The most-looked-up terms in the AI Dictionary this week, ranked by real page views.
Trending this week — most-looked-up terms, ranked by real page views.
meatproxy
A meat proxy is a person who forwards AI-generated text, code, or other output without reading, understanding, or validating it, acting only as a relay to the recipient.
Claude-ese · Opus-ese · token vomit
Claudish is the dense, jargon-heavy register Claude and similar coding agents default to — compressed metaphors, invented compound nouns, and objects doing action verbs.
Claude is Anthropic's family of large language models, built with an emphasis on helpfulness, harmlessness, and honesty as core training objectives.
CLM
A Contrastive Language Model (CLM) is a System One model that trains a state encoder and an action encoder with a contrastive loss, then picks the candidate action whose embedding best matches the current state.
GEV · WorldView globe · gods-eye-view
God's Eye View is an MIT-licensed browser globe by Bilawal Sidhu that fuses public aircraft, ship, satellite, and camera feeds with an optional OpenAI Realtime voice agent.
OpenJev · authored144
SemIf is an open-source decision-scoring project (formerly OpenJev) that evaluates runtime-defined yes/no, choice, and rubric questions by reading option logits from open models instead of generating JSON text.
Open Bot · CopilotKit OpenBot
OpenBot is CopilotKit's open-source, self-hosted platform for AI coworkers: each Bot gets its own computer, and a gateway decides and records every action before it runs.
OpenAI Decisions API · Luna Decisions API · Perplexity Decisions API
A closed-set decision surface: you send a question and finite answers (or a yes/no), and the API returns a selection with probabilities instead of prose.
GLiNER2.5 Decide
A 340M-parameter open-weight encoder from Fastino Labs that answers typed questions under user-defined rules and returns probabilities.
Diarization · Speaker Attribution
Speaker diarization is the step in a transcription pipeline that answers "who spoke when", splitting audio into segments and labelling each one with a distinct speaker identity.
InfoNCE
InfoNCE is a contrastive loss that trains a model to score a true positive pair higher than many negative pairs in the same batch, pulling matching items together in embedding space and pushing others apart.
RLCD · calibrated decision training
RLCD is a training method that optimizes a model to produce well-calibrated confidence scores on its decisions — so a model that says '70% confident' is actually correct about 70% of the time it says that.
SendFeedback tool · feedbackDrafts
Claude-drafted feedback is a Claude Code feature in which the agent writes its own bug report when something fails, queues it locally, and sends nothing until the user reviews and approves it.
agent skill · capability
A reusable, self-contained capability that an agent can invoke — such as "search the web," "write code," or "analyze an image" — typically defined as a tool, prompt template, or sub-workflow.
AI Model · Machine Learning Model
A model is the trained set of parameters and computations that maps inputs to outputs after learning from data.
Agent · Agents · AI agents · AI agent
An AI agent is a system that observes context, chooses actions, and uses their results to pursue a goal.
GitHub is a Microsoft-owned platform for hosting Git repositories, reviewing pull requests, and running CI/CD workflows.
Julia-1 · Supersonic Labs Julia 1
Julia 1 is Supersonic Labs' Apache-2.0 decision model, a 144.3 million parameter encoder that picks among 2 to 20 supplied answers and runs on a CPU or in a WebGPU browser.
TypeSafe AI Jev · TypeSafe Jev
Jev is TypeSafe AI's first System One Model, launched September 15, 2026 — it takes a state and typed questions and returns a Choice, Score, or Noul decision with calibrated confidence, instead of generating text.
OpenAI Dots · ChatGPT Dots · Dot
Dots are OpenAI's always-on ChatGPT agents, each powered by GPT-6 Astra with its own cloud computer, browser, and optional app plugins.
Cline AI · Cline Desktop · Claude Dev
Cline is an open-source coding agent available through IDE extensions, a command-line interface, a desktop app, and an SDK, with support for multiple hosted and open-weight models.
Fast Takeoff · Hard Takeoff
Foom is shorthand, popular in AI safety circles, for a hypothetical fast takeoff where an AI system rapidly self-improves from roughly human-level to vastly superhuman intelligence.
DSec
DeepSeek Elastic Compute (DSec) is DeepSeek's sandbox platform for agentic training and evaluation, unifying function-call sandboxes, containers, microVMs and full VMs behind one interface.
Pixel Leak · Glow PixelLeak
PixelLeak is Glow Labs' name for AI coding agents publishing internal screenshots to public GitHub so pull-request reviewers can see visual diffs.
pi-durable · @earendil-works/pi-durable
Pi Durable is Earendil's experimental MIT library for long-running agent sessions that stay alive outside the Pi terminal UI.
ART
Array-associated reverse transcriptases (ART) are a candidate enzyme system in bacteriophages: a reverse transcriptase, an accessory protein of unknown function, and a long array of repeating non-coding DNA resembling a CRISPR array.
Sign in With ChatGPT · Continue with ChatGPT · Login with ChatGPT
OpenAI's official way to sign into partner products with a ChatGPT account and, for Plus and Pro, spend included plan usage there.
Exploit Gym
ExploitGym is an 898-task cybersecurity benchmark that scores whether an AI agent can turn a known vulnerability into a working exploit and retrieve a flag.
pplx-decider · Perplexity decider · pplx decider 27b
Perplexity's Apache-2.0 ~26B multimodal decision fine-tune of Qwen3.8-27B, also served on a Decisions API at $0.04 per million input tokens.
OpenAI Rosalind Workbench · Rosalind NGS Analysis Workbench · rosalind-workbench
Rosalind Workbench is OpenAI's research-preview life-sciences environment, built on the GPT-Rosalind model and available inside the ChatGPT app and Codex.
