gpt-4

Sebastien Bubeck presents the Sparks of AGI paper at MIT, showing early experiments with GPT-4 across many domains and asking whether artificial intelligence has arrived. It includes the TikZ unicorn example.

A podcast episode in which GPT-4 red teamer Nathan Labenz describes using the model before release, finding it human-level but not human-like, and discussing its weaknesses and worrying capabilities. It closes with speculation about GPT-5.

OpenAI's livestream demo of GPT-4 with Greg Brockman, showing its capabilities and limitations. Highlights include writing code, turning a hand-drawn sketch into a website, and a tax-question example.

Yannic Kilcher explains how he trained a language model on three years of 4chan's /pol/ board and let it post on the site. He evaluates how good the model was and discusses the ethics and fallout.

Lex Fridman's long interview with OpenAI CEO Sam Altman about GPT-4, ChatGPT, and the future of AI. Topics include models as reasoning engines, AI risk, and consciousness.

Tom Gally, a former professional translator and University of Tokyo professor, discusses in March 2023 whether GPT-4 and similar models could translate literary and popular fiction. He gives his assessment of Japanese-to-English quality.

A demo of a Skyrim mod that uses Inworld AI to let players hold free-form conversations with NPCs. It previews LLM-driven characters in role-playing games.

Computerphile interviews AI safety researcher Rob Miles about ChatGPT: how it was trained with human feedback, why it tends to produce confident nonsense, and where it may or may not be dangerous.

Tom Gally, a former professional translator and University of Tokyo professor, discusses in April 2023 how GPT-4 and similar models may affect translators' careers. He draws on Japanese-English translation experience.

OpenAI cofounder Greg Brockman explains the design principles behind ChatGPT and demos unreleased plug-ins at TED 2023. In the follow-up, TED head Chris Anderson presses him on the timeline of development and the risks of releasing such a tool.

Rob Miles discusses GPT-3 on Computerphile, covering how scaling parameters kept improving results over GPT-2 and how the model nearly learned arithmetic without anyone knowing exactly how.

A short demo of a GPT-powered Linux shell that guesses the bash command you meant and sometimes hallucinates absurd results. The creator stresses that it is real and should not be trusted.

Lex Fridman interviews OpenAI CEO Sam Altman about GPT-4, GPT-5, Sora, the OpenAI board saga, Elon Musk, Ilya Sutskever, power and AGI. The conversation runs nearly two hours.

Andrej Karpathy's three-hour-plus general-audience deep dive into how ChatGPT-style LLMs are built, covering the full training stack from pretraining to post-training. It also gives mental models for their quirks, like hallucination, and how to use them well.

Andrej Karpathy builds a GPT from scratch in code, following Attention Is All You Need and GPT-2/3, in about two hours using PyTorch. He connects it to ChatGPT and assumes familiarity with his earlier makemore videos.

Andrej Karpathy builds the GPT tokenizer from scratch, implementing byte pair encoding along with encode and decode. He shows how many odd LLM behaviors trace back to tokenization, and covers Unicode byte encodings along the way.

Andrej Karpathy walks through how GPT assistants are trained, from tokenization and pretraining to supervised finetuning and RLHF, and gives practical prompting strategies and a view of the tooling ecosystem.

Alex O'Connor shows that ChatGPT cannot draw a wine glass filled to the brim, and ties the failure to David Hume's missing shade of blue. It asks what the failure says about how image models learn.

Computerphile video in which Mike Pound explains a University of Maryland paper on watermarking LLM output. The method biases token choices so generated text can later be detected statistically.

Jeremy Howard's 90-minute practical tour of language models. It covers how they work, GPT-4's capabilities and limits, using the OpenAI API, code writing, data analysis and fine-tuning.

Tom Scott describes trying ChatGPT to fix his Gmail labels and being unsettled by how capable it was. He argues that we cannot know where we are on the technology curve, comparing it to Napster.

Rob Miles explains "glitch" tokens, odd strings in large language model vocabularies that make the model produce gibberish. He covers how they likely arise from tokenizer training data, such as Reddit usernames, that never appeared in the model's own training.

UC Berkeley EECS colloquium by OpenAI's John Schulman on reinforcement learning from human feedback. He covers why RL helps with truthfulness, hallucination, and the limits of behavior cloning.

A step-by-step tutorial on making a Harry Potter by Balenciaga style AI video, using tools such as ChatGPT, Midjourney and Eleven Labs for script, images and voices.

Part two of a StrictlyVC interview with OpenAI CEO Sam Altman, focused on AI. He discusses regulation, commodification of AI, Google's reluctance to release its own models, and best- and worst-case futures.

Chubbyemu presents a case study of a man who followed AI health advice, replaced salt with sodium bromide and developed bromism with severe neurological effects. It is told as a dramatized medical case by a doctor.

3Blue1Brown's visual explanation of how transformers and GPT-style language models work, tracing data through tokens, embeddings and attention.

Tristan Harris and Aza Raskin of the Center for Humane Technology present why current AI capabilities pose catastrophic societal risks. They argue companies are racing to deploy without adequate safety, and that institutions must adapt. It was given before GPT-4's launch.

Andrej Karpathy builds a bigram character-level language model from scratch, introducing torch.Tensor and the framework of training, sampling and loss evaluation. It is the first step toward a GPT-style Transformer in his makemore series.

A short cartoon explaining how language models like ChatGPT work, from autocomplete and conditional probabilities to neural networks, gradient descent and backpropagation. It shows how networks act as universal function approximators.

Apple's WWDC 2024 keynote introducing Apple Intelligence along with iOS 18, iPadOS 18, macOS Sequoia, watchOS 11 and visionOS 2.

A short AI-generated parody video that imagines Harry Potter as a Balenciaga fashion campaign, made with Midjourney, ChatGPT and Eleven Labs voices. It became a viral meme.

A short Charlie Rose interview clip of Eric Schmidt discussing Google's direction for search. It is remembered for his remark that returning more than one result is a bug.

Andrej Karpathy gives a one-hour general-audience introduction to large language models, covering what they are, how they are trained, comparisons to operating systems, and security challenges such as jailbreaks.

3Blue1Brown's visual walkthrough of the attention mechanism inside transformers, covering self-attention, multiple heads and cross-attention. It is chapter 6 of his deep learning series.

Leo Dirac explains how transformer models have replaced LSTMs in NLP. He gives background from bag of words through RNNs and LSTMs, then a technical deep dive into multi-headed self-attention and positional encoding, with sample code.

Yannic Kilcher walks through the Attention Is All You Need paper, which introduced the Transformer, a sequence model based solely on attention without recurrence or convolutions. He reviews the architecture and the paper's translation results.

Sentdex demonstrates GAN Theft Auto, a GameGAN-based neural network that recreates GTA V so the network itself is the playable environment. He shows the results and releases the code.

Corridor Digital makes a roughly seven-minute anime short using Stable Diffusion and related AI tools, based on a technique they teach in a longer tutorial. It was an early showcase of AI-assisted animation.

Geoffrey Hinton's 2025 Royal Institution Discourse on what AI has learned from biological intelligence and how it might surpass it. He explains neural networks and argues that human language use resembles what backpropagation-trained networks do.