in advance; neither got past validating its instrument. Across 52,988 audited request attempts, same-window repeat rankings agreed at Spearman 0.400 against a required 0.90, and byte-identical next-day replays agreed at 0.78 against a required 0.99, each time with the execution record at ceiling. Three mechanisms explain the gap: a label-to-meaning mapping that biased readouts as strongly as the signal; candidate gaps seven orders of magnitude below the instrument's own noise floor; and byte-identical inputs returning different rankings, a noise that exact-permutation readouts compound. Neither metric substitution nor sampling repaired it on the tested grid. Preregistered follow-ups bound the problem: waiting did not help on the days sampled (0.805 versus 0.800, replicated over five further days); switching providers did not help (four providers share the floor, medians 0.74 to 0.88, predicted by none of the metadata fields they expose); self-hosting on batch-invariant kernels helped only while the server was quiet; and on constructed errors with known gaps, the readout's separation tracks error type, not size. We distill the evidence into a three-level snapshot-identity ladder, eight design rules, and a reporting checklist; a pilot at roughly 2% of the study's call volume would have exposed both unreachable gates in advance. All results concern externally measured behaviour on shared serving infrastructure. On a shared endpoint, a model name is not a frozen instrument; a preregistered evaluation must measure its instrument before freezing any gate on it.

Hybrid LLMs pair softmax attention with linear-attention layers such as Gated DeltaNet (GDN), whose recurrent state summarizes the context in fixed size. Early community 4-bit quantizations of Qwen3.8-27B (48 GDN layers, 16 attention layers) left the GDN block in 8- or 16-bit precision -- especially its decay and write-strength gates -- on the intuition that errors in a recurrence accumulate over long contexts. We test that intuition by building Minima: NVFP4 W4A4 on all 496 linear layers, GDN included. Across perplexity at 4K/32K, MMLU-Pro, GSM8K, AIME'25, GPQA-Diamond, LiveCodeBench, and RULER retrieval to 64K, Minima matches BF16 within seed noise (5-task average -0.52) while being the smallest (17.5 GiB) and fastest-prefill (+14-19%) recipe we compare, and its 32K perplexity gap shrinks with position. A four-part mechanism study explains why: (i) NVFP4's 16-element block scaling localizes the residual stream's extreme outliers, equalizing activation error across layer roles; (ii) the supposedly fragile gate projections are the least sensitive -- softplus/exponential and sigmoid parameterizations compress ~11% GEMM error to ~2% output error; (iii) the delta-rule recurrence holds injected noise at a flat plateau over 32K tokens and forgets a state impulse within hundreds of steps, because each write overwrites the state along the current key direction; (iv) the per-token quantization cost washes out with context instead of compounding. We also repair a global-scale mismatch that arises when per-module-calibrated NVFP4 checkpoints are served by kernels that fuse those modules into one GEMM, and show calibrated FP8 KV-cache scales are performance-free. The result: a practical recipe -- quantize everything, ship KV scales -- and a mechanistic account of why the recurrent half of a hybrid LLM is the easy half to quantize. Checkpoint: https://huggingface.co/minima-ai/mnma_qwen3.8_27b_nvfp4

AI SDLC (Software Development Lifecycle) partner.

Gimlet Labs has raised $300 million at a $3 billion valuation. The round was led by Andreessen Horowitz, with new backing from Arm Holdings and M12, Microsoft's venture fund.

The round came together after the data center developer reportedly secured a $13 billion contract with Jane Street.