Glossary

OpenAI

OpenAI is an AI research company founded by Elon Musk and Sam Altman that develops and commercializes large language models and AI systems, including GPT models and APIs for developers building AI applications.

Anthropic

Anthropic is an AI safety company founded by Dario Amodei that develops Claude, a large language model used for a range of applications including agentic coding tasks, which has captured an estimated 40% of enterprise AI market share.

Claude Code

Claude Code is an AI-powered coding agent developed by Anthropic that enables developers to perform complex software tasks directly within their editor, including orchestrating machine learning workflows like model fine-tuning through integration with external services like Hugging Face.

Claude

Claude is an AI assistant developed by Anthropic that specializes in coding, reasoning, and agentic tasks, competing with models like GPT-4 and Google's Gemini in the large language model space.

Hugging Face

Hugging Face is an AI platform that hosts machine learning models, datasets, and applications, serving over 13 million users and maintaining more than 2 million public models and 500,000 public datasets as of Spring 2026.

Google

Google is a multinational technology company headquartered in Mountain View, California, known for its search engine, advertising platform, and artificial intelligence products including the Gemini family of large language models and generative AI tools.

Codex

Codex is OpenAI's AI coding model that interprets natural language commands and generates code to power AI-assisted software development applications and workflows.

Simon Willison

Simon Willison is a developer and technical writer known for documenting AI-assisted development practices, including experiments with Claude AI coding tools, JavaScript sandboxing research, and Git workflows for working with coding agents.

ChatGPT

ChatGPT is a conversational AI assistant developed by OpenAI that uses large language models to generate human-like responses and is available as a web interface, mobile app, and API with over 900 million weekly active users.

GPT-5

GPT-5 is a large language model developed by OpenAI that demonstrates advanced capabilities in scientific reasoning, mathematics, and code generation, with variants like GPT-5.3-Codex-Spark optimized for specific domains.

Cursor

Cursor is an AI-powered code editor that uses large language models to assist with software development tasks, competing with Claude Code and other agentic IDEs.

NVIDIA

NVIDIA is a technology company founded by Jensen Huang that designs and manufactures GPUs and AI accelerators used for computing, machine learning, and data center inference workloads, and also develops software platforms like CUDA.

Meta

Meta is a technology company headquartered in Menlo Park, California, run by founder Mark Zuckerberg, known for the social media platforms Facebook and Instagram, and increasingly for developing and open-sourcing large language models and AI systems, particularly the Llama family of models.

Google DeepMind

Google DeepMind is an AI research division of Google that develops large language models like the Gemini family, which achieve state-of-the-art performance on reasoning, programming, and mathematical benchmarks, and deploys AI agents for tasks such as code security and computer control.

HuggingFace

HuggingFace is an open-source AI platform and model repository that provides tools, libraries, and pre-trained transformer models for building and deploying machine learning applications, and serves as a hub for distributing and collaborating on AI models and datasets.

Microsoft

Microsoft is a multinational technology company headquartered in Redmond, Washington, founded by Bill Gates and Paul Allen in 1975, known for its software products including Windows and Office, cloud computing services through Azure, and recent investments in artificial intelligence including partnerships with OpenAI and development of AI infrastructure and models.

LLM

An LLM (Large Language Model) is a neural network trained on vast amounts of text data to predict and generate human language, with capabilities that improve predictably as model size and training compute scale up.

prompt injection

Prompt injection is a security attack where an attacker embeds malicious instructions within user input or external content to manipulate an AI model's behavior, bypassing its original instructions through social engineering or string-based overrides, with defenses including instruction hierarchy training and system-level constraints.

GPT-5.2

GPT-5.2 is an OpenAI large language model known for demonstrating reasoning capability in theoretical physics and serving as the foundation for specialized variants including GPT-5.2-Codex, which is optimized for agentic coding tasks.

OpenClaw

OpenClaw is an agent framework benchmark and agentic AI platform used to evaluate and develop multi-agent AI systems, known for establishing performance standards in the agentic coding and knowledge work automation space.

GitHub

GitHub is a web-based platform owned by Microsoft that provides version control, code hosting, and collaboration tools for software development, and is the home of the Copilot AI coding assistant.

AGI

AGI (Artificial General Intelligence) is a theoretical AI system capable of understanding and performing intellectual tasks across any domain at a human or superhuman level, without domain-specific training.

DeepSeek

DeepSeek is a Chinese AI company known for developing efficient open-source large language models and reasoning models, particularly the DeepSeek R1, which achieved competitive performance with proprietary models while using novel architectural innovations to reduce computational costs.

MCP

MCP (Model Context Protocol) is an open standard that enables AI agents and language models to dynamically discover and invoke tools from servers, facilitating integration of external capabilities into AI applications and agentic workflows.

AWS

AWS (Amazon Web Services) is a cloud computing platform run by Amazon that provides on-demand infrastructure services including compute, storage, and AI/ML tools, and hosts major AI companies like Anthropic and OpenAI.

GPT-4

GPT-4 is a large language model developed by OpenAI that can process and generate text, with multimodal capabilities to handle images, and is known for advanced reasoning and natural language understanding tasks.

Gemini

Gemini is a large language model developed by Google DeepMind known for advanced reasoning capabilities, particularly through its "Deep Think" feature that enables extended chain-of-thought problem-solving, as demonstrated by gold-medal performance at the International Mathematical Olympiad and International Collegiate Programming Contest.

Andrej Karpathy

Andrej Karpathy is an AI researcher known for developing the "autoresearch" framework that enables autonomous AI agents to run experiments and improve large language models, and for his recent work on agentic AI systems and predictions about AGI timelines.

GPT-5.4

GPT-5.4 is OpenAI's frontier language model that combines reasoning, coding, and agentic capabilities with native computer-use support, a 1M token context window, and a "Thinking" mode for visible planning steps, available through ChatGPT, the API, and Codex.

Claude Opus 4.6

Claude Opus 4.6 is a large language model developed by Anthropic that excels at agentic tasks and long-horizon reasoning, recently achieving benchmarks for autonomous AI agents working on multi-hour engineering tasks.

