Подтвердите e-mail

Для публикаций, комментариев, реакций и сообщений подтвердите адрес.

←

Публикация

Предыдущая публикация
Baldur Bjarnason

A thing to note, which Hank Green shares with so so many other “smart” people in science or tech, is that his words show he doesn’t understand how these systems work He’d continue to use these systems if they hadn’t triggered what looks like the now common chatbot mental health episode

Kotaku

YouTuber Hank Green Admits He Has Unhealthy Relationship With AI After Facing Backlash Over Recent Video 'The level of dopamine I've been getting from interacting with LLMs is not healthy for me,' said Green

47120
Baldur Bjarnason

For example, a statistical text model doesn’t give you “access” to a paper. It gives you a lossy, part-fictional approximation of a paper, which is a very, VERY different thing

Обсуждение

4 прямых ответов · 6 сообщений

Soyweiser

I saw that the models struggled with negation so I would not use they for summaries at all. Somebody explained that that isn't strange as text statistically wise the sentences 'that is true' and 'that is not true' are pretty similar.

Ответ для Soyweiser
Amy Hoy

this is absolutely not me saying “they’re just like brains” bc they aren’t but interestingly humans also struggle with negation it’s why repeating the lie with “not” is just awful at persuasion

gchance60

I spend my day reminding pretty smart people of the same thing. Two priors shape their trust. Search engines that showed valid content quickly and using other software that is fact based. They “trust” computers (and I guess the internet) to be accurate.

Amy Hoy

IF you are diligent and careful, it can be a powerful way to surface semantic connections between concepts in the corpus that are not linked out in normal space. but my rule is “trustworthiness == forum rando” anyone intoxicated is neither diligent or careful

Amy Hoy

and before you ask, ofc i’ve asked doctors about this first. they don’t know but linking semantic concepts in an enormous corpus of ingested research is something llms can do better than people, as long as you take it as about as trustworthy as a random forum member

Benjamin Harnett

Psychologically fascinating is how angry they get when you try to walk them down this road and they ultimately have to admit they just don’t care and that it works well enough for everybody else and why am I such a complainer?

Ответ для Benjamin Harnett
Joseph Kohlmann

The Cliff’s Notesification / reblogification of everything