Resources

Foundations

The explainers that make everything after them click.

1. 1.Neural Networks: Zero to Hero (opens in new tab) · Andrej Karpathy. Builds a neural net, then a GPT, from scratch in code. The single best on-ramp there is.
2. 2.The Illustrated Transformer (opens in new tab) · Jay Alammar. The picture-first walkthrough of attention that most people first understood transformers from.
3. 3.The Unreasonable Effectiveness of Recurrent Neural Networks (opens in new tab) · Andrej Karpathy, 2015. The post that made a generation of engineers fall for sequence models.
4. 4.Understanding LSTM Networks (opens in new tab) · Christopher Olah, 2015. The canonical, diagram-driven explanation of how gated recurrent memory works.

Landmark Papers

Must reads.

1. 5.Attention Is All You Need (opens in new tab) · Ashish Vaswani et al., 2017. Introduced the transformer. The paper the modern era is built on.
2. 6.Efficient Estimation of Word Representations in Vector Space (opens in new tab) · Tomas Mikolov et al. (Google), 2013. Word2Vec. Showed words could be turned into meaningful vectors, the root of embeddings.
3. 7.Improving Language Understanding by Generative Pre-Training (opens in new tab) · Alec Radford et al. (OpenAI), 2018. GPT-1. First showed pre-train then fine-tune on a transformer.
4. 8.BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding (opens in new tab) · Jacob Devlin et al. (Google), 2018. The bidirectional encoder that dominated NLP benchmarks before the GPT line pulled ahead.
5. 9.Language Models are Unsupervised Multitask Learners (opens in new tab) · Alec Radford et al. (OpenAI), 2019. GPT-2. Scaled the idea up and hinted at zero-shot ability.
6. 10.Language Models are Few-Shot Learners (opens in new tab) · Tom Brown et al. (OpenAI), 2020. GPT-3. The one that showed scale alone unlocks new behaviour.
7. 11.ImageNet Classification with Deep Convolutional Neural Networks (opens in new tab) · Alex Krizhevsky, Ilya Sutskever, Geoffrey Hinton, 2012. AlexNet. The result that set off the modern deep-learning era.
8. 12.Sequence to Sequence Learning with Neural Networks (opens in new tab) · Ilya Sutskever, Oriol Vinyals, Quoc Le, 2014. Seq2Seq. The encoder-decoder idea that set the path toward modern language models.
9. 13.Deep Residual Learning for Image Recognition (opens in new tab) · Kaiming He et al., 2015. ResNet. Made truly deep networks trainable and reset the field.
10. 14.Training Language Models to Follow Instructions with Human Feedback (opens in new tab) · Long Ouyang et al. (OpenAI), 2022. InstructGPT. The RLHF recipe that turned raw models into assistants.

Courses

If you want to go from reading to doing.

1. 15.Practical Deep Learning for Coders (opens in new tab) · fast.ai. Top-down and hands-on. Ship a working model in the first lesson.
2. 16.CS231n: Deep Learning for Computer Vision (opens in new tab) · Stanford. The course Karpathy designed. Still the gold standard for the fundamentals.
3. 17.CS224n: NLP with Deep Learning (opens in new tab) · Stanford. The rigorous path through language models, from embeddings up.
4. 18.Machine Learning Specialization (opens in new tab) · Andrew Ng · DeepLearning.AI. The gentlest on-ramp to the maths and intuitions underneath it all.

Watch & Listen

Talks, channels, and podcasts. Sit back and let someone explain it.

1. 19.Neural Networks (series) (opens in new tab) · 3Blue1Brown. The most beautiful visual intuition for what a network actually does.
2. 20.[1hr Talk] Intro to Large Language Models (opens in new tab) · Andrej Karpathy, 2023. The clearest one-hour mental model of how LLMs work and where they go.
3. 21.Let's build GPT: from scratch, in code (opens in new tab) · Andrej Karpathy. Watch a GPT get built line by line. Everything demystifies at once.
4. 22.AI Explained (opens in new tab) · YouTube channel. Calm, careful breakdowns of every major model release and benchmark.
5. 23.Theo, t3.gg (opens in new tab) · YouTube channel. Opinionated, fast-moving takes on AI coding tools and how they change the work.
6. 24.Prof G Markets (opens in new tab) · Scott Galloway · YouTube channel. The business and money side of the AI boom, markets, valuations, and who's winning.
7. 25.YC Paper Club (opens in new tab) · Y Combinator. Researchers and founders break down the latest AI papers. A great way to keep up.
8. 26.AI Engineer (opens in new tab) · swyx & the AI Engineer team. Talks from the AI Engineer conferences. Practitioners showing what they actually shipped.
9. 27.Dwarkesh Podcast (opens in new tab) · Dwarkesh Patel. Deeply researched, unusually hard questions for the people building frontier AI.
10. 28.Latent Space (opens in new tab) · Shawn Wang (swyx) & Alessio Fanelli. The AI engineer's trade publication in audio form. Real implementation, not hype.
11. 29.No Priors (opens in new tab) · Sarah Guo & Elad Gil. Frontier AI from the investor and product-builder angle. Where the field is heading.
12. 30.Lex Fridman Podcast (opens in new tab) · Lex Fridman. Long, unhurried conversations with many of the field's defining figures.
13. 31.The 80,000 Hours Podcast (opens in new tab) · Rob Wiblin & team. In-depth interviews on AI safety and the field's biggest risks, now on video.
14. 32.TLDR News (opens in new tab) · TLDR News. Concise, plain explainers on the day's politics, tech, and economics.
15. 33.TLDR News Global (opens in new tab) · TLDR News. The global-affairs channel, for stories beyond the UK and US.
16. 34.TLDR News EU (opens in new tab) · TLDR News. The European politics and policy channel from the same team.

