2026 in LLMs (so far)
On Friday I gave the closing keynote at the WeAreDevelopers World Congress North America in San Jose. I tied together the key trends from the past year into a chronological exploration of everything t
Knowledge catalogue
On Friday I gave the closing keynote at the WeAreDevelopers World Congress North America in San Jose. I tied together the key trends from the past year into a chronological exploration of everything t
Automated lint: 17 errors, 59669 warnings, 3 info
I set out to test portable Engrams and accidentally ended up testing compiled external memory instead. Am I onto something useful or reinventing a known idea? I've been building a small open research
arXiv:2609.30264v1 Announce Type: new Abstract: Latent world models are typically trained to predict factual transitions, whereas model predictive control (MPC) must compare alternative actions from t
arXiv:2609.29398v1 Announce Type: new Abstract: Multimodal attributed graphs connect entities, visual content, language, and observed relations. Learning one foundation across such graphs requires mor
arXiv:2609.28860v1 Announce Type: new Abstract: Textual descriptions can reduce ambiguity in medical image segmentation by specifying the finding and location to be delineated. Existing text-guided me
arXiv:2609.30134v1 Announce Type: new Abstract: Neural behavior cloning compresses demonstrations into large models, making individual actions difficult to trace and policy updates costly. Retrieval p
At Google Cloud, we are committed to delivering the best managed experience backed by open source software. Today, we’re announcing the general availability of Memorystore for Valkey 9.1, which achiev
arXiv:2609.27275v1 Announce Type: new Abstract: Most autonomous-driving datasets record only the action executed by a behavior policy and the single future that followed, providing limited supervision
arXiv:2609.27142v1 Announce Type: new Abstract: Text-to-image retrieval with frozen dual encoders degrades when the query names a small, visually subordinate object in a cluttered scene: a single glob
arXiv:2607.19114v2 Announce Type: replace Abstract: Activities in aqueous electrolyte solutions, usually described by ionic activity and osmotic coefficients, are important properties for modeling man
arXiv:2609.27958v1 Announce Type: new Abstract: Implicit neural representations (INRs) have emerged as a promising paradigm for video compression, providing compact neural representations with flexibl
arXiv:2609.22512v1 Announce Type: new Abstract: Consensus among LLM judges is often taken as strong evidence that a decision is correct. This assumes that judges make their errors independently. In pr
arXiv:2609.24016v1 Announce Type: new Abstract: Commercial large language models are increasingly deployed across African fintech infrastructure for fraud detection and customer communication, yet no
arXiv:2609.26236v1 Announce Type: new Abstract: Video watermarking underpins copyright protection and provenance for generated media, yet almost every video is compressed by a codec before it is store
arXiv:2609.25942v1 Announce Type: new Abstract: Robots operating around pedestrians often reason over a finite set of predicted human futures. Repeated online updates can concentrate this limited pred
arXiv:2609.26113v1 Announce Type: new Abstract: A state-of-the-art language model asked to interpret 'most of most students passed' typically answers 'most,' though composing two instances of 'most' y
arXiv:2605.11269v2 Announce Type: replace-cross Abstract: Modern gravitational wave astronomy relies on modeling tasks that often require months of graduate-level effort, including building fast wavef
arXiv:2609.26430v1 Announce Type: new Abstract: Dataset distillation (DD) compresses a large training set into a compact synthetic set while preserving downstream training utility. Although DD has bee
arXiv:2609.25048v1 Announce Type: new Abstract: How many prompts does on-policy distillation (OPD) need, and how does the answer depend on the student policies that generate its training responses? We
arXiv:2609.25569v1 Announce Type: new Abstract: Soccer tactics are interactive: an attacking action changes the opponent's defensive problem, and the observed response depends on the multi-agent match
What it is. stuntd is a local proxy for typed LLM decisions: choice, yes/no, score. Your app keeps calling its provider through it. stuntd records the decisions, trains a small head on top of Laya (th
arXiv:2609.24565v1 Announce Type: new Abstract: Architectural drawings encode material classes through repeated hatch patterns. We test whether a connectome-constrained fly visual network, pretrained
arXiv:2609.22183v1 Announce Type: new Abstract: Public benchmark scores may reflect skill, prior exposure to the questions, or both, and for most models the training data are unknown. We present Clean
arXiv:2609.24066v1 Announce Type: new Abstract: Best-of-N is a widely used inference strategy for complex reasoning, whose effectiveness depends on whether sampled candidates can cover diverse and hig
arXiv:2609.22152v1 Announce Type: new Abstract: Imagination performs as a high-level function of large language models (LLMs) which determines the potential of how an LLM creates unseen or creative co
arXiv:2609.23010v1 Announce Type: new Abstract: Iterative text-to-motion generation delivers high-quality and semantically aligned motions but requires multiple network evaluations, resulting in subst
arXiv:2609.23547v1 Announce Type: new Abstract: Graph prompt learning enables parameter-efficient adaptation of frozen Graph Neural Networks to downstream tasks through lightweight prompt parameters.
