Exposing biases, moods, personalities, and abstract concepts hidden in large language models
A new method developed at MIT could root out vulnerabilities and improve LLM safety and performance.
A new method developed at MIT could root out vulnerabilities and improve LLM safety and performance.
The last time I wrote a blog post for Three Geeks was in 2021. It was the height of COVID, and I had just hosted my firs…
By minimizing the need to drive around looking for a parking spot, this technique can save drivers up to 35 minutes — an…
Preface This essay argues that rational people don’t have goals, and that rational AIs shouldn’t have goal…
(How to Create a Claude Skill or Plugin for Law and Use It in Claude Cowork) On February 3, 2026, a single product annou…
This episode is recorded live, and is best enjoyed on YouTube. Watch the episode here . While Bob is visiting San Franci…
The context of long-term conversations can cause an LLM to begin mirroring the user’s viewpoints, possibly reducing accu…
The billable hour has survived a lot of threats, from alternative fee arrangements to client procurement, but this episo…
This story has been retracted
Project AI Evidence will connect governments, tech companies, and nonprofits with world-class economists at MIT and acro…
OpenAI's new GPT‑5.3‑Codex‑Spark is 15 times faster at coding than its predecessor.
Associate Professor Rafael Gómez-Bombarelli has spent his career applying AI to improve scientific discovery. Now he bel…
Zoë Hitzig resigned on the same day OpenAI began testing ads in its chatbot.
Many employees don’t struggle with too much information — they struggle with information that lives in to…
Driven by overuse and misuse of antibiotics, drug-resistant infections are on the rise, while development of new antibac…
Welcome to Remarkable People. We're on a mission to make you remarkable. Helping me in this episode is Suzy Welch. Suzy …
Opening a new window on the brainstem, a new tool reliably and finely resolves distinct nerve bundles in live diffusion …
MIT Sports Lab researchers are applying AI technologies to help figure skaters improve. They also have thoughts on wheth…
I have just posted a trio of new research white papers to SSRN. These represent the latest output from the Kennedy Idea …
An investigation into why serious AI work depends less on clever prompts and more on defending invariants, boundaries, a…
A fresh Anthropic announcement set off a week of market jitters and existential questions: what happens when the big mod…
Removing just a tiny fraction of the crowdsourced data that informs online ranking platforms can significantly change th…
The $20,000 experiment compiled a Linux kernel but needed deep human management.
MIT faculty join The Curiosity Desk to discuss football, math, Olympic figure skating, AI and the quest to cure ovarian …
Incident is at least the third time the exchange has been targeted by thieves.
Claude Opus 4.6 and OpenAI Frontier pitch a future of supervising AI agents.
EnCompass executes AI agent programs by backtracking and making multiple attempts, finding the best set of outputs gener…
Copilot connectors extend Microsoft 365 Copilot and Copilot Search to your external business systems. They bring line of…
Sam Altman calls AI competitor "dishonest" and "authoritarian" in lengthy post on X.
Publishers are rolling out more aggressive defenses.