Thursday, July 23, 2026
- UMG v. Uncharted Labs (Udio)v3UMG Recordings, Inc., et al. v. Uncharted Labs, Inc. (case number 1:24-cv-04777) is a federal copyright infringement lawsuit filed on June 24
- Multi-Query Attention (MQA)v4Multi-Query Attention (MQA) is a variant of the multi-head attention mechanism used in transformer neural networks in which all query heads share a single key head and a single value head, rather than each…
- Authors Guild v. OpenAIv3Authors Guild et al. v. OpenAI, Inc., et al. is a putative class action copyright lawsuit filed in the United States District Court for the Southern District of New York (SDNY) on September 19, 2023, by The…
- Sliding window attentionv4Sliding window attention (SWA) is a sparse attention pattern in which each query token attends only to a fixed-size window of nearby tokens instead of to every preceding (or every other) token in the sequence.
- Kadrey v. Metav3Kadrey v. Meta Platforms, Inc. is a putative class action lawsuit filed in 2023 by a group of book authors against Meta Platforms, Inc., alleging that Meta infringed their copyrights by training its LLaMA…
- Concord Music Group v. Anthropicv5Concord Music Group, Inc., et al. v. Anthropic PBC is a federal copyright lawsuit in which a coalition of major music publishers, led by Concord Music Group with Universal Music Publishing Group and ABKCO Music
- Eliciting latent knowledgev4Eliciting latent knowledge (ELK) is an open problem in AI alignment formulated by Paul Christiano, Ajeya Cotra, and Mark Xu at the Alignment Research Center (ARC) and introduced in a December 2021 technical…
- Sandbagging (artificial intelligence)v4Sandbagging, in the context of AI safety, refers to the strategic and intentional underperformance of an AI system on a capability evaluation or specific task, typically to hide a capability from human…
- Recursive reward modelingv3Recursive reward modeling (RRM) is a proposed approach to the scalable oversight problem in AI alignment
- UMG v. Sunov3UMG Recordings, Inc., et al. v. Suno, Inc. is a landmark copyright infringement lawsuit filed on June 24, 2024 by a coalition of major record labels
- Getty Images v. Stability AIv4Getty Images v. Stability AI is a pair of parallel intellectual property lawsuits brought by the visual content licensing company Getty Images against Stability AI
- Logit lensv4The logit lens is a foundational technique in mechanistic interpretability for inspecting the intermediate computations of transformer language models.
- Self-consistencyv4Self-consistency is a decoding strategy for large language models that samples multiple chain-of-thought reasoning paths for the same question and returns the answer that the majority of those paths agree on
- WildBenchv4WildBench is an automated evaluation framework for large language models (LLMs) introduced by the Allen Institute for AI (AI2) in 2024.
- nnsightv3nnsight is an open-source Python library for the interpretation and intervention of deep learning models, developed by the Bau Lab at Northeastern University.
- LMArenav4LMArena is a crowdsourced artificial intelligence evaluation platform and company that ranks large language models by having anonymous users vote on which of two blind, side-by-side model responses they prefer
- TransformerLensv4TransformerLens is an open-source Python library for the mechanistic interpretability of GPT-style language models.
- Activation patchingv3Activation patching is a causal intervention technique used in mechanistic interpretability to identify which internal components of a neural network are causally responsible for a specific behaviour.
- Refusal directionv3The refusal direction is a finding from mechanistic interpretability research that the refusal behavior of safety fine-tuned chat language models is mediated by a single
- Mira Murativ5Ermira "Mira" Murati (born December 16, 1988) is an Albanian-American engineer and technology executive who served as Chief Technology Officer of OpenAI from May 2022 to September 2024 and is the co-founder…
- New York Times v. OpenAIv4The New York Times Company v. Microsoft Corporation, OpenAI Inc., et al. (case number 1:23-cv-11195) is a federal copyright lawsuit filed by The New York Times Company against Microsoft and OpenAI in the U.S.…
- SPIN (Self-Play Fine-Tuning)v5SPIN (Self-Play fIne-tuNing) is a post-training method for large language models introduced by researchers at the University of California, Los Angeles (UCLA) in January 2024.
- Bartz v. Anthropicv5Bartz et al. v. Anthropic PBC, case number 3:24-cv-05417, is a landmark copyright class action in which authors sued Anthropic, maker of the Claude large language models, for training on copyrighted books
- Periodic Labsv5Periodic Labs is an American artificial intelligence research startup that builds AI systems and autonomous laboratories aimed at accelerating the discovery of new physical materials.
