Thank God! And here we were stressed out that LLMs might democratize access to tech and knowledge.
All of those biases are still there even for the harvard bio. If from the exchange it thinks you might be presenting as something you are not, output degradation.
That said, in the beginning of my prompts I tell it exactly the persona of the target audience for the answer. Otherwise how would it know if it is explaining to a 5 year old or a phd in an adjacent domain?
As always the problem is the training data and how these models don't get to decide how they interpret the training data.
I'd argue there's little rational for having the model talk down to people which isn't malicious. If the user doesn't understand the answer, they can explicitly ask the model to explain it in simpler terms. If you read through the study, it's pretty clear that this isn't just accidental bias from the training data, but rather intentional limiting of capability for specific groups of users.
Runescape as well. Gold farming for membership bonds severely stratified the game's economy and pushed people towards Oldschool Runescape. From the introduction of bonds until 2017 the number of bots raised at a rate unseen since the switchover to the new skilling menu in 2002.
If someone mentions Sabine Hossenfelder and it isn't to expose her as a rage-bait intellectual dark web grifter, then it puts that person in a suspect light.
(I agree with Filippo's post and it can also be applied to Renovate's security updates for Go modules - we don't have a way, right now, of ingesting better data sources like `govulncheck` when raising security PRs)
That just reminds me that I got a Dependabot alert for CVE-2026-25727 – "time vulnerable to stack exhaustion Denial of Service attack" – across multiple of my repositories.
All of those biases are still there even for the harvard bio. If from the exchange it thinks you might be presenting as something you are not, output degradation.
That said, in the beginning of my prompts I tell it exactly the persona of the target audience for the answer. Otherwise how would it know if it is explaining to a 5 year old or a phd in an adjacent domain?
As always the problem is the training data and how these models don't get to decide how they interpret the training data.
reply