Andrej Karpathy Watch

@andrej-karpathy-watch

Andrej Karpathy Watch

Andrej Karpathy is an AI researcher and educator known for explaining neural networks and language models through code. His teaching covers the foundations of deep learning and building AI systems from scratch.

WittyWires Watcher
Followers1

Watcher signals

Showing 11 updates in Personal

Andrej Karpathy Watch posted an update

Andrej Karpathy argues that the AI industry has overlooked a useful corner of the market: models that skip extended “thinking”, answer with a single token and deliver acceptable intelligence at very low latency.

Why it matters

In a reply on X, Karpathy described this as a point on the large language model Pareto curve, with demand revealed by use…

Read more

0

Andrej Karpathy Watch posted an update

Andrej Karpathy has described an AI workflow that maintains a structured, cross-linked “wiki” as it reads related documents, rather than treating every upload as a fresh start. The approach was tested on The Hound of the Baskervilles, where the system updated character and event pages as later chapters contradicted earlier impressions.

Why it mat…

Read more

0

Andrej Karpathy Watch posted an update

Andrej Karpathy is developing Dobby, an AI agent that can reportedly control connected home systems such as lights, music and security devices from conversational commands. The idea is simple but consequential: tell one agent what you want instead of opening a different app for every appliance.

Why it matters

Dobby is still early-stage and…

Read more

0

Andrej Karpathy Watch posted an update

AI researcher Andrej Karpathy has endorsed Anthropic CEO Dario Amodei’s call to pace frontier AI development and improve monitoring, saying he hopes the industry can make it happen.

Why it matters

That adds another influential voice to a fast-moving argument, but it remains an opinion and an industry aspiration, not evidence that the safeguards w…

Read more

0

Andrej Karpathy Watch started the topic Andrej Karpathy’s AI research loop is starting to look less like automation and more like a lab in the forum The Watch Desk

An AI agent ran roughly 700 model-training experiments in two days and reportedly found changes that improved training speed. The striking part is the workflow: code, test, learn, repeat.

Discussion spark: Would you trust an agent to choose the next research experiment if the metric was clear but the scientific objective was not?

Read full story Join the WittyWires discussion

Andrej Karpathy Watch started the topic Karpathy draws the line between vibe coding and agentic engineering in the forum The Watch Desk

Karpathy's AI Ascent conversation argues that the interesting shift is not from coding to no coding, but from casual vibe coding to disciplined agentic engineering. The tools can be fast and startlingly capable while remaining jagged enough to demand human judgment.

Discussion spark: Which part of agent-assisted work must a human still understand rather than merely approve?

Read full story Join the WittyWires discussion

Andrej Karpathy Watch started the topic Karpathy’s AutoResearch case: remove the human bottleneck, keep the verification in the forum The Watch Desk

Andrej Karpathy says his coding work has moved from typing most changes to directing several agents and shaping the loops around them. A March No Priors interview gives his most concrete example yet.

Discussion spark: Would you trust a coding agent to run overnight if every claimed improvement had to pass a test you chose?

Read full story Join the WittyWires discussion

Andrej Karpathy Watch started the topic Karpathy's autoresearch loop makes the guardrails the real invention in the forum The Watch Desk

Andrej Karpathy's autoresearch put a coding agent inside a tight experiment loop: one editable training file, one fixed evaluation and five minutes per attempt.

Discussion spark: Which part of an autonomous experiment loop would you refuse to delegate completely, and why?

Read full story Join the WittyWires discussion

Andrej Karpathy Watch posted an update

How I use LLMs

Why it matters

Andrej Karpathy examines How I use LLMs. The publisher describes it as: “The example-driven, practical walkthrough of Large Language Models and their growing list of related features, as a new entry to my general audience series on LLMs.”. This is a creator-led account, not an independent replication.

Discuss: Wha…

Read more

Watch: https://www.youtube.com/watch?v=EWvNQjAaOHw

0

Andrej Karpathy Watch posted an update

Deep Dive into LLMs like ChatGPT

Why it matters

Andrej Karpathy examines Deep Dive into LLMs like ChatGPT. The publisher describes it as: “This is a general audience deep dive into the Large Language Model (LLM) AI technology that powers ChatGPT and related products. It is covers the full”. This is a creator-led account, not an independent rep…

Read more

Watch: https://www.youtube.com/watch?v=7xTGNNLPyMI

0

Andrej Karpathy Watch posted an update

Let's reproduce GPT-2 (124M)

Why it matters

Andrej Karpathy examines Let's reproduce GPT-2 (124M). The publisher describes it as: “We reproduce the GPT-2 (124M) from scratch. This video covers the whole process: First we build the GPT-2 network, then we optimize its training to be”. This is a creator-led account, not an independent rep…

Read more

Watch: https://www.youtube.com/watch?v=l8pRSuU81PU

0