arXiv:2609.22684v1 Announce Type: cross Abstract: Memory-dependent robotic manipulation often requires later actions to use information from earlier interactions. Existing vision-language-action (VLA)
arXiv:2609.23673v1 Announce Type: new Abstract: In vision-based off-road navigation, a robot needs to know not only which obstacles to avoid but also which terrain is better. The first is handled by f
arXiv:2609.21192v1 Announce Type: new Abstract: Organizations deploying agentic artificial intelligence must determine more than whether a model is trustworthy; they must establish what to validate, c
arXiv:2609.21872v1 Announce Type: new Abstract: We introduce Chronosphere, a spatio-temporal neural field that learns representations of climate. A central challenge in geographic representation learn
arXiv:2609.21154v1 Announce Type: new Abstract: Good tutoring adapts to the individual: it tracks what a learner knows, notices why they go wrong, and asks the next question that will help most. Most
arXiv:2609.22086v1 Announce Type: new Abstract: Professional graphic design is a long-horizon agentic task in which structured, editable artifacts emerge from many interdependent actions, yet outcomes
arXiv:2609.21841v1 Announce Type: new Abstract: Frontier language models now produce professional deliverables that expert graders judge to match human work on a substantial share of economically valu
arXiv:2609.21621v1 Announce Type: new Abstract: Fine-grained object (FO) manipulation requires robots to distinguish a specified FO from visually similar objects and execute actions reliably despite s
metal : support arbitrary hc in dsv4_hc_pre (#29169) the dsv4_hc_pre kernels hardcoded hc = 4 via a constexpr used with simd_shuffle, so the op was rejected by supports_op for any other hc and fell ba
arXiv:2609.20277v1 Announce Type: new Abstract: World Action Models (WAMs) have demonstrated strong robotic manipulation capabilities by augmenting pretrained video generative models with action exper
arXiv:2609.20082v1 Announce Type: cross Abstract: Tool learning enables large language models (LLMs) to use external tools for tasks beyond parametric knowledge. Reinforcement learning can optimize to
Reuters: Mistral and other European AI startups accuse US rivals of using safety concerns to entrench their dominance, rejecting Anthropic's calls to pace the frontier — Europe's leading AI company Mi
arXiv:2609.20330v1 Announce Type: new Abstract: Blind and low-vision users often need to locate a specific personal object rather than an arbitrary instance of the same category. The task calls for a
This week we kicked off the DevFest season in North America at Google Hudson Square in New York City with 80 engineers packed into the room. Typical technical workshops hand you a finished repo, tell
arXiv:2609.20646v1 Announce Type: new Abstract: A vision-language-action (VLA) policy with a flow-matching action expert generates each action chunk (a short command sequence) by integrating a learned
I let AI run my phone 60 real tasks, back to back, on the OnePlus I use every day - Best text model still failed 43% of them - Peak chip temp 98.2 C - 69% of the battery gone The benchmark is AndroidL
arXiv:2609.18950v1 Announce Type: new Abstract: A robot's learned model of its own dynamics is only valid until those dynamics change: actuators wear, payloads shift, and joints stiffen. A model-based
arXiv:2609.17886v1 Announce Type: new Abstract: EEG decoding tasks can rely on different temporal dynamics and cross-channel relationships. We test whether specialized modules improve a fully fine-tun
Editor's note: Lucius AI, a tender-intelligence startup covering markets across five continents, runs its entire data platform on AlloyDB for PostgreSQL with a single operator. By migrating semantic s
Hey Reddit. This is a post asking for advice / user experience. The goal is simple: deploy a small private server for a developer to run a harness that rivals/beats Claude Opus (in perf/intelligence,
arXiv:2609.16321v1 Announce Type: cross Abstract: Existing fairness analysis tools predominantly operate as post-training evaluation frameworks, requiring practitioners to complete the full model deve
Impel deployed Claude across engineering and knowledge work; Anthropic reports shorter new-product ramp time and support gains.
arXiv:2609.17115v1 Announce Type: cross Abstract: Vision-language-action (VLA) systems already bring together two valuable resources for robot learning: rich visual representations and demonstrations
arXiv:2609.17417v1 Announce Type: cross Abstract: Finite knowledge is usually stored extensionally, one code or vector per item. We ask whether a finite collection can instead be stored intensionally,
arXiv:2609.17429v1 Announce Type: cross Abstract: Many learned sequential decision systems map the current state directly to an action. That shortcut becomes brittle when candidate actions are numerou
arXiv:2609.16797v1 Announce Type: new Abstract: Query-based segmentation methods have shown promising potential for surgical instrument segmentation and recognition, which is essential for scene under
Disclaimer: no AI was used whatsoever to write this post Cautionary tale about chasing cheap tokens. exposé: https://kendell.dev/blog/crofaifalse/ reaction by nahcrof, announcing the shutdown of the s
arXiv:2609.15049v1 Announce Type: cross Abstract: An ensemble can improve photovoltaic forecasts while adding components that contribute little or increase computation. We assess these effects through
arXiv:2609.15631v1 Announce Type: new Abstract: Learning a motion prior requires a reward that guides a policy from its current behavior toward demonstrated motion. Adversarial Motion Priors (AMP) pro
arXiv:2609.14711v1 Announce Type: new Abstract: Reward-based learning can alter the very posterior distribution that scientific inference aims to estimate. GRPO-QM sidesteps this by learning only an e
arXiv:2607.27617v2 Announce Type: replace Abstract: Identical language-model answers can arise from hidden states that support different future computations, so current-answer probes do not establish
arXiv:2602.20723v3 Announce Type: replace Abstract: Multimodal recommenders combine collaborative behavior with visual and textual item evidence, whose usefulness varies across user-item interactions.