Tools & Products

The AI products worth actually using, across code, writing, and media.

1. 35.Cursor (opens in new tab) · Anysphere. The AI-native code editor most developers reach for first.
2. 36.Claude Code (opens in new tab) · Anthropic. Anthropic's terminal coding agent, the pick for serious work.
3. 37.Codex (opens in new tab) · OpenAI. OpenAI's coding agent, in the terminal, IDE, and the cloud.
4. 38.Devin (opens in new tab) · Cognition. Autonomous software engineer that plans and ships code on its own.
5. 39.v0 (opens in new tab) · Vercel. Generates working web apps and UI from a prompt.
6. 40.Lovable (opens in new tab) · Lovable. Prompt-to-app builder for shipping full prototypes fast.
7. 41.Base44 (opens in new tab) · Wix. No-code app builder that grew to a Wix acquisition in six months.
8. 42.Perplexity (opens in new tab) · Perplexity AI. AI answer engine that cites its sources as it searches.
9. 43.Exa (opens in new tab) · Exa AI. Search API that gives AI agents real-time, structured web data.
10. 44.Granola (opens in new tab) · Granola. AI meeting notes that transcribe your calls without a bot joining.
11. 45.NotebookLM (opens in new tab) · Google. Grounds answers in your own documents, and turns them into audio.
12. 46.Whisper (opens in new tab) · OpenAI. The open speech-to-text model much of the ecosystem is built on.
13. 47.ElevenLabs (opens in new tab) · ElevenLabs. The leading text-to-speech and voice-cloning platform.
14. 48.Suno (opens in new tab) · Suno. Generates complete songs, vocals and all, from a text prompt.
15. 49.Runway (opens in new tab) · Runway. AI video generation and editing tools for creative work.
16. 50.Midjourney (opens in new tab) · Midjourney. Still the benchmark for aesthetic AI image generation.

Evals & Dev Tooling

Where to see who's actually ahead, and what to build and monitor with.

1. 51.LMArena (opens in new tab) · LMArena (ex-LMSYS). Crowdsourced head-to-head voting that ranks models by human preference.
2. 52.Artificial Analysis (opens in new tab) · Artificial Analysis. Independent leaderboards comparing models on intelligence, speed, and price.
3. 53.SWE-bench (opens in new tab) · SWE-bench. The standard test of whether agents can fix real GitHub issues.
4. 54.Epoch AI (opens in new tab) · Epoch AI. Research and data on AI trends, compute, and where the frontier is heading.
5. 55.Hugging Face (opens in new tab) · Hugging Face. The hub for open models and datasets, and home to many leaderboards.
6. 56.LangChain (opens in new tab) · LangChain. Agent framework (LangGraph) plus LangSmith for tracing and evals.
7. 57.Langfuse (opens in new tab) · Langfuse. Open-source observability and evals for LLM apps. Now part of ClickHouse.
8. 58.Weights & Biases (opens in new tab) · Weights & Biases. Experiment tracking and model tooling, widely used across ML teams.

People to Follow

Track these and you'll never fall too far behind. Follow a few on X and you'll feel the field move in real time.

1. 59.Andrej Karpathy (opens in new tab) · @karpathy · Anthropic (ex-OpenAI, ex-Tesla). The field's best teacher. Read and watch everything he makes.
2. 60.Sam Altman (opens in new tab) · @sama · OpenAI. OpenAI's CEO. Sets expectations for the field one post at a time.
3. 61.Elon Musk (opens in new tab) · @elonmusk · xAI. Runs xAI, and X itself. Unavoidable, for better and worse.
4. 62.Dario Amodei (opens in new tab) · @DarioAmodei · Anthropic. Anthropic's CEO. His long essays on scaling and safety set the terms of the debate.
5. 63.Demis Hassabis (opens in new tab) · @demishassabis · Google DeepMind. Frontier research and the science applications, from AlphaFold to Gemini.
6. 64.Greg Brockman (opens in new tab) · @gdb · OpenAI. OpenAI's president. Infrastructure scale, demos, and engineering culture.
7. 65.Yann LeCun (opens in new tab) · @ylecun · Turing laureate. Turing laureate and the field's most reliable contrarian on LLMs.
8. 66.Alexandr Wang (opens in new tab) · @alexandr_wang · Meta. Founded Scale AI, now Meta's Chief AI Officer leading its superintelligence push.
9. 67.Jack Clark (opens in new tab) · Anthropic. His Import AI newsletter is the most trusted weekly read on where the field is going.
10. 68.Lilian Weng (opens in new tab) · Ex-OpenAI. Deep, survey-grade posts that become the reference on their topic.
11. 69.Simon Willison (opens in new tab) · Independent. The most reliable running commentary on what's actually new and useful.
12. 70.Christopher Olah (opens in new tab) · Anthropic. Makes the inside of neural networks legible. Pioneer of interpretability.

Frontier AI Labs

Who's building the models, grouped by region and in no particular order. As of Aug 2026.

Hyperscalers, Neoclouds & Compute

The clouds and chips the models actually run on.