Popular facts are memorised more deeply during pretraining and resist removal longer than rare ones, yet existing LLM unlearning methods apply uniform gradient pressure regardless of training-data frequency. We propose the AdaPop (Adaptive Popularity) method, which combines local token confidence with a per-fact popularity-dependent exponent derived from an external proxy (e.g., Wikidata sitelinks
Open-weight language models are fine-tuned, quantized, pruned, and merged, yet their provenance is often undocumented. We study data-free white-box lineage verification: can weights alone reveal whether two compatible model checkpoints share ancestry? Residual training produces a shared identity-aligned component in branch products, so this structure alone cannot establish ancestry. We remove it a
712 points, 279 comments
968 points, 495 comments
411 points, 166 comments
When enterprises transition from using simple chat assistants to autonomous, agentic workloads, they quickly run into a hard truth: Agents are prone to inaccurate insights when working with directly raw tables.
At Sierra, we build agents using goals and guardrails so they can think and reason independently, whether originating a mortgage, disputing a charge, or helping select the right pair of skis. That freedom is what makes them so powerful — and the guardrails they are given so important.
Two distinct Einstein--Hilbert sectors are related by a covariant correspondence bridge that maps stress-energy through bitensor kernels. The Bianchi identities enforce conservation, with bridge stresses balancing cross-sector exchange. For positive gravitational couplings and paired Minkowski backgrounds, a fixed metric-independent bridge leaves the quadratic action block diagonal, so the vacuum
220 points, 95 comments
744 points, 310 comments
130 points, 185 comments
Learn how startups use GPT-5.6 to build faster, more cost-efficient AI agents with smarter model selection and new Responses API capabilities.
The problem of computing \emph{diverse} solutions has recently emerged as an important area of study, motivated by applications in fairness, robustness, and security. Instead of returning a single feasible or optimal solution, the goal is to output a \emph{collection} of meaningfully different solutions, often measured by symmetric differences. Diverse variants have been studied using sparsificati
Preview Ultrafast, a new OpenAI API service tier that runs GPT-5.6 Sol up to 14× faster. Powered by Cerebras, it delivers up to 750 output tokens per second.
120 points, 89 comments
August 13, 2026
Agent Skills are today either hand-authored or produced in a single LLM generation pass, and consequently possess no closed loop through which they might improve from the interaction failures they actually cause. Recent work does close this loop, but derives its feedback from single-turn question-answering evaluation. The consequence is a sharp asymmetry: once the first round has patched the gaps
Industrial recommender systems commonly use cascaded retrieval, ranking, and re-ranking pipelines. Although efficient, these pipelines fragment information and objectives across modules, rely on rigid rules, and have limited awareness of real-time intent, leaving session-level shifts among browsing, comparison, and purchase insufficiently addressed. We present DREAM (Developing Recommender Engine
Recent advances in foundation models have enabled AI scientists to automate increasingly complete research workflows, from hypothesis generation and code execution to manuscript preparation. Yet workflow coverage alone does not provide access to the full evidence on which scientific discovery depends. Existing systems typically reason over text, code, labels, or precomputed summaries, leaving scie
Open-ended real-world interaction admits multiple valid behaviors: an agent may answer directly, ask for clarification, provide progress updates, or confirm before acting. This flexibility breaks a core assumption behind group-based RL: rollouts compared within a group are no longer guaranteed to be behaviorally comparable. As a result, reward-model preferences over interaction style can distort r
Should you replace your text-embedding pipeline with a large language model? We answer this with a controlled, cost-aware comparison of ten LLMs across six families and 26 embedding models (118M to 14B parameters) on 37 tasks spanning classification, semantic textual similarity (STS), clustering, pair classification, and retrieval. In aggregate the two paradigms are effectively tied: the best LLM
LLM coding agents issue Bash commands through interfaces that may serialize, wrap, and reparse model output. Matched execution scores alone cannot distinguish command-generation errors from failures introduced after generation. QuoteBench measures this boundary with exact final-state validation on 56 one-shot tasks from 14 incident-derived families, crossing the generation contract with the execut
Electrocardiography (ECG), photoplethysmography (PPG), and phonocardiography (PCG) provide complementary views of the same cardiac cycle, yet existing cardiac foundation models are trained for a single sensing modality, leaving the shared physiology across sensors unexploited. We introduce CardioState-JEPA, a cardiac foundation model to learn a single shared representation jointly across ECG, PPG,
Latent video generation relies on autoencoders to define a compact space in which generative models operate. Although video autoencoder architectures have evolved substantially, their latent spaces are still optimized primarily for pixel-level reconstruction and provide limited high-level semantic organization. A reconstruction-optimal latent space, however, need not be well suited to generative m
Long-form video understanding encompasses tasks that go beyond retrieving isolated events, including tracking an evolving narrative and interpreting social meaning that may remain implicit. However, existing benchmarks rarely evaluate these capabilities jointly, particularly in high-context, non-English media. To address this gap, we introduce NARU, a benchmark designed to evaluate Narrative evolu
343 points, 416 comments
1041 points, 452 comments
632 points, 615 comments
304 points, 228 comments
Introducing sign-language-to-text (SL2T), our breakthrough model powering new sign language features for Deaf and hard of hearing users.
