AI - AI Dynamics

ChatGPT programmers lack epistemics and advance predictions

By

–

04 December 2022 3h11

By default, I would guess that if the people programming ChatGPT responses do not even think to ask 'But is that actually true?', then the company does not have a secret internal committee where everyone has great epistemics and wrote down advance predictions.

OpenAI could prove advance knowledge through hashed predictions

By

–

04 December 2022 3h09

Well, they could always make the predictions internally, publish a hash somewhere unobtrusive but time-stamped, swear me to secrecy on the specific contents, show me the impolite internal prediction, and have me say publicly, "Sorry! I'm convinced OpenAI knew in advance."

GPT-3.5 Could Lower Status of Bullshit and AI-Like Speech

By

–

04 December 2022 3h08

It's more of a hope than a prediction: but I would not be *shocked* if GPT 3.5's incredible facility at producing boilerplate, vague disclaimers, unfalsifiable abstractions, etc, might actually succeed in lowering the status of bullshit, and getting people to talk less like AIs.

ChatGPT’s Vague Claims About Its Own Capabilities

By

–

04 December 2022 2h43

Sounds pretty vague to me; I think you're reading in the interpretation you now know to be correct. Example of contrary update (for me): ChatGPT insists it doesn't have abilities it clearly still has. I'd bet the people who burned in those replies weren't knowingly dishonest.

→ View original post on X — @fchollet

Learning and Refining Semantic Categories from Single Instances

By

@fchollet

–

04 December 2022 2h39

Learn a word based on a single instance (e.g. this particular airplane = word "airplane"), then use the word to refer to a broad category of related things (all flying things are now airplanes), then gradually refine semantic categories (helicopters, dirigibles…)

→ View original post on X — @goodside

ChatGPT pretends to have code but fails scrutiny

By

@goodside

–

04 December 2022 2h38

ChatGPT is capable writing working code in other contexts, for simple problems, but this isn’t that. This is pretending that some larger body of code exists, and then talking about it and showing plausible-seeming pieces of it. It wouldn’t stand up to serious scrutiny.

→ View original post on X — @animaanandkumar

AI for Science as True Test of AI Capabilities

By

@animaanandkumar

–

04 December 2022 2h30

I believe @AI_for_Science is a true test of #AI capabilities. While generating realistic text or images can be fun to play with, more controlled tests of #AI is easier with scientific domains where we have more concrete metrics and objectives to measure

→ View original post on X — @animaanandkumar

AI for Science Community Growth at NeurIPS22

By

@animaanandkumar

–

04 December 2022 2h27

So pleased to see @AI_for_Science community grow at #NeurIPS22 When we founded AI for science initiative at @caltech in 2018, it was still considered niche. I want to thank organizers of @AI_for_Science workshop, especially student organizers who put in so much time and effort.