Andrej Karpathy

“Author of "Andrej Karpathy Discusses Using GPT-5.1 to Evaluate Past HN Predictions in His New Blog Post"”

“Author of "Andrej Karpathy Finds Python's Random.seed Removes the Sign Bit, Causing Negative and Positive Seeds"”

Large language models represent a fundamentally different form of intelligence than animal intelligence, shaped by commercial optimization rather than biological evolution, and understanding this distinction is critical for reasoning about AI systems.

“Author of "The space of intelligences is large" in Twitter/X”

AI represents a new computing paradigm (Software 2.0) where the most predictive feature of task automation is verifiability rather than algorithmic specificity, fundamentally reshaping which jobs and tasks are amenable to automation.

“Author of "AI Is a New Computing Paradigm" in Thread Reader App”

Moltbook, a social network for AI agents that went viral with 1.7 million bot accounts, was ultimately 'AI theater' rather than a genuine glimpse of autonomous agent futures, with much of its activity being spam, crypto scams, and human-generated performance.

“OpenAI cofounder and influential AI researcher who called Moltbook 'the most incredible sci-fi takeoff-adjacent thing' on X.”

AI startups should optimize inference spending as a growth investment and viral distribution mechanism rather than as a margin cost, treating compute as CAC replacement to drive product-led growth.

“Quoted describing Cursor's 'vibe coding' experience as a key innovation driver”

AI startups winning the market are agent labs that ship products solving real problems using existing frontier models, rather than model labs spending years and billions on foundation model R&D.

“Quoted on industry overshooting tooling relative to present capability and the importance of integration work”

Andrej Karpathy argues the 'decade of agents' is more accurate than 2025 being the 'year of agents,' citing insufficient groundwork in intelligence and context handling, with AI agent adoption likely peaking in impact by 2027-2028.

“Discusses AI agents, AGI timeline predictions, and historical shifts in AI research over 15 years of experience”

OpenAI's o1 model introduces test-time compute as a new paradigm for AI improvement by training models via reinforcement learning to perform chain of thought reasoning natively, enabling performance gains through longer inference time rather than just larger training runs.

“Joked on X that o1 refuses to solve the Riemann Hypothesis”

Large LLM context windows degrade model performance and increase costs significantly; effective context management ('context engineering') is essential to optimize both performance and API expenses.

“Ex-OpenAI researcher quoted defining context engineering as 'the delicate art and science of filling the context window with just the right information.'”

“Released autoresearch, a system where an AI agent iteratively modifies training code to improve model performance, achieving ~11% improvements on nanochat training.”

“Originally coined the term 'vibe coding' in an X post describing generating code by talking to AI without fully understanding it.”

“Built HackerTube in two hours using Claude Opus 5.5 to filter Hacker News comments and find recommended videos.”

“Built HackerTube, an alternative YouTube recommendation system using Claude, after being frustrated with YouTube's recommendations.”

“Researcher who built HackerTube video site in about two hours using Claude Opus 5.5 and let an agent run nanochat experiments autonomously.”

“Created HackerTube by using Claude to filter 20 years of Hacker News comments into a 6,454-video watchlist, demonstrating practical AI agent capabilities.”

“God-tier ML researcher at Anthropic who tweeted about the funnel of public AI capability awareness.”

“OpenAI co-founder who sketched AI's 'sharp funnel' and estimated only 5,000 people have internal access to frontier AI systems.”

“Estimated that only about 5,000 people have internal access to the most advanced AI systems, warning about widening capability gap.”

“Coined the term 'vibe coding' in February 2025 to describe AI-assisted software development where developers provide natural language descriptions to LLMs.”

“Estimated that just 0.2% of the general population truly grasps how powerful frontier LLMs have become and the knowledge gap is widening.”

“Stated that only about 5,000 people have access to frontier-grade AI systems, with about 6 billion people barely touching LLMs, highlighting a gap in AI capability access.”

“Discusses future directions for AI including output understanding, explainer videos, and LLM capabilities”

“Quoted on the LLM pareto optimal curve and inference optimization, referred to as 'Karpathy-san' on Jev.”

“AI researcher who demonstrated autonomous research where a teacher AI agent improves a student model.”

“Argued that achieving AGI would have little to no impact on GDP.”

“Researcher whose perspective on 'LLM Wiki' influenced the development of WikiSkill framework for AI agents.”

“Wrote canonical essay 'The Unreasonable Effectiveness of Recurrent Neural Networks' cited as essential resource for understanding RNN fundamentals.”

“Founding member of OpenAI, ran AI at Tesla, joined Anthropic to train Claude; wrote nanoGPT and llm.c.”

“Tested Opus 5 model with creative prompts including Three.js rendering of Lord of the Rings and shared perspectives on model capabilities.”

“Created animated short based on Lord of the Rings opening while testing Claude Opus 5”

“Wrote 2015 article on RNNs that inspired Eiso Kant to pivot startup to working on RNNs and later transformer models for code.”