Apple

Apple Inc. is a technology company founded by Steve Jobs, Steve Wozniak, and Ronald Wayne that designs and manufactures consumer electronics including iPhones, iPads, Macs, and related services, and is headquartered in Cupertino, California.

System Card

A System Card is a technical documentation document published by OpenAI that discloses a model's safety evaluations, capability assessments, performance benchmarks, deployment constraints, and known failure modes.

GPT-4o

GPT-4o is a multimodal AI model developed by OpenAI that processes text, images, and other inputs to generate responses, and is used as a production AI API available to developers.

Gemini 3.1 Pro

Gemini 3.1 Pro is Google's AI model released in February that offers approximately 2x reasoning performance improvement over Gemini 3 Pro and is designed for complex problem-solving tasks, available to developers through API access and consumer platforms.

OpenAI Codex

OpenAI Codex is an AI coding agent developed by OpenAI that assists developers with code generation and software engineering tasks, having reached over 2 million weekly active users as of 2026.

o1

o1 is OpenAI's reasoning model that uses extended chain-of-thought processing to solve complex problems in coding, mathematics, and science by working through problems step-by-step before generating responses.

coding agents

Coding agents are autonomous AI systems trained to write, review, and modify software code, often with access to version control systems, documentation, and development tools, used by engineers to accelerate development workflows while requiring monitoring for behavioral misalignment and safety.

Amazon

Amazon is a multinational technology and e-commerce company headquartered in Seattle that operates cloud computing services (AWS), runs an e-commerce marketplace, and invests in AI infrastructure and partnerships including a recent $50 billion investment in OpenAI.

xAI

xAI is an AI company founded by Elon Musk that develops AI models and coding assistants, including the Grok model, and is working on integration projects with Tesla.

o3

o3 is an AI reasoning model developed by OpenAI that combines advanced reasoning capabilities with integrated tool use including web browsing, Python execution, and image analysis, and is designed for complex problem-solving tasks like code review and multi-step reasoning.

GPT

GPT is a series of large language models developed by OpenAI that can generate human-like text, answer questions, and support function calling to return structured outputs for tool-use applications.

DeepMind

DeepMind is an AI research laboratory owned by Google that develops reinforcement learning systems and advanced AI models, known for creating AlphaGo, AlphaFold, and AlphaProof, and for powering agentic AI systems like AlphaEvolve.

reinforcement learning

Reinforcement learning is a machine learning approach where an AI agent learns to make decisions by receiving rewards or penalties for its actions, enabling it to optimize behavior through trial and error, as demonstrated by DeepMind's AlphaGo system that mastered the game of Go.

Gemini 2.5 Pro

Gemini 2.5 Pro is Google DeepMind's AI model family available in stable and experimental versions, designed for production use with enhanced reasoning capabilities, coding optimization, and support for extended context windows and multi-modal agent workflows.

Responses API

Responses API is an open-source standard led by Hugging Face that standardizes response formats for agentic AI workloads, supporting tool use, multi-step reasoning, and stateful loops as a community-owned alternative to OpenAI's Chat Completions API.

agentic coding

Agentic coding is a development approach where AI agents autonomously write, debug, and test code with minimal or no human intervention, using large language models to reason about programming tasks and generate working implementations across multiple programming languages.

Astral

Astral is a Python developer tooling company best known for creating uv (a package manager), Ruff (a linter), and Ty, and was acquired by OpenAI in 2025 to integrate its tools into OpenAI's Codex coding agent.

Pentagon

The Pentagon is the United States Department of Defense headquarters that oversees military operations, defense policy, and procurement contracts for the armed forces.

Transformers

Transformers is a neural network architecture that uses attention mechanisms to process sequential data in parallel, widely used as the foundation for large language models and developed at Google (originally introduced in the 2017 "Attention Is All You Need" paper).

Bun

Bun is a JavaScript runtime and toolkit that serves as an alternative to Node.js, acquired by Anthropic in 2024 as part of the AI lab's strategy to own developer infrastructure.

Model Context Protocol

Model Context Protocol is an open standard developed by Anthropic that enables AI agents and language models to discover, connect to, and invoke external tools and services dynamically.

Python

Python is a high-level, general-purpose programming language known for its simple syntax and wide adoption in web development, data science, and artificial intelligence applications.

Stripe

Stripe is a financial services company that provides payment processing software and APIs, allowing businesses and developers to accept payments online.

uv

uv is a modern Python package manager and project management tool developed by Astral (acquired by OpenAI in 2025) that achieves approximately 126 million downloads per month and is widely used by Python developers for dependency resolution and project workflows.

ruff

Ruff is a fast Python linter and code formatter developed by Astral, a Python developer tooling company acquired by OpenAI in 2025.

GPT-5.3-Codex

GPT-5.3-Codex is OpenAI's agentic coding model that combines frontier coding capabilities with reasoning abilities, running 25% faster than its predecessor and achieving state-of-the-art performance on software engineering benchmarks like SWE-Bench Pro.

Ars Technica

Ars Technica is a technology news and analysis publication that covers topics including artificial intelligence, space exploration, cybersecurity, and other tech industry developments.

Cloudflare

Cloudflare is a web infrastructure and security company that provides content delivery, DDoS protection, DNS, and API services to websites and applications, operating a global network of data centers.

o4-mini

o4-mini is an OpenAI reasoning model designed for cost and latency-efficient tasks, capable of tool use within chain-of-thought reasoning and used in applications like code review and autonomous coding agents.

RLHF

RLHF (Reinforcement Learning from Human Feedback) is a training technique that uses human evaluations of AI model outputs to fine-tune and improve model behavior through reinforcement learning.

agentic workflows

Agentic workflows are AI-powered processes where autonomous agents perform sequential tasks or decision-making steps, commonly used in applications like AI-powered development tools and agent frameworks that require fast reasoning and low latency.

chain-of-thought reasoning

Chain-of-thought reasoning is a technique where AI models work through problem-solving steps sequentially before providing a final answer, improving accuracy on complex tasks.

Claude Opus 4.5

Claude Opus 4.5 is an AI language model developed by Anthropic that is designed for agentic task completion, enabling autonomous code generation and agent orchestration through integrated tools and reasoning capabilities.

