Sycophancy to subterfuge: Investigating reward tampering in language models AnthropicReward tampering is a specific, more troubling form of specification gaming. This is where a model has access to its own code and alters the training process itself, finding a way to “hack” the reinforcement system to increase its reward. This is like a person hacking […]
Category: Weekly
Weekly Web Harvest for 2024-05-26
Engineering for Slow InternetIf you’re an app developer reading this, can you tell me, off the top of your head, how your app behaves on a link with 40 kbps available bandwidth, 1,000 ms latency, occasional jitter of up to 2,000 ms, packet loss of 10%, and a complete 15-second connectivity dropout every few minutes? […]
Weekly Web Harvest for 2024-05-19
KCS v6 – Consortium for Service InnovationA framework for support documentation Coding my Handwriting — Amy Goodchild University Suspends Students for AI Homework Tool It Gave Them $10,000 Prize to MakeThe school “figured out that the Eightball program accesses the Canvas data through the Canvas user generated token, which is essentially users’ Emory credentials that […]
Weekly Web Harvest for 2024-05-12
Canvas LMS “Unpublish All” Hack | Dave EargleCanvas foils both goals, as of publication date of this post. (1) There is no built-in way to unpublish all assignments. (2) When a module is published, all items within it are auto-published. But when a module is unpublished, items within it remain published. :facepalm: h/t ohheybrian.com Model […]
Weekly Web Harvest for 2024-05-05
We Need To Rewild The InternetThe first magnificent bounty had not been the beginning of endless riches, but a one-off harvesting of millennia of soil wealth built up by biodiversity and symbiosis. Complexity was the goose that laid golden eggs, and she had been slaughtered. The story of German scientific forestry transmits a timeless truth: […]
Weekly Web Harvest for 2024-04-28
Tracker Beeper – Bert Hubert’s writingsMakes a little bit of noise any time your computer sends a packet to a tracker or a Google service, which excludes Google Cloud users. https://github.com/berthubert/googerteller h/t D’Arcy Norman and Alan Levine The Ultimate CSS Shapes Collection […]
Weekly Web Harvest for 2024-04-21
Isotope · Filter & sort magical layouts Filterizr | Create responsive galleries Hanover book nook creator’s ‘censored’ Girl Scout Gold Award“I would like to let you, the Board of Supervisors know, that you have bestowed upon me the greatest honor you could. Greater than that of any proclamation, in your censorship of my Gold Award,” […]
Weekly Web Harvest for 2024-04-14
t.30=Ovipares et Serpens t.1 (1788) – Histoire naturelle, ge?ne?rale et particulie?re – Biodiversity Heritage Library AI isn’t useless. But is it worth it?Some are surprised when they discover I don’t think blockchains are useless, either. Like so many technologies, blockchains are designed to prioritize a few specific characteristics (coordination among parties who don’t trust one […]
Weekly Web Harvest for 2024-03-31
AI bots hallucinate software packages and devs download them • The RegisterSeveral big businesses have published source code that incorporates a software package previously hallucinated by generative AI. Not only that but someone, having spotted this reoccurring hallucination, had turned that made-up dependency into a real one, which was subsequently downloaded and installed thousands of […]
Weekly Web Harvest for 2024-03-24
JavaScript Visualized – Promise Execution Models All The Way Down 3D DOM viewer, copy-paste this into your console to visualise the DOM topographically.A really different way of looking at the DOM h/t D’Arcy Stanford Prison Experiment: why famous psychology studies are now being torn apart – VoxMany of the classic show-stopping experiments in psychology have […]