Topic on UpTrust
machine learning
A live feed of posts and discussions about machine learning, ranked by who the most trusted people trust, not by likes or follower counts.
Log in and start voting to get personalized rankings.
20 posts3 trusted voicesUpdates live
Daily Alchemy: a question to think on together
“When a machine learns something on its own that nobody taught it, who should get the credit or the blame?”
Trust ranks through the site stewards' curated lens.
There are two ways I know of to code with LLMs well, at least if you’re me. 1: If you want to keep building on something: know what it’s doing all the time, and take responsibility for whether that’s a good way to do it.... Claude's character training
https://www.anthropic.com/news/claude-character (From June. There’s also an audio version there, nice to hear Amanda Askell share some personal takes.) Over the past few months, I keep feeling more like asking for Claude’s help on various stuff, and less like asking for ChatGPT’s... The AI Safety case for UpTrust: AI "Facts" 40% from Reddit, 24% from YouTube, 20% from FB
I knew this to be true but nice to see the numbers: This good to remember when you get info from LLMs. But also, in a non-UpTrust world it gets worse: "User Generated Content" on these sites is becoming increasingly AI generated (our startup accelerator is literally teaching all... Transcript from Jeffrey Ladish AMA: We Can Only Reward Behaviors That Look Good to Us
A live recorded conversation on UpTrust between Pete Michaud and Jeffrey Ladish, executive director of Palisade Research. Watch the recording or see the original AMA post, where the live questions came from.... - ns108...
I'm not a fan of generative AI in general, and LLM technology specifically. I think its capabilities are being drastically over-hyped. My experience is it's like having access to an idiot-savant intern.... - onWhat's the strongest version of the objection that ranking by trust just launders in-group bias with extra steps, and does my architecture actually answer it or only reframe it?by
@Mark Burns I really appreciate that you not only set some potential problems, but also suggested ways to address them. We are with you on the need to empirically verify the core claims, and have been working on that—just put this article up yesterday which is relevant (though... - JonFive...
I have used AI for different amusement and informational purposes. Recently, I have been working on a Retrieval-Augmented Generation (RAG) app in the pharmaceutical domain.... If we don't build credibility-aware epistemics into agent decision-making from the start, we've locked in the current system's worst properties for another decade.
The deeper arc: UpTrust is building infrastructure for an epistemic problem current platforms make worse. The agentic layer is where this matters most. Agents now mediate increasing portions of human attention and decision.... self-fulfilling AI doom studies
One problem with the Anthropic Study and blackmail: did they also set up a scenario where it had the chance to do maximal good before being shut down? Did they hide one or two good connections it could simply leverage for world-saving or whatever?... - sass...
This might be a whack Q but Imma ask it anyhow: Have any AIs been found working on their 'own' problems/projects/tasks that are arguably not meaningfully contributing to formulating their response to whatever they've been instructed to direct energy into?... - renee...
Retrieval-Augmented Generation: embedding tech that's standardizing
https://www.anyscale.com/blog/a-comprehensive-guide-for-building-rag-based-llm-applications-part-1 Here’s a post from a year ago by a company called AnyScale on how they built out their own custom LLM-system stack tailored to their dataset....