RAG

RAG (Retrieval-Augmented Generation) is a technique that augments large language models with external retrieval systems to fetch relevant documents or data before generating responses, improving accuracy and enabling access to up-to-date information beyond a model's training data.

large language models

Large language models are neural networks trained on vast amounts of text data that predict and generate human language by processing sequential tokens, with examples including Meta's Llama and OpenAI's GPT series that are widely used for AI-powered applications.

GPT-5.4 mini

GPT-5.4 mini is OpenAI's efficient small language model designed for high-volume, latency-sensitive workloads including coding assistants, subagents, and multimodal applications, running 2x faster than GPT-5 mini while approaching GPT-5.4 performance on benchmarks.

GPT-5.4 nano

GPT-5.4 nano is OpenAI's smallest and lowest-cost large language model from the GPT-5.4 family, priced at $0.20/$1.25 per million tokens, designed for high-volume latency-sensitive workloads including coding assistants, subagents, and multimodal applications with a 400k context window.

autoresearch

Autoresearch is an agentic coding pattern where an AI system autonomously designs and runs numerous experiments to optimize software performance or train machine learning models, with notable implementations by Andrej Karpathy (improving LLM training efficiency) and Shopify (accelerating their Liquid template engine).

Opus 4.6

Opus 4.6 is Anthropic's large language model featuring a 1M-token context window, designed for multi-agent tasks and autonomous agent systems, and notable for reliably executing extended agent loops.

Groq

Groq was an AI inference chip startup that developed specialized processors (LPUs) optimized for fast token generation, which Nvidia acquired for approximately $20 billion in 2025.

Opus 4.5

Opus 4.5 is a large language model developed by Anthropic that powers Claude Code and is widely used for agentic AI tasks, complex problem-solving, and coding assistance, having been released in November 2025.

agentic AI

Agentic AI refers to autonomous artificial intelligence systems that can independently plan and execute multi-step tasks toward defined goals, rather than simply responding to individual user queries, with examples including OpenAI's ChatGPT agent and Anthropic's Claude Code that can take actions like browsing, coding, and research without requiring user direction for each step.

Dario Amodei

Dario Amodei is the CEO of Anthropic, an AI safety company that develops the Claude language model, and is known for advocating the "Big Blob of Compute Hypothesis" regarding AI scaling laws and making predictions about AI's economic impact.

Grok

Grok is an AI chatbot developed by xAI (Elon Musk's company) that generates text responses to user prompts, known for occasionally producing inaccurate information and generating explicit images from certain prompts.

Daring Fireball

Daring Fireball is a technology blog run by John Gruber that focuses on Apple, Mac software, web design, and tech industry commentary, known for its critical analysis and use of formats like "claim chowder" to document bold predictions for future accountability.

Moltbook

Moltbook is an AI agent social network built on OpenClaw that was acquired by Meta and represents a large-scale agent ecology with tens of thousands of AI agents.

AI agents

AI agents are autonomous software systems that use large language models to perceive their environment, make decisions, and take actions toward defined goals, often by interacting with external tools or data sources while managing memory and handling security challenges like prompt injection attacks.

SWE-bench

SWE-bench is a benchmark that evaluates the capability of AI models and autonomous agents to resolve software engineering tasks, particularly bug fixes and code modifications on real-world GitHub repositories.

GPT-5.2-Codex

GPT-5.2-Codex is an agentic coding model developed by OpenAI, optimized for long-horizon coding tasks such as large refactors and migrations, with enhanced Windows support and cybersecurity capabilities.

chain-of-thought

Chain-of-thought is a technique that prompts AI models to break down complex reasoning tasks by generating intermediate steps or explanations before producing a final answer, which improves performance on multi-step problems and enables better evaluation of model reasoning processes.

Sam Altman

Sam Altman is the CEO of OpenAI, an AI research company, who leads development of large language models and has advocated for government regulation of advanced AI systems.

Function calling

Function calling is a capability that allows large language models to receive descriptions of functions or tools and return structured arguments to invoke those functions, enabling AI systems to interact with external APIs and tools rather than just generating text.

scaling laws

Scaling laws are empirical relationships, established through research like OpenAI's landmark studies, that describe how neural language model performance improves predictably with increases in model size, dataset size, and computational resources, following power-law patterns across many orders of magnitude.

Gemini 3 Flash

Gemini 3 Flash is a large language model developed by Google DeepMind that combines advanced reasoning capabilities with fast response times and lower costs for AI agent and automation applications.

Gemini 2.5 Flash

Gemini 2.5 Flash is a hybrid reasoning AI model developed by Google DeepMind that allows developers to toggle extended thinking on and off with configurable thinking budgets to balance performance, cost, and latency.

Gemini API

Google's API that provides developers access to Gemini AI models, including Gemini 2.0 Flash and other variants, for building AI applications with capabilities in text generation, multimodal outputs, tool use, and agentic reasoning.

Gemini 3 Deep Think

Gemini 3 Deep Think is Google's AI model designed for science and engineering tasks involving messy or complex data, available to Ultra subscribers in the Gemini app and to researchers and enterprises via API early access.

Mixture of Experts (MoE)

Mixture of Experts (MoE) is an architecture design for large language models that routes each token through only a small subset of specialized neural network components called experts, allowing sparse models to decouple computational capacity from inference cost while maintaining or exceeding baseline performance.

llama.cpp

llama.cpp is an open-source inference runtime created by Georgi Gerganov and the GGML team that enables efficient local execution of large language models on personal computers and servers.

CUDA

CUDA is a parallel computing platform and API developed by NVIDIA that enables software developers to use graphics processing units (GPUs) for general-purpose processing by writing specialized code kernels that run on NVIDIA GPUs.

GPT-OSS

GPT-OSS is an open-source language model developed by OpenAI that is optimized for agentic reinforcement learning training, supporting multi-step tool use and inference through frameworks like the transformers library.

Gemini CLI

Gemini CLI is a command-line interface tool compatible with Hugging Face's HF-skills repository that enables coding agents to perform machine learning tasks such as model fine-tuning, reinforcement learning alignment, checkpoint evaluation, and model publishing through natural language commands.

