New Year resolution

[dmm]

To read my https://twitter.com/home more regularly (that's absolutely the best source of info at the moment).

A small fraction of today's catch:

New work by Janus

A new involved take on AI safety/alignment

(What's the right way to organize all that information?)

Links are in the comments (I think the new work by Janus is more important even for alignment, and is just overall more important of the two topics of this post)...

Comments

[dmm]

> A new involved take on AI safety/alignment

https://theinsideview.ai/david

via https://twitter.com/MichaelTrazzi/status/1611824181636644864

[dmm]

"David Krueger On Academic Alignment"

"happy to finally share my 3h conversation with David Krueger, Cambridge Professor, about the AI Alignment research going on at his lab, coordination, takeoff speeds and why he doesn't have a research agenda

transcript: https://theinsideview.ai/david
youtube: [...]"

[dmm]

A mess, though, not a neatly organized thing...

[dmm]

https://www.lesswrong.com/posts/3ZdfoezXhKZHakEzy/ethan-caballero-on-broken-neural-scaling-laws-deception-and

https://arxiv.org/abs/2210.14891 "Broken Neural Scaling Laws"

*****

"Unifying Grokking and Double Descent" https://openreview.net/forum?id=JqtHMZtqWm

*****

"Assistance with large language models", https://openreview.net/forum?id=OE9V81spp6B

[dmm]

https://neurips2022.mlsafety.org/

[dmm]

https://theinsideview.ai/david#coordination-is-neglected-and-there-are-many-low-hanging-fruits

'you shouldn’t just say, “Here are my timelines,” and it’s a number, because timelines aren’t a number, they’re distribution. You might want to invest in something that looks like more likely to pay off over long timelines, even if you think timelines are likely to be short. You still have a significant amount of mass on long timelines'