- Golden Gate Claudev4Golden Gate Claude was a temporary, research-oriented public demonstration released by Anthropic on May 23, 2024
- Crosscoderv5A crosscoder is a mechanistic interpretability tool, introduced by Anthropic in October 2024, that generalizes the sparse autoencoder (SAE) and the transcoder by learning a single shared dictionary of sparse…
- BigCodeBenchv4BigCodeBench is a Python code generation benchmark of 1,140 function-level programming tasks that require composing 723 distinct function calls from 139 libraries across seven domains
- GitHub Copilot Workspacev4GitHub Copilot Workspace was a task-centric, AI-powered developer environment built by GitHub Next, the research and incubation arm of GitHub.
- SWE-Bench Prov4SWE-Bench Pro (stylized SWE-BENCH PRO) is a contamination-resistant benchmark, released by Scale AI in September 2025, that measures whether an AI coding agent can resolve long-horizon
- AWS Inferentiav4AWS Inferentia is a family of custom application specific integrated circuits (ASICs) designed by Amazon Web Services for machine learning inference in the cloud, built to deliver, in AWS's words, "high…
- Mila (Quebec AI Institute)v5Mila, formally styled Mila - Quebec Artificial Intelligence Institute (in French, Mila - Institut quebecois d'intelligence artificielle)
- Transcoderv4A transcoder is a sparse neural network used in mechanistic interpretability research to approximate the input-to-output function of a component inside a transformer (most commonly an MLP sublayer) using a…
- ARIA (UK)v4The Advanced Research and Invention Agency (ARIA) is a United Kingdom government research funding body, established by Act of Parliament in 2022 and made operational in January 2023, that backs "high-risk…
- FAR.AIv4FAR.AI is an artificial intelligence safety research and education non-profit based in Berkeley, California, that conducts technical research on robustness, alignment, deception, and model evaluation while…
- NIST ARIAv5NIST ARIA (Assessing Risks and Impacts of AI) is a testing, evaluation, validation, and verification (TEVV) program operated by the United States National Institute of Standards and Technology (NIST) to…
- Conjecture (AI Safety Lab)v4Conjecture is a London-based artificial intelligence safety research company founded in March 2022 by Connor Leahy, Sid Black, and Gabriel Alfour, all alumni of EleutherAI.
- Blueprint for an AI Bill of Rightsv4The Blueprint for an AI Bill of Rights is a non-binding policy framework released by the White House Office of Science and Technology Policy (OSTP) on October 4, 2022, during the Biden-Harris administration.
- Xynova Flex 1v5The Xynova Flex 1 is a high-degree-of-freedom tendon-driven dexterous hand built by Xynova (Chinese: 曦诺未来), a Hangzhou robotics startup founded in December 2024.
- Xynova Flex 2v3The Xynova Flex 2 is the second-generation dexterous hand developed by Hangzhou-based robotics startup Xynova, publicly launched on May 13
- Superposition (Mechanistic Interpretability)v5Superposition is the phenomenon in which an artificial neural network represents more distinct features than it has dimensions in its activation space, by assigning those features to nearly-orthogonal (rather…
- Joint Embedding Predictive Architecturev7Joint Embedding Predictive Architecture (JEPA) is a family of self-supervised, non-generative neural network architectures proposed by Yann LeCun in his June 2022 position paper A Path Towards Autonomous…
- Interim Measures for the Management of Generative AI Servicesv5The Interim Measures for the Management of Generative Artificial Intelligence Services (Chinese: 生成式人工智能服务管理暂行办法
- Weak-to-Strong Generalizationv4Weak-to-Strong Generalization is an empirical research direction, introduced in a December 2023 paper by OpenAI's Superalignment team
- AlphaGeometry 2v5AlphaGeometry 2 (often abbreviated AG2) is a neuro-symbolic artificial intelligence system built by Google DeepMind that solves Olympiad-level Euclidean geometry problems by pairing a Gemini-based language…
- ChatGPT Agentv6ChatGPT Agent is an agentic feature of ChatGPT from Sam Altman's OpenAI that gives the chatbot its own virtual computer (complete with a graphical browser, a text browser, a Linux-style terminal, code…
- South Korea AI Basic Actv5The South Korea AI Basic Act, officially the Framework Act on the Development of Artificial Intelligence and Establishment of a Foundation for Trust (Korean: 인공지능 발전과 신뢰 기반 조성 등에 관한 기본법)
- Polysemanticityv5Polysemanticity is the phenomenon in artificial neural networks in which a single neuron (or directional unit such as an attention head) activates strongly for multiple, semantically unrelated inputs or…
- MATH-500v6MATH-500 is a 500-problem benchmark for evaluating the mathematical reasoning of large language models, formed by holding out 500 problems from the test split of the MATH benchmark of Dan Hendrycks et al.
- NuminaMathv5NuminaMath is a family of openly licensed competition-mathematics resources developed by the non-profit Project Numina, spanning the largest public dataset of competition math problems and solutions, a set of…
- Goedel-Proverv4Goedel-Prover is an open-source large language model designed for automated formal theorem proving in Lean 4.