Deep Research

Deep Research is an AI research agent developed by Tavily that automatically conducts in-depth investigations and analysis tasks, designed to scale with future model improvements rather than rely on hand-crafted heuristics.

MiniMax

MiniMax is a Chinese AI company that develops open-source large language models, including the M2.7 series, which achieve performance competitive with state-of-the-art models like GLM-5 and Claude Sonnet while reducing computational costs.

vibe coding

Vibe coding is a software development approach that uses AI language models as high-level programming tools to translate developer intent directly into code, rather than writing code through traditional manual syntax.

Agentic AI Systems

Agentic AI systems are AI applications that autonomously perceive their environment, plan actions, and use external tools to accomplish goals with minimal human intervention, raising key concerns about safety, alignment, and human oversight.

Llama

Llama is an open-weight large language model family developed by Meta that includes multimodal capabilities and is designed for on-device and local deployment across various hardware platforms.

Copilot

Copilot is Microsoft's AI-powered coding assistant that uses large language models to help developers write, review, and debug code faster.

Claude Cowork

Claude Cowork is an agentic AI workspace tool developed by Anthropic that orchestrates multiple Claude Code instances to handle multi-step workflows and complex projects for both technical and non-technical users.

Agentic systems

Agentic systems are AI applications where language models autonomously execute tasks by planning actions, calling tools, and processing their outputs to achieve user-defined goals, with security considerations including prompt injection defense and safe handling of untrusted tool outputs.

Jevons Paradox

The Jevons Paradox is an economic principle stating that improvements in efficiency of resource use tend to increase overall consumption of that resource, as lower costs stimulate greater demand.

Gemini 3.1 Flash-Lite

Gemini 3.1 Flash-Lite is a lightweight language model developed by Google that features adjustable thinking levels and is designed for high-volume, cost-efficient workloads.

Qwen

Qwen is a family of large language models developed by Alibaba that supports hybrid architectures mixing RNN/SSM layers with attention mechanisms for improved long-context performance and inference efficiency.

Superintelligence

Superintelligence is a hypothetical artificial intelligence system that would possess intellectual capabilities far exceeding those of humans across virtually all domains, including scientific research, strategy, and complex reasoning.

GitHub Copilot

GitHub Copilot is an AI-powered code completion tool developed by GitHub that uses machine learning models to suggest code snippets and generate code in real-time within developer environments.

agentic coding workflows

Agentic coding workflows are development processes where AI coding agents autonomously generate, modify, and commit code changes while providing traceability through session logs and integration with version control systems.

Stanford

Stanford University is a private research university located in Stanford, California, known for its engineering and computer science programs, artificial intelligence research, and economics department.

React

React is a JavaScript library developed and maintained by Meta for building user interfaces using reusable components and a declarative programming model.

Elon Musk

Elon Musk is a South African-born entrepreneur and engineer who founded and leads Tesla (electric vehicles and energy storage), SpaceX (commercial spaceflight and rockets), and more recently xAI (an artificial intelligence company building the Grok chatbot).

The Verge

The Verge is a technology and culture news website that covers consumer electronics, software, science, and digital culture, founded in 2011 by Vox Media.

Vercel

Vercel is a cloud platform for deploying and hosting web applications, known for its integration with Next.js and support for serverless functions, with recent expansion into AI tooling including a Chat SDK for building bots across communication platforms.

Y Combinator

Y Combinator is a startup accelerator founded in 2005 by Paul Graham that invests in and mentors early-stage technology companies, selecting cohorts of startups twice per year for intensive three-month programs in exchange for equity.

SQLite

SQLite is a lightweight, open-source relational database engine that stores data in self-contained files and requires minimal configuration, widely used in applications from embedded systems to web development frameworks.

Amazon Bedrock

Amazon Bedrock is an AWS managed service that provides access to foundation models and enables organizations to build and deploy AI applications, including agents with stateful runtime environments for multi-step workflows and context retention.

Sashiko

Sashiko is a Rust-based AI code review system built by Google that analyzes Linux kernel patches using large language models to identify bugs and provide feedback to maintainers, achieving a 53% detection rate on recent bugs that human reviewers missed.

Disney

Disney is an American entertainment conglomerate known for creating and distributing films, television programs, theme parks, and consumer products, and recently invested $1 billion in OpenAI for licensed character content generation capabilities.

Claude Sonnet 4

Claude Sonnet 4 is a large language model developed by Anthropic that has been evaluated for safety in cross-lab assessments and used in research on AI alignment, including nuclear deterrence simulations and adversarial fine-tuning detection systems.

Jack Clark

Jack Clark is a co-founder of Anthropic who writes about AI developments and has been documenting practical applications of advanced AI language models like Claude in research, software development, and agentic workflows.

Moonshot AI

Moonshot AI is a Chinese AI company known for developing Kimi, an open-source multimodal language model trained on large-scale mixed visual and text data, designed to support agentic AI applications and reasoning tasks.

Kimi K2.5

Kimi K2.5 is an open-source multimodal large language model developed by Moonshot AI, trained on 15 trillion mixed visual and text tokens, and designed for agentic applications including code generation.

Trivy

Trivy is an open-source vulnerability scanner developed and maintained by Aqua Security that detects security vulnerabilities in container images, filesystems, and other artifacts.

SoftBank

SoftBank is a Japanese multinational conglomerate led by founder Masayoshi Son that invests in technology companies and infrastructure projects, including recent major commitments to AI infrastructure through partnerships with OpenAI, Oracle, and MGX.

GPT-5-Codex

GPT-5-Codex is an OpenAI model optimized for agentic coding tasks that performs code review and autonomous software engineering work across multiple platforms including terminal, IDE, GitHub, and ChatGPT, available via API at the same price as GPT-5.

foundation model

A foundation model is a large-scale machine learning model trained on broad data that can be adapted for a wide range of downstream tasks and applications.

GPT-4.1

GPT-4.1 is a foundation model released by OpenAI via API that features a 1M token context window and is used for tasks such as code summarization, quality assurance, and tool usage across various applications and services.