With Sierra’s long-running Horizon agents, insurers can now proactively engage every prospective customer over days, weeks, or months — staying with them until they buy or move on.
1009 points, 936 comments
Affine frequency division multiplexing (AFDM) is a promising chirp-based multicarrier waveform for high-mobility integrated sensing and communication (ISAC). Accurate angle, delay, and Doppler estimation is essential for AFDM sensing. Since target delays and Doppler shifts are generally continuous-valued, representing them on a discrete delay--Doppler grid causes energy leakage and peak displaceme
In this work, we construct fast eighth-order Pade schemes for the direct Zakharov-Shabat scattering problem. The schemes are based on an eighth-order exponential integrator obtained from the Magnus expansion. A direct extension of the conventional fast Pade representation to the eighth-order case leads to insufficiently accurate fast variants in the considered tests. To overcome this difficulty, w
Generative AI
We investigate how small a co-propagating transverse-scalar component can be resolved by Taiji in an already identified bright tensor chirp. Using a source-tracked tensor-null response, we formulate the problem in terms of the minimum resolvable scalar strain fraction and evaluate it over the sky. For a one-year benchmark chirp with tensor signal-to-noise ratio $ρ_T=1000$, we find that Taiji can r
OpenAI research reveals how enterprises are adopting agentic AI, using ChatGPT and Codex, and how frontier firms are pulling ahead in AI adoption.
See how RingCentral uses ChatGPT Work and Codex to accelerate AI product development and centralize operational intelligence across engineering and operations.
AI-powered search products such as ChatGPT search, Google's AI Overviews, and Perplexity provide LLM-synthesized answers grounded in live web results. We developed OreoLook (formerly lixSearch), an open-source answer engine using automated browser agents and provider-routed LLM inference. Its local search, caching, session-management, and embedding stack runs on commodity CPU hardware; answer synt
Creative agents still lack an effective way to learn from high-quality human films, limiting their ability to produce cinematic-grade videos. A key challenge is the absence of a structured video representation that is both faithful to film content and directly usable for agentic reasoning and manipulation. To address the challenge, we propose the Agentic Video Auto-Encoder (AVA-Encoder), a framewo
On 2025 August 18, the LIGO-Virgo-KAGRA collaboration reported a sub-threshold gravitational-wave (GW) candidate, S250818k, consistent with a binary neutron star (NS) merger potentially involving a sub-solar-mass compact object. Follow-up electromagnetic (EM) observations identified a Type IIb supernova, SN 2025ulz within the broad localization area of the GW signal. This potential link between a
Multimodal retrieval and classification across different types of media, spanning text, images,video and audio, has traditionally relied on dual-encoder models that align visual and textual representations through contrastive learning. The March 2026 release of Gemini Embedding 2, Google's first natively multimodal embedding model to map text, images, video, audio, and documents into a single shar
350 points, 334 comments
Health & Bioscience
440 points, 543 comments
For organizations deploying AI agents at scale, there’s often a critical divide between structured and unstructured data. While large language models (LLMs) excel at parsing text documents, emails, and PDFs, they can struggle when presented with raw enterprise databases. Meanwhile, standard natural-language-to-SQL (NL
The autonomous platforms, scheduled for delivery as early as September, will be operated directly by the Portuguese Air Force. The systems will be integrated into maritime surveillance, security, and defense missions, as well as monitoring of the Exclusive Economic Zone (EEZ). TEKEVER, a pan-European technology company specializing in autonomous systems based on artificial intelligence, […]
Agents built on Sierra can now navigate the most challenging IVR systems out of the box: they understand when to call, how to get through a multi-level menu, when to stay silent, and how to recognize when they’ve reached a person.
524 points, 489 comments
Professor Abate will support the AI Office’s scientific assessment of general-purpose AI models, alongside its work on AI innovation, testing and model evaluation.
OpenAI and AWS are making Daybreak cybersecurity capabilities available through Amazon Bedrock to support enterprise security workflows.
# Abstract Thermodynamic concepts are increasingly used in nonlinear photonics to describe Rayleigh--Jeans thermalization, optical wave turbulence, condensation, negative-temperature states, and statistical mode locking. This raises a broader question: how far can thermodynamic reasoning be extended to localized structures maintained far from equilibrium by a balance of gain, loss, dispersion, and
Soniox TTS v2 brings extraordinary voice quality, expressive control, exceptional precision, high-quality voice cloning, and low-latency streaming in 60+ languages.
938 points, 988 comments
Soniox TTS v2 brings extraordinary voice quality, expressive control, exceptional precision, high-quality voice cloning, and low-latency streaming in 60+ languages.
451 points, 425 comments
The space-based detector Laser Interferometer Space Antenna (LISA) will observe inspiralling black hole binaries in the mHz band, many of which may retain orbital eccentricity. In contrast to quasicircular binaries, eccentric systems radiate through multiple orbital harmonics, while relativistic periastron precession introduces a secular phase structure in the waveform. We exploit this structure t
9551 articles sourced historically · 100 per page