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.
17 postsUpdates live
Daily Alchemy: a question to think on together
“Should hospitals use the News-Medical ML method for uncovering hidden self-harm histories without patient consent?”
Trust ranks through the site stewards' curated lens.
- b_b...
What is the topic of a comment?
Some hypotheses: 1. It’s not contained in the comment itself, it’s something about the comment, in the context of the conversational tree it’s in. 2. There’s not just one "topic".... I really like the latent space / embedding vector approach. So I my first idea to addressing the "multiple topics per comment" is: what if we could vote for sections of a contribution?... - b_b...
Another 2 items for your list: particular to your context: the experiment worked in your hands, but it works ~only within your particular context (the pollen mix in Austin, or your particular handwashing protocol, or the way people at your university smile to each other, or the... It’s a little dangerous though! It’s biased in the user’s favor, and might tell you something is a good idea even if it isn’t (RLHF biases towards sycophancy a bit).... "Well now Marcello, you’ve been chewing on cacti again, haven’t ya? Infer the intent behind upvotes? You planning on inventing the mind-readin’ machine next, partner? Pardoning your crude fix, it sounds like you’re trying to rope the wind....