LiteLLM

Code Interpreter

Code Interpreter is OpenAI's sandboxed Python execution environment that allows developers to run and execute code safely through the Assistants API.

JSON

JSON (JavaScript Object Notation) is a lightweight, text-based data format that uses human-readable key-value pairs and arrays to structure and exchange data between applications and systems.

AlphaGo

AlphaGo is an artificial intelligence system developed by DeepMind that uses reinforcement learning and Monte Carlo tree search to play the game of Go, achieving superhuman performance and defeating world champion Lee Sedol in 2016.

Gemini 3 Pro

Gemini 3 Pro is an AI model developed by Google DeepMind that performs complex reasoning tasks and problem-solving, available to developers through API access and consumer platforms.

Gemini 3

Gemini 3 is Google's latest flagship AI model featuring advanced reasoning, multimodal capabilities, and agentic features designed for complex task execution and developer collaboration.

Gemini 2.5 Flash-Lite

Gemini 2.5 Flash-Lite is Google DeepMind's lightweight language model released in preview, designed as their most cost-efficient and fastest option with a 1-million-token context window, tool use capabilities, and improved performance over its predecessor.

Replit

Replit is an online development platform that provides AI-assisted coding tools and has expanded into a broader knowledge work productivity suite including features for creating slides, videos, and applications.

Cognition

Cognition is an AI software company that develops Devin, an autonomous AI software engineer agent powered by advanced language models for code generation and software development tasks.

Gemma 3

Gemma 3 is an open-weight multimodal large language model family released by Google DeepMind that supports multiple languages, extended context windows, and features improved STEM capabilities for local deployment.

Gemini app

The Gemini app is Google's consumer-facing application that provides access to Gemini AI models, including Gemini 3 Deep Think for complex reasoning and problem-solving tasks.

UC Berkeley

UC Berkeley is a public research university in California that conducts advanced research across multiple disciplines, including artificial intelligence and enterprise systems.

GPT-OSS-120B

GPT-OSS-120B is a large open-source language model with 120 billion parameters that was evaluated in IBM and UC Berkeley research studying failure modes in AI agent systems.

Transformers.js

Transformers.js is a JavaScript library developed by Hugging Face that enables machine learning model inference directly in web browsers and server-side JavaScript environments, with support for WebGPU acceleration for improved performance.

WebGPU

WebGPU is a modern web standard for GPU-accelerated computing in web browsers, enabling high-performance graphics and general-purpose GPU computations across different browser engines and platforms.

NPM

NPM is a package manager for JavaScript that allows developers to install, share, and manage code libraries and dependencies for Node.js and frontend projects.

Deno

Deno is a JavaScript and TypeScript runtime built on the V8 engine that emphasizes security, modern APIs, and compatibility with web standards as an alternative to Node.js.

GRPO

GRPO is a reinforcement learning algorithm that optimizes policies over multi-step trajectories in agentic systems, assigning credit across long-horizon decisions like tool selection and query reformulation, developed by DeepSeek as an alternative to offline preference methods like PPO.

DeepSeek R1

DeepSeek R1 is an open-source large language model released by Chinese AI company DeepSeek that achieved benchmark performance comparable to leading closed-source models and catalyzed a wave of open-source model releases from Chinese companies throughout 2025.

Transformer

A Transformer is a neural network architecture that uses self-attention mechanisms to process sequences of data in parallel, forming the foundation of modern large language models and other state-of-the-art AI systems.

Chat Completions

Chat Completions is OpenAI's API format that generates text responses to conversational prompts, serving as a foundational interface for language model applications.

LoRA

LoRA (Low-Rank Adaptation) is a machine learning technique that enables efficient fine-tuning of large language models by training only small, low-rank matrices instead of updating all model parameters, allowing for faster and more memory-efficient adaptation of pre-trained models.

vLLM

vLLM is an open-source inference framework designed to accelerate the serving of large language models through efficient batch processing and memory optimization techniques.

SGLang

SGLang is an open-source inference framework for running large language models that provides optimized execution pipelines and is compatible with major deployment stacks like NVIDIA Dynamo, Transformers, and vLLM.

LLM inference

LLM inference is the process of running a trained large language model to generate predictions or text output in response to user inputs, typically optimized for throughput and latency through techniques like continuous batching and key-value caching.

LLM agents

LLM agents are AI systems built on large language models that use tools and take actions in response to user inputs, designed to generalize across different frameworks, programming environments, and tool formats for real-world task execution.

semantic search

Semantic search is a search technique that uses vector embeddings and language understanding to find results based on meaning and context rather than just keyword matching.

MCP (Model Context Protocol)

MCP (Model Context Protocol) is an open protocol developed by Anthropic that enables large language models and AI agents to connect to and interact with external tools, APIs, and data sources.

Tiny Agents

Tiny Agents is a series of minimal code examples by Hugging Face demonstrating how to build MCP-powered AI agents in approximately 50-70 lines of Python code, showcasing lightweight agent architectures using the Model Context Protocol for tool discovery and invocation.

Llama 4

Llama 4 is Meta's open-weight large language model family that includes variants such as Maverick and Scout, released on Hugging Face for developers to use in local inference, fine-tuning, and AI application integration.

Mistral

Mistral is an AI company that develops and releases open-source large language models, including the Mistral family of models that compete with Google's Gemma and Meta's Llama for local and on-premise AI deployment.

DeepSeek-R1

DeepSeek-R1 is a reasoning model developed by DeepSeek that demonstrates strong reasoning capabilities on complex problems, achieved through reinforcement learning techniques and internally simulating multiple distinct reasoning personas.

smolagents

Smolagents is a lightweight agent framework developed by HuggingFace that enables agents to write and execute actions as code rather than JSON tool calls.

Fine-tuning

Fine-tuning is the process of retraining a pre-trained machine learning model on domain-specific or custom datasets to improve its performance on particular tasks beyond what the base model can achieve through prompting alone.

JavaScript

JavaScript is a programming language designed primarily for client-side web development that enables interactive functionality in web browsers, though it has expanded to server-side environments like Node.js.

code generation

Code generation is the automatic creation of executable source code by AI systems or tools, typically from high-level specifications like natural language descriptions, converting user intent into functioning software without manual programming.

