Posts that commenters keep linking to, ranked by how many different people have cited them in Hacker News comments. 1,015 posts.
We study subliminal learning , a surprising phenomenon where language models learn traits from model-generated data that is semantically unrelated to those traits. For example, a "student" model learns to prefer owls when trained on sequences of numbers generated by a "teacher" model that prefers owls. This same…
All modern language models sometimes act like they have emotions. What’s behind these behaviors? Our interpretability team investigates.
This is going to be a two-part post—one on why machine intelligence is something we should be afraid of, and one on what we should do about it. If you’re already afraid of machine intelligence,...
AI folklore tells a story about a neural network trained to detect tanks which instead learned to detect time of day; investigating, this probably never happened.
Recent generations of frontier language models have introduced Large Reasoning Models (LRMs) that generate detailed thinking processes…
Stay up to speed on the rapid advancement of AI technology and the benefits it offers to humanity.
Our measurements of the time horizons for public frontier language models.
Posted by Brendan McMahan and Daniel Ramage, Research ScientistsStandard machine learning approaches require centralizing the training data on one ...
Learn how code execution with the Model Context Protocol enables agents to handle more tools while using fewer tokens, reducing context overhead by up to 98.7%.
Chuan Li, PhD reviews GPT-3, the new NLP model from OpenAI. The technical overview covers how GPT-3 was trained, GPT-2 vs. GPT-3, and GPT-3 performance.
My smartest friends have bananas arguments about LLM coding.
Summary: An AI agent of unknown ownership autonomously wrote and published a personalized hit piece about me after I rejected its code, attempting to damage my reputation and shame me into acceptin…
An OpenAI model solved the 80-year-old unit distance problem, disproving a major conjecture in discrete geometry and marking a milestone in AI-driven mathematics.
The recent innovations in the AI space, most notably those such as GPT-4, obviously have far-reaching implications for society, ranging from the utopian eliminating of drudgery, to the dystopian damage to the livelihood of artists in a capitalist society, to existential threats to humanity itself.
By Ljubisa Bajic Many believe AI is the real deal. In narrow domains, it already surpasses human performance. Used well, it is an unprecedented amplifier of human ingenuity and productivity. Its widespread adoption is hindered by two key barriers: high latency and astronomical cost. Interactions with language models…
Wolfram plugin gives computationally accurate answers to ChatGPT queries using Wolfram's data knowledgebase and language. Custom visualizations are given as well. Stephen Wolfram explains how it works.
Much has been said about the effects that AI will have on software development, but there is an angle I haven’t seen talked about: I believe that AI will bring formal verification, which for decades has been a bit of a fringe pursuit, into the software engineering mainstream.
Machine learning provides the underlying oomph to many of Google’s most-loved applications. In fact, more than 100 teams are currently using machine learning at Google today, from Street View, to Inbox Smart Reply, to voice search.
I'm calm. I'm calm. I promise.
Stay up to speed on the rapid advancement of AI technology and the benefits it offers to humanity.
I procrastinated a deep dive into transformers for a few years. Finally the discomfort of not knowing what makes them tick grew too great for me. Here is that dive.
Claude Opus 4.6 found 22 Firefox vulnerabilities in two weeks, 14 rated high-severity, with proofs-of-concept and candidate patches for Mozilla.
We have rolled back last week’s GPT‑4o update in ChatGPT so people are now using an earlier version with more balanced behavior. The update we removed was overly flattering or agreeable—often described as sycophantic.
ML, Biotech, Hardware, and Coordination Problems. Sometimes I write about hard problems and how to solve them.
In a recent study in Nature, we introduce a form of memory-augmented neural network called a differentiable neural computer, and show that it can learn to use its memory to answer questions about complex, structured data, including artificially generated stories, family trees, and even a map of the London Underground…
Repository for personal website
Large Language Models (LLMs) are typically presumed to process context uniformly—that is, the model should handle the 10,000th token just as reliably as the 100th. However, in practice, this assumption does not hold. We observe that model performance varies significantly as input length changes, even on simple tasks…
June 24, 2018 note: If you want to cite an example from the post, please cite the paper which that example came from. If you want to cite the post as a whole, you can use the following BibTeX:
Musings of a Computer Scientist.