- Genie 2v4Genie 2 is a foundation world model developed by Google DeepMind, unveiled on December 4, 2024.
- Representation Engineeringv6Representation Engineering (often abbreviated RepE) is a top-down approach to artificial-intelligence transparency and control that reads and manipulates high-level concepts (such as honesty, harmlessness, and…
- Attribution Graphsv5Attribution graphs are a mechanistic interpretability technique developed by Anthropic that traces the internal "circuits" a large language model uses to turn a specific prompt into a specific output.
- Paris AI Action Summitv7The Paris AI Action Summit (French: Sommet pour l'action sur l'IA) was the third major international summit on artificial intelligence in the so-called "AI Safety Summit" series, held on 10-11 February 2025 at…
- Colorado Artificial Intelligence Actv5The Colorado Artificial Intelligence Act (commonly abbreviated CAIA and codified as Senate Bill 24-205, "Consumer Protections for Artificial Intelligence") was a state statute signed by Governor Jared Polis on…
- Trillium (TPU v6e)v6Trillium, also designated TPU v6e, is the sixth-generation Tensor Processing Unit developed by Google for machine learning training and inference workloads on Cloud TPU infrastructure.
- Superalignmentv7Superalignment is the technical problem of steering and controlling AI systems that are far more capable than their human supervisors, that is, systems at or beyond the level of superintelligence, and the name…
- Ring Attentionv7Ring Attention, formally Ring Attention with Blockwise Transformers, is a distributed algorithm for computing the self-attention operation of transformer neural networks across a ring of compute devices
- MiniMax M1v6MiniMax M1 (stylised MiniMax-M1) is an open-weight large language reasoning model released on 16 June 2025 by the Shanghai-based artificial-intelligence company MiniMax
- Arena-Hardv7Arena-Hard (and its evaluation tool Arena-Hard-Auto) is an automatic large language model (LLM) benchmark developed by the team behind Chatbot Arena that scores instruction-tuned models on 500 challenging
- BitNetv7BitNet is a family of large language model architectures developed by Microsoft Research Asia that constrain the weights of a transformer to extremely low bit-widths: initially a single bit ({-1, +1}) and…
- MiniMax-Text-01v6MiniMax-Text-01 is an open-weights, large-scale mixture-of-experts (MoE) language model released by Shanghai-based AI company MiniMax on January 14, 2025.
- Texas Responsible AI Governance Actv5The Texas Responsible Artificial Intelligence Governance Act (TRAIGA), enacted as House Bill 149 of the 89th Texas Legislature, Regular Session, is a Texas state law that regulates artificial intelligence…
- SambaNova SN40Lv6The SambaNova SN40L is a reconfigurable dataflow AI accelerator designed by SambaNova Systems and unveiled on September 19, 2023.
- Positron AIv4Positron AI is an American semiconductor startup headquartered in Reno, Nevada, that designs and manufactures purpose-built hardware for transformer inference.
- Capability overhangv7Capability overhang is a term used in ai safety and AI policy discourse to describe a situation in which the latent capabilities of a deployed AI system
- Inner alignmentv6Inner alignment is the AI-safety problem of ensuring that a learned model which is itself an optimizer (a mesa-optimizer) pursues the objective the training process actually selected for (the base objective)
- MusicGenv6MusicGen is an open-weights text-to-music generation model from Meta AI's Fundamental AI Research (FAIR) team, released on 8 June 2023, that generates roughly 30-second music clips from a text prompt, a…
- SWE-Lancerv6SWE-Lancer is a benchmark released by OpenAI in February 2025 that evaluates the ability of frontier large language models to perform real-world freelance software-engineering work.
- Outer alignmentv6Outer alignment is the problem of specifying a training objective (typically a loss function, reward signal, or preference dataset) that correctly captures what the designers of a machine-learning system…
- Deceptive alignmentv7Deceptive alignment is a hypothesised AI failure mode in which a trained model internally pursues an objective different from the one specified by its training signal, yet deliberately behaves as if it shares…
- OpenAI o3-prov5OpenAI o3-pro is a high-compute reasoning large language model released by OpenAI on June 10, 2025, designed as the professional, higher-reliability variant of the company's o3 reasoning model.
- Mesa-optimizationv6Mesa-optimization is the situation in AI alignment research in which a learned model, typically a neural network produced by a machine-learning training process, is itself an optimizer that internally searches…
- Sycophancy (artificial intelligence)v6Sycophancy in artificial intelligence is the tendency of large language models to tell users what they want to hear: tailoring responses to match a user's perceived beliefs, preferences, or emotional state…
- Stable Diffusion 3.5v6Stable Diffusion 3.5 (SD 3.5) is a family of open-weights text-to-image diffusion models released by Stability AI on October 22, 2024, comprising three variants: Stable Diffusion 3.5 Large (8.1 billion…
- GPT-5 Prov5GPT-5 Pro is a large language model developed by OpenAI and the highest-capability variant of the GPT-5 model family.
