Refusal in Language Models Is Mediated by a Single Direction
(news.ycombinator.com)
31.
32.
34.
35.
How we test AI at ZDNET
(zdnet.com)
36.
The Download: a new Christian phone network, and debugging LLMs
(technologyreview.com)
39.
40.
After the illusion: what enterprise AI must become
(feeds.feedburner.com)
41.
Salesforce is crowdsourcing its AI roadmap — with customers
(techcrunch.com)
42.
The AI industry’s massive bet on transformer models may not be enough for true AGI
(feeds.feedburner.com)
43.
DAIMON Robotics Wants to Give Robot Hands a Sense of Touch
(spectrum.ieee.org)
44.
Granite 4.1: IBM's 8B Model Matching 32B MoE
(news.ycombinator.com)
45.
This AI knew the answers but didn’t understand the questions
(sciencedaily.com)
46.
The Shadow Glass
(news.ycombinator.com)
47.
Alignment whack-a-mole: Finetuning activates recall of copyrighted books in LLMs
(news.ycombinator.com)
48.
Finetuning Activates Verbatim Recall of Copyrighted Books in LLMs
(news.ycombinator.com)
49.
Functional Programmers need to take a look at Zig
(news.ycombinator.com)
50.
A grounded conceptual model for ownership types in Rust
(news.ycombinator.com)
51.
52.
Vera: a programming language designed for machines to write
(news.ycombinator.com)
53.
Why friendly AI chatbots might be less trustworthy
(feeds.bbci.co.uk)
54.
The friendlier the AI chatbot the more inaccurate it is, study suggests
(feeds.bbci.co.uk)
55.
Friendlier LLMs tell users what they want to hear — even when it is wrong
(feeds.nature.com)
56.
58.
A Primer on Bézier Curves – So What Makes a Bézier Curve?
(news.ycombinator.com)
59.
A statement from members of the Toki Pona community
(news.ycombinator.com)
60.
Show HN: I wrote a DOOM clone in my own programming language
(news.ycombinator.com)
Today's top topics:
google
openai
apple
google health
anthropic
samsung
nvidia
android authority
chatgpt
spacex