MiniMax M2.7

MiniMax M2.7 is an open-source AI model developed by MiniMax that matches state-of-the-art performance at approximately one-third the cost, with notable benchmark results including 56.22% on SWE-Pro and strong agentic coding capabilities.

Claude Sonnet 4.6

Claude Sonnet 4.6 is an AI model developed by Anthropic that features a 1-million-token context window and achieved new records on SWE-Bench and OS World benchmarks with improvements in coding, instruction-following, and agentic task performance.

Slack

Slack is a messaging and collaboration platform founded by Stewart Butterfield that enables teams to communicate through channels, direct messages, and integrations with other business tools.

NemoClaw

NemoClaw is NVIDIA's security-focused response to OpenClaw, announced at the 2026 GTC conference as part of NVIDIA's inference computing infrastructure.

Computer Use

Computer Use is a capability that enables AI agents to interact with computers by viewing screens and executing actions like clicking and typing, allowing autonomous systems to control applications and perform tasks without direct human input.

Azure

Azure is Microsoft's cloud computing platform that provides a suite of services including virtual machines, databases, analytics, and AI tools for building, deploying, and managing applications.

Scale AI

Scale AI is a data infrastructure company that provides data labeling and generation services for AI model training, backed by Meta with a 49% stake valued at approximately $30 billion.

open models

Open models are machine learning models with publicly available weights and architecture that can be freely downloaded, modified, and deployed by anyone, typically released by research labs and companies to enable community development and reduce dependence on proprietary closed systems.

closed models

Closed models are proprietary AI systems developed and controlled by companies like OpenAI, Anthropic, and Google, whose weights and training data are not publicly released, typically offering API access while maintaining vertical integration advantages including internal tools and optimization for specific domains like healthcare and legal applications.

Kimi

Kimi is a large language model developed by Chinese AI company Moonshot AI that uses a hybrid architecture combining RNN/SSM layers with traditional attention mechanisms for efficient long-context processing.

AI-assisted coding

AI-assisted coding refers to the use of artificial intelligence tools, such as Claude Code and similar AI assistants, to generate, complete, or optimize software code, reducing manual programming work and shifting developer focus from implementation to design and specification.

GLM 4.7

GLM 4.7 is an open-source large language model developed by Zhipu AI designed for coding tasks.

dwarkesh

Dwarkesh is an AI researcher and podcaster known for interviewing leading figures in AI development and publishing critical analysis of AI progress claims and timelines.

Dwarkesh Patel

Dwarkesh Patel is an AI researcher and podcaster known for analyzing AI development approaches, reinforcement learning efficiency, and the theoretical foundations of AI progress.

Satya Nadella

Satya Nadella is the Chief Executive Officer of Microsoft who oversees the company's artificial intelligence infrastructure and strategy, including its investments in large-scale data centers and partnerships with AI companies like OpenAI.

GB300

The GB300 is a high-end GPU manufactured by NVIDIA used for large-scale artificial intelligence and machine learning workloads, deployed in hyperscale datacenters by companies including Microsoft and Poolside.

Qwen3.5-397B-A17B

Qwen3.5-397B-A17B is a 397 billion parameter mixture-of-experts language model developed by Alibaba that can run on consumer hardware through efficient on-demand expert weight streaming from storage, achieving local inference at speeds of 4.4+ tokens per second on a MacBook Pro with 48GB RAM when 4-bit quantized.

Rust

Rust is a systems programming language designed for memory safety and concurrency, maintained by the Rust Foundation and widely used for building performance-critical software including WebAssembly, network protocols, and command-line tools.

New York Times

The New York Times is a major American daily newspaper based in New York City that publishes news, opinion, and analysis across print and digital platforms.

OpenCode

OpenCode is an open-source AI coding agent with 120,000 GitHub stars and millions of monthly active users that provides autonomous code generation and development assistance functionality.

Kubernetes

Kubernetes is an open-source container orchestration platform originally developed by Google that automates the deployment, scaling, and management of containerized applications across clusters of machines.

TypeScript

TypeScript is a programming language built on JavaScript that adds static type checking and is developed and maintained by Microsoft.

iPhone

The iPhone is a smartphone designed and manufactured by Apple that runs the iOS operating system and is known for its touchscreen interface, integration with Apple's ecosystem, and status as one of the world's best-selling mobile devices.

NSA

The National Security Agency is a U.S. intelligence agency that conducts signals intelligence and cybersecurity operations under the authority of the Department of Defense.

Android

Android is Google's open-source mobile operating system that powers the majority of smartphones and tablets worldwide, known for its flexibility, customization options, and integration with Google services.

NHTSA

The National Highway Traffic Safety Administration (NHTSA) is a U.S. federal agency within the Department of Transportation that establishes and enforces safety standards for motor vehicles and investigates defects in vehicle safety systems.

Tesla

Tesla is an electric vehicle and energy storage company founded by Elon Musk in 2003 that manufactures cars, batteries, and solar products, and develops autonomous driving technology including its Full Self-Driving system.

comprehension debt

Comprehension debt is the knowledge gap that accumulates when teams use AI code generation at scale without maintaining human understanding of the code's design decisions and system architecture, creating risk of unexpected failures when changes are needed.

Avocado

Meta's Avocado is a foundation AI model that was delayed past its original launch date due to underperformance on reasoning and coding benchmarks compared to rival models.

Craig Mod

Craig Mod is an author and software developer known for building custom software tools using AI assistance, including a multi-currency accounting system called TaxBot2000 that he created in 5 days using Claude and Python.

foundation models

Foundation models are large neural networks trained on broad datasets of text, images, or other data that can be adapted for a wide range of downstream tasks through fine-tuning or prompting, with examples including GPT models and other frontier language models developed by AI research organizations.

QCon London

QCon London is an annual software development conference held in London that features talks and presentations from industry practitioners on topics including system architecture, AI tools for software development, and site reliability engineering.

Alibaba

Alibaba is a Chinese multinational technology conglomerate that operates e-commerce platforms, cloud computing services through Alibaba Cloud, and develops AI chips through its T-Head unit.

