med-mastodon.com is one of the many independent Mastodon servers you can use to participate in the fediverse.
Medical community on Mastodon

Administered by:

Server stats:

413
active users

#yannlecun

0 posts0 participants0 posts today
Continued thread

The invited talk by #LeslieKaelbling has been excellent. She was able to find a justification of many classic #AI ideas within the current discourse and gave a robot demo that had some of the spirit of the good old demo of #Shakey the robot.

The invited talk by #YannLeCun has been interesting in the sense that he is opening up to ideas such as online optimization (i.e. at the time of acting).

I also liked the panel in honor of #DougLenat and the information that was given about #Cyc. ⤵️

Deep Neural Nets: 33 yrs ago & 33 yrs f. now
karpathy.github.io/2022/03/14/
Disc'n: news.ycombinator.com/item?id=3

* Interesting retrospective / perspective / prospective f. phenom Andrej Karpathy: en.wikipedia.org/wiki/Andrej_K
* founding member, OpenAI: en.wikipedia.org/wiki/OpenAI | ChatGPT, GPT-4, ...)
* "most important trend ... training NN f. scratch ... quickly becoming outdated due to finetuning esp. w. emergence of foundation models like GPT ..."

karpathy.github.ioDeep Neural Nets: 33 years ago and 33 years from nowMusings of a Computer Scientist.
Replied in thread

@deanpierce @craignewmark #LLM (Large Language Models) and #NN (Neural Networks) are, to me, built like black boxes. In essence, you can't ask it to explain why or cite its "sources" without it throwing the kitchen sink back because that's how they are trained.
I'm no expert in #LLM #NN but tool like #ChatGPT has definitely put it on my radar to try to learn more.
Heated debates between experts like #GradyBooch (calling ChatGPT "statistical nonsense at scale") & #YannLeCun are teaching me lots.