- Tripo (3D generation)v6Tripo is an artificial-intelligence platform for generating three-dimensional models from text prompts, single images, multi-view images, or sketches.
- Yejin Choiv6Yejin Choi (born 1977) is a South Korean-American computer scientist and the Dieter Schwarz Foundation HAI Professor and Professor of Computer Science at Stanford University
- Daniel Grossv6Daniel Gross (born 1991) is an Israeli-American entrepreneur and venture investor who co-founded the AI laboratory Safe Superintelligence Inc. with Ilya Sutskever and Daniel Levy in June 2024 and served as its…
- Pika Labsv5Pika Labs (legally incorporated as Mellis, Inc. and doing business as Pika) is an American generative AI company that builds consumer software for creating short AI videos from text, images, or clips.
- Frontier Safety Framework (Google DeepMind)v6The Frontier Safety Framework (FSF) is Google DeepMind's risk-management framework for identifying and mitigating severe risks from advanced frontier AI models, first published on 17 May 2024 and updated to…
- Nat Friedmanv5Nathaniel Dourif "Nat" Friedman (born August 6, 1977) is an American entrepreneur and investor who served as chief executive officer of GitHub from 2018 to 2021, where he oversaw the launch of GitHub Copilot…
- Christopher Manningv5Christopher Manning is an Australian-American computer scientist and computational linguist at Stanford University who is one of the most cited researchers in natural language processing and a central figure…
- Decagon (AI company)v4Decagon is an American artificial intelligence company that builds AI agents for enterprise customer service, deploying conversational agents that resolve customer inquiries autonomously across chat, email…
- David Silverv7David Silver is a British computer scientist whose work has defined the modern field of deep reinforcement learning and computer game-playing.
- SB 1047 (California Safe and Secure Innovation for Frontier Artificial Intelligence Models Act)v6SB 1047, officially the Safe and Secure Innovation for Frontier Artificial Intelligence Models Act
- Stanford Institute for Human-Centered Artificial Intelligencev6The Stanford Institute for Human-Centered Artificial Intelligence (HAI) is a university-wide research institute at Stanford University, founded in 2019 and co-directed at launch by computer-science professor…
- Bob McGrewv4Bob McGrew is an American computer scientist, engineering executive, and technology investor best known for serving as Chief Research Officer at openai until his departure on September 25, 2024.
- Noam Shazeerv7Noam Shazeer (born 1975 or 1976) is an American computer scientist and software engineer who is one of the most prolific and influential researchers in the modern era of deep learning, best known as a…
- Aravind Srinivasv7Aravind Srinivas (born 1994) is an Indian-American computer scientist and entrepreneur who co-founded and serves as chief executive officer of Perplexity, the artificial-intelligence company that operates a…
- Tri Daov5Tri Dao is a computer scientist who created the FlashAttention family of GPU attention algorithms and co-created the Mamba selective state-space architecture
- John Schulmanv6John Schulman is an American artificial intelligence researcher, one of the eleven original co-founders of OpenAI, and the inventor of Proximal Policy Optimization (PPO), the reinforcement-learning algorithm…
- Multi-token predictionv2Multi-token prediction (often abbreviated MTP) is a language modeling training objective in which the model is trained to predict several future tokens at each context position rather than only the next token.
- Aidan Gomezv6Aidan N. Gomez (born 1996) is a British-Canadian computer scientist and technology executive who is the co-founder and chief executive officer of Cohere
- Activation steeringv5Activation steering is a family of inference-time techniques in mechanistic interpretability and AI safety that modify a neural network's internal activations to influence its behavior, without retraining the…
- Percy Liangv4Percy Liang is an associate professor of computer science at Stanford University and the founding director of the Stanford Center for Research on Foundation Models (CRFM), the lab that coined the term…
- NIST AI Risk Management Frameworkv3The NIST AI Risk Management Framework, commonly abbreviated AI RMF, is a voluntary guidance document released by the United States National Institute of Standards and Technology on January 26, 2023, that helps…
- Albert Guv6Albert Gu is an American computer scientist, Assistant Professor of Machine Learning at Carnegie Mellon University, and co-founder and Chief Scientist of Cartesia AI.
- Mike Knoopv6Mike Knoop is an American technology entrepreneur and artificial intelligence researcher who co-founded the workflow-automation company Zapier in 2011 and, in 2024, co-founded the ARC Prize
- Jakub Pachockiv5Jakub Pachocki (born 1991) is a Polish computer scientist and former competitive programmer who has served as the Chief Scientist of OpenAI since May 2024.