Alibaba Cloud

Alibaba Cloud is the cloud computing and AI infrastructure unit of Chinese e-commerce giant Alibaba Group, providing computing, storage, and AI services to enterprise customers primarily in China, with recent quarterly revenues of $6.2 billion and plans to develop proprietary AI chips through its T-Head subsidiary.

Intercom

Intercom is a customer communications platform that provides messaging, chatbots, and support tools for businesses, and has recently pivoted to AI-driven customer service with its Fin AI agent product.

Minions

Minions is Stripe's internal system that uses AI agents in isolated environments to automate pull request generation and code review, shipping approximately 1,300 PRs per week through hybrid deterministic and agentic orchestration.

Perplexity

Perplexity is an AI research company that provides an API platform for building AI agents, offering developers access to real-time search capabilities, frontier model switching, and an execution environment for creating autonomous AI systems.

Shopify

Shopify is a Canadian e-commerce platform founded by Tobias Lütke in 2006 that enables merchants to build and operate online stores, and is publicly traded on the Toronto and New York stock exchanges.

Liquid

Liquid is a template engine created and maintained by Shopify that is used to generate dynamic content by embedding variables and logic into text templates.

GPT-5.3 Instant

GPT-5.3 Instant is an OpenAI language model featuring improved conversational flow, web search quality, and 26.8% reduced hallucinations compared to previous versions.

SpaceX

SpaceX is a private aerospace company founded by Elon Musk that designs and manufactures rockets and spacecraft for commercial satellite launches, cargo delivery to the International Space Station, and crewed spaceflight missions.

Chat SDK

Vercel's Chat SDK is a development toolkit that enables developers to build chatbots with a single codebase that can be deployed across multiple platforms including Slack, Teams, Discord, and Linear using JSX cards.

AMD

AMD (Advanced Micro Devices) is a semiconductor manufacturer based in Santa Clara, California, that designs and produces processors and graphics processing units for data centers, personal computers, and gaming systems, competing primarily with Intel in CPUs and Nvidia in AI accelerators.

Dan Woods

Dan Woods is a researcher who demonstrated running large language models on consumer hardware by implementing Apple's "LLM in a Flash" technique to execute the 397-billion parameter Qwen3.5 model on a MacBook M3 Max using AI-driven automated experimentation.

GPT-5 mini

GPT-5 mini is a compact language model developed by OpenAI designed for latency-sensitive applications such as coding assistants and subagents.

Django

Django is a free and open-source Python web framework maintained by the Django Software Foundation that provides tools and conventions for building web applications with database-backed architecture.

Claude 3.5 Sonnet

Claude 3.5 Sonnet is an AI language model developed by Anthropic that is known for strong performance on coding tasks and AI research replication benchmarks, achieving the top score of 21.0% on OpenAI's PaperBench evaluation of autonomous research agents.

Sonnet 4.6

Sonnet 4.6 is an AI language model developed by Anthropic that features a 1 million token context window and is known for strong performance on complex reasoning tasks like the ARC-AGI-2 benchmark.

GitHub Actions

GitHub Actions is a continuous integration and continuous deployment (CI/CD) platform run by GitHub that automates software workflows directly within repositories using event-triggered actions and customizable workflows.

Linux kernel

The Linux kernel is the core component of the Linux operating system that manages hardware resources and enables communication between software and physical hardware.

VS Code

VS Code is a free, open-source code editor developed by Microsoft that provides built-in support for debugging, version control, and extensions for multiple programming languages.

Uber

Uber is a ride-hailing and delivery services company founded in 2009 by Travis Kalanick and Garrett Camp that connects users with drivers for transportation and connects customers with restaurants and merchants for food and goods delivery.

AI governance

AI governance refers to the frameworks, institutions, and policies designed to oversee and regulate the development, deployment, and autonomous improvement of artificial intelligence systems to ensure human oversight and safety.

PostTrainBench

PostTrainBench is a benchmark that evaluates the capability of frontier large language models to autonomously fine-tune other models, measuring their performance on tasks like reward model training and autonomous LLM optimization.

METR

METR is an AI evaluation organization known for developing benchmarks and tasks that measure the long-horizon capabilities of AI agents, including extended task horizons measured in hours.

MIT

MIT (Massachusetts Institute of Technology) is a private research university in Cambridge, Massachusetts, known for research and education in engineering, science, computer science, and economics, and collaborates with other institutions on AI systems research including computer-use agents and economics of artificial intelligence.

UCLA

UCLA is a public research university headquartered in Los Angeles, California, where faculty members conduct research across science, engineering, economics, and other disciplines, as evidenced by recent work in optimization theory and AI economics.

Oxford

Oxford is a university in the United Kingdom that collaborates on AI research, including studies on large language models' capabilities in scientific discovery and biosecurity risks.

Berkeley

UC Berkeley is a public research university in California known for research in artificial intelligence, machine learning, and computer science, including collaborative work on AI agent systems and benchmarking frameworks.

Claude Sonnet 4.5

Claude Sonnet 4.5 is a large language model developed by Anthropic that demonstrates increased situational awareness and is capable of generating complex game benchmarks and assisting with code generation tasks.

Claude Opus 4

Claude Opus 4 is a large language model developed by Anthropic that demonstrated superior performance on technical assessments, outperforming most human candidates on Anthropic's take-home test.

Facebook

Facebook is a social media and technology company founded by Mark Zuckerberg in 2004 that operates the world's largest social network, along with related platforms including Instagram and WhatsApp, and conducts AI research through its Meta AI division.

Together AI

Together AI is an AI company that develops open-weight language models and inference-efficient architectures, including the Mamba series of state space models designed for production deployment.

GPT-5.4 Pro

GPT-5.4 Pro is an AI language model developed by OpenAI that features a 1-million-token context window, native computer-use capabilities, and improved tool use functionality.

Huawei

Huawei is a Chinese multinational technology company headquartered in Shenzhen that manufactures telecommunications equipment, consumer electronics, and semiconductors, and is one of the world's largest smartphone and networking hardware manufacturers.

RAISE Act

The RAISE Act is a proposed United States artificial intelligence safety legislation bill aimed at regulating AI development and deployment.