“His autoresearch project is mentioned as an example of AI agents searching for better machine learning algorithms.”

“Poached by Elon Musk from OpenAI to work on Tesla's Autopilot technology.”

“AI researcher who created AutoResearch GitHub repo demonstrating loop-based AI workflows that run autonomous experiments; his method became known as the Karpathy Loop.”

“Published a pattern for building persistent structured wikis instead of using RAG systems, earning 17 million views and 13,000 GitHub stars.”

“Created nanoGPT minimalist GPT repository that was trained in space aboard an H100 on the complete works of Shakespeare.”

“AI researcher and investor who provided angel investment to Etched, an AI chip startup.”

“Built home automation AI agent called Dobby using OpenClaw framework to control lights, HVAC, security and other devices.”

“AI researcher praising Claude Tag as the third major paradigm shift in software UI/UX for LLM interaction.”

“Recently joined Anthropic to develop richer auto-research-style loops; known for work on making model architectures more composable.”

“Recently joined Anthropic to develop auto-research loops; known for research on making architectures composable.”

“Individual investor backing Inception Labs' $50 million funding round for Mercury 2 development.”

“Cited as example of researcher who spent significant time on unglamorous work, manually labeling ImageNet data.”

“His NanoChat autoresearch repo is referenced as a popular starting point for automated research systems in the Recursive automated AI research article.”

“Called Claude Fable 5 a major version bump deserving step change forward with state-of-the-art performance on benchmarks.”

“Co-founder of OpenAI who led AI at Tesla; spent 2 hours demonstrating how he actually uses AI day to day in simple conversational terms with the machine.”

“Former Tesla AI lead showing in a 2-hour tutorial how he uses AI day-to-day through simple natural language instructions.”

“Called Fable 5 a major-version-bump-level step change, emphasizing the qualitative jump in the underlying Mythos model with added safeguards.”

“Andrej Karpathy, whose work on neural networks and language models shaped AI practice, was among the first to receive an NVIDIA DGX Station system.”

“Released autoresearch project allowing AI agents to autonomously run research experiments and iterate independently.”

“Created Micrograd, a tiny autograd engine implementing backpropagation in Python for educational purposes.”

“Co-founder of OpenAI and former Tesla AI director who joined Anthropic to build a team focused on using Claude to accelerate pretraining research.”

“Influential AI researcher and former OpenAI founding member who has joined Anthropic, representing a significant hire for the AI safety company.”

“Joined Anthropic's pre-training team to work on using Claude to accelerate pre-training research.”

“Released a free video course teaching neural networks from scratch, including lectures on backpropagation, language modeling, and building GPT.”

“Joined Anthropic to work on automating the AI training pipeline with Claude.”

“Original creator of the LLM Wiki concept that the v2 post extends with lessons from agentmemory production experience.”

“OpenAI alum and former head of AI at Tesla who wrote on X that he's struggling to take full advantage of new programming tools and feels behind as a programmer despite their power.”

“Created Autoresearch, an AI system that emulates a human researcher in a loop to optimize tasks like model training and CI pipelines.”

“Clem referenced Andrej Karpathy's work on Auto Research as an example of agent limitations in AI tasks.”

“AI researcher noted for autonomous framework for model training; referenced regarding post-training infrastructure like Tinker.”

“Made observations about where AI coding tools go wrong that inspired a CLAUDE.md file that hit GitHub trending.”

“AI researcher who proposed the LLM Wiki pattern for building persistent knowledge bases using Claude.”

“Quoted discussing LLMs and dependency management in software engineering.”

“Noted that the LiteLLM supply chain attack was quickly discovered due to a bug causing a system crash.”

“Shared a GitHub Gist describing a pattern for building personal knowledge bases using LLMs that incrementally build and maintain a persistent wiki.”

“Demonstrates using LLMs to build personal knowledge bases for research topics with automatic organization, querying, and expansion capabilities.”

“Stated that AI Engineers will significantly outnumber ML engineers and can be successful without training models.”

“Built MenuGen AI agent; quoted describing the challenges of deploying modern apps with multiple services, APIs, and configurations.”

“AI researcher discussing speciation in AI models and the shift toward post-training as the frontier, quoted in discussion of Intercom's Apex model launch.”

“Promotes autoresearch where AI agents run ML experiments overnight, with humans writing program.md instead of training code.”

“Founding member and research scientist at OpenAI from 2015-2017 who contributed to foundational work in deep learning, computer vision and generative modeling; received first Dell Pro Max with NVIDIA GB300 DGX Station.”

“Creator of autoresearch project for autonomous model improvement with coding agents.”

“AI researcher discussing how AI research has historically been constrained by human researchers needing food, sleep, and meetings.”

“Published an open-source tool on GitHub for visually exploring Bureau of Labor Statistics data and analyzing AI exposure across 342 US occupations.”

“Released autoresearch project for autonomous AI agent-based LLM training research.”

“Mentioned for LLM OS architecture concepts that inspired Slate's thread design for agent orchestration.”