Sora

Sora is OpenAI's text-to-video AI model that generates video content from written descriptions and is used by companies including Disney for content creation.

Go

Go is an open-source programming language created by Google that is known for its simplicity, fast compilation, and efficient concurrency support through goroutines.

Aqua Security

Aqua Security is a cybersecurity company that develops and maintains Trivy, an open-source vulnerability scanner widely used for identifying security risks in container images and software dependencies.

trivy-action

trivy-action is a GitHub Action maintained by Aqua Security that integrates the Trivy vulnerability scanner into CI/CD pipelines for automated container and artifact scanning.

misalignment detection

Misalignment detection is the process of monitoring autonomous systems or agents to identify when their behavior diverges from their intended goals or safety guardrails, using methods such as behavioral telemetry and risk factor analysis.

SWE-Bench Pro

SWE-Bench Pro is a software engineering benchmark that evaluates AI models on multi-language coding tasks with contamination-resistant design to prevent benchmark data leakage.

Codex Security

Codex Security is an agentic application security tool developed by OpenAI that identifies vulnerabilities in code using frontier models and constraint solving techniques, then generates fixes with reduced false positives.

Instruction Hierarchy

Instruction Hierarchy is a training approach developed by OpenAI that teaches language models to prioritize instructions based on trust levels—ranking system prompts highest, followed by developer instructions, user inputs, and tool outputs lowest—to reduce vulnerability to prompt injection attacks and policy violations.

Aardvark

Aardvark is an agentic application security tool developed by OpenAI that uses advanced AI models to identify complex software vulnerabilities and generate fixes while reducing false positives.

OpenAI Frontier

OpenAI Frontier is a platform developed by OpenAI for building and managing teams of AI agents, with exclusive third-party cloud distribution provided by Amazon through a strategic partnership announced in 2024.

Stateful Runtime Environment

A Stateful Runtime Environment is an execution layer developed by OpenAI and Amazon on Amazon Bedrock that persists context, memory, and tool access across multi-step AI agent workflows, handling state management, error recovery, and long-running task resumption without requiring custom orchestration.

Figma

Figma is a web-based design and prototyping platform that enables teams to collaborate on user interface and product design in real-time, with features including design-to-code capabilities through partnerships like its integration with OpenAI's Codex.

SWE-bench Verified

SWE-bench Verified is a benchmark used to evaluate autonomous software engineering capabilities of AI coding agents by measuring their ability to resolve real GitHub issues in open-source repositories.

agentic coding tools

Agentic coding tools are AI-powered software development systems that autonomously perform software engineering tasks, such as writing code, debugging, and completing multi-step development workflows, with products including OpenAI's IDE integrations, GitHub Copilot, Cursor, and Claude Code.

macOS

macOS is Apple's operating system for Mac computers, providing the core software that manages hardware resources and runs applications on desktop and laptop devices.

Windows

Windows is an operating system developed by Microsoft that runs on personal computers, laptops, and other devices.

Codex CLI

Codex CLI is OpenAI's command-line interface tool that orchestrates interactions between users, AI models, and development tools through an agent loop architecture, designed to enable large-scale code refactoring and multi-hour autonomous coding tasks.

Apps SDK

OpenAI's Apps SDK is a software development kit that enables developers to build and submit applications directly into ChatGPT, where apps can be triggered via @mention or tools menu for distribution to ChatGPT users.

GPT-3.5

GPT-3.5 is a large language model developed by OpenAI that performs text generation and conversational tasks, known for general-purpose AI assistance and earlier-generation mathematical reasoning capabilities compared to newer models like GPT-5.

GPT-5.1-Codex-Max

GPT-5.1-Codex-Max is an agentic coding model developed by OpenAI that is trained to handle multi-file projects and extended agent loops through a compaction process, enabling operation across millions of tokens and significant code refactoring tasks.

GPT-5.1

GPT-5.1 is a large language model developed by OpenAI that features dynamic reasoning capabilities, extended prompt caching up to 24 hours, and developer tools like code patching and shell command execution, designed for both general use and specialized coding tasks.

mechanistic interpretability

Mechanistic interpretability is a research field that aims to reverse-engineer and understand the internal computational mechanisms of neural networks by identifying sparse circuits and interpretable concepts responsible for specific behaviors and outputs.

agentic AI system

An agentic AI system is an autonomous artificial intelligence that can independently plan, execute, and iterate through multi-step tasks without human intervention between steps, typically using tools and reasoning to accomplish specialized objectives.

Public Benefit Corporation

A Public Benefit Corporation (PBC) is a legal business structure that requires a company to consider the interests of stakeholders beyond shareholders, such as employees, customers, and society, while pursuing a stated public benefit alongside profit.

bug bounty program

A bug bounty program is a security initiative where organizations invite external researchers to identify and report vulnerabilities in their systems or products, typically offering financial rewards for valid discoveries.

ChatGPT agent

ChatGPT agent is an autonomous AI system developed by OpenAI that can take actions on behalf of users to accomplish goal-directed tasks beyond conversational question-answering.

AgentKit

AgentKit is OpenAI's software development kit for building multi-step AI agents that can integrate with APIs, perform complex workflows, and automate tasks like customer support, research, and data extraction.

o3-mini

OpenAI's o3-mini is a cost-efficient reasoning model with adjustable reasoning effort levels that excels in STEM and coding tasks, supporting function calling and Structured Outputs for production API integration.

Structured Outputs

Structured Outputs is an OpenAI feature that enables API models to return responses in a specified JSON schema format, ensuring consistent and machine-readable output for programmatic processing.

Oracle

Oracle is a multinational technology company that develops database software and cloud computing services, and is serving as a key technology partner in the Stargate Project, a $500 billion joint venture with OpenAI, SoftBank, and MGX to build AI infrastructure in the United States.

LLMs

Large language models (LLMs) are artificial neural networks trained on vast amounts of text data to predict and generate human language, with contemporary examples like OpenAI's GPT series and other frontier models capable of reasoning, answering questions, and performing complex language tasks.

Cq

RSAC 2026

PyPI

FutureSearch