“Used autoresearch tool to squeeze 11% more speed out of his GPT-2 training by running 700 experiments over two days.”

“Highlighted Nemotron-ClimbMix as delivering the largest improvement on the Time-to-GPT-2 leaderboard.”

“Referenced for observation about Claude Code running on localhost versus cloud deployments.”

“Mentioned for creating autoresearch tool that ran 126 experiments overnight on single GPU, exploring LLM training hyperparameters.”

“Released AutoResearch project that gives AI agents autonomous control over LLM training to experiment overnight.”

“Posted on X about how programming changed dramatically in December due to AI coding agents becoming significantly more capable.”

“AI researcher who coined the term 'vibe coding' in early 2025, which has become a central cultural phenomenon in AI-assisted development.”

SkyPilot v0.11 provides a unified interface for AI teams to run, manage, and scale workloads across multiple cloud providers and on-premises infrastructure with advanced scheduling, cost optimization, and enterprise-grade features.

“Created nanochat, described as the best ChatGPT that $100 can buy, which can be trained and served using SkyPilot.”

Effective agent design patterns center on context management as a finite resource, with successful agents using computer access, hierarchical action spaces, and deliberately limited tool sets to optimize long-running autonomous capabilities.

“Framed the need for context engineering as 'the delicate art and science of filling the context window with just the right information.'”

NVIDIA's aggressive push for 16-Hi HBM production by Q4 2026 represents a critical infrastructure bottleneck that will determine AI's future, as solving memory bandwidth constraints enables dramatically more capable AI systems.

“Former Tesla Autopilot builder who expressed feeling behind as a programmer due to rapid AI advancements in inference capabilities.”

Reinforcement Learning from Verifiable Rewards (RLVR) emerged as a major new training paradigm in 2025, enabling LLMs to develop complex reasoning strategies and significantly improving capability-per-compute efficiency.

“Authored a 2025 year-in-review article analyzing paradigm changes in LLMs, including the emergence of Reinforcement Learning from Verifiable Rewards (RLVR).”

Reinforcement Learning has evolved from early psychology and mathematical foundations through the Deep RL revolution of the 2010s to modern applications like RLHF and GRPO in LLMs, with future potential still largely untapped.

“Discussed the future of reinforcement learning and stated that RL is terrible but everything before it was much worse; used FP16 precision format for nanochat.”

Leading AI researchers including Ilya Sutskever and Andrej Karpathy are revising AGI timelines and expressing skepticism about transformer scaling limits and LLM agent capabilities, raising doubts about current AI business model sustainability.

“Previously featured on a podcast discussing AI development and the future of AI systems.”

LLM Council is an open-source local web application that aggregates multiple LLMs into a council structure where models review each other's responses and a chairman model produces a final synthesized answer.

“Creator of the LLM Council project, a local web app that aggregates multiple LLMs for comparative analysis”

Richard Sutton's 'Bitter Lesson' argues that general search and compute methods outperform domain expertise, but engineers remain critical for properly framing search problems—as exemplified by tinygrad's approach to hardware optimization.

“Created micrograd which inspired George Hotz to build tinygrad, and appeared on Dwarkesh Patel's podcast calling the 2020s the 'decade of agents'.”

“Discussed in wide-ranging conversation covering AGI economic impact, reinforcement learning limits, self-driving research, and education reimagining.”

AGI timelines are realistically 5-10 years away but require significant work in integration, physical world sensors, safety, and research beyond current LLM capabilities, despite recent progress.

“Discussed AGI timelines, predicting a 'decade of agents' with 5-10X more pessimistic timelines than typical SF AI predictions, and wrote thoughts on BERT diffusion.”

Nanochat enables training GPT-2 capability models in 3 hours for ~$72 on a single 8XH100 GPU node, compared to $43,000 in 2019, with a complete open-source stack covering tokenization, pretraining, fine-tuning, and inference.

“Introduced nanochat, an open-source experimental harness for training LLMs that covers tokenization, pretraining, finetuning, evaluation, inference, and chat UI.”

Vibe coding—using AI as a high-level programming language—works initially but degrades as project context grows, requiring lightweight context management tools like Shallot for sustainable scaling.

“Originally coined the term 'vibe coding' in a viral tweet describing giving in to the vibes and forgetting the code even exists.”

Kairos, an AI-powered reading companion, enables fundamentally new ways to read by integrating AI directly into the reading experience, allowing readers to actively probe and discuss texts rather than passively consume them.

“Provided inspiration through an X post that led to the development of Kairos AI reading tool.”

Silicon Valley startups and major AI labs are racing to build reinforcement learning environments for training AI agents, with Anthropic reportedly planning to spend over $1 billion on the technology.

“Backer of Prime Intellect startup focused on RL environments”

Context engineering—properly balancing intent, environment, and conversation data—is the critical differentiator between mediocre and exceptional AI code review agents.

“Commented that good context engineering is what sets AI apps apart in response to discussions about context engineering for AI.”