PhD student in reinforcement learning and interpretability at University of Bath.
Karolis Jucys
Karma: 31
Any advice on how to use ANKI for 3yos? I recently started a deck for times tables, Chinese Hanzi and the lyrics of The Elements and The Love Song of the Physical Anthropologist. The 3yo already knew/knows both songs by heart. I just want to make sure they stay in memory as I sing them as lullabies less and less...
A little hack that I love is to have a keyboard shortcut that saves any currently selected text to a file for later ankification.
Very nice idea! I use a bookmark folder for the same purpose. It’s a (mostly) FIFO queue with 165 potential future cards and growing though...
The Hubinger et al. link in the first paragraph points to arXiv /pdf/, not /abs/. This makes it not show preview on hover. Same with “model organisms” link later.
For some reason my brain just said “you should google Tom Lehrer LessWrong”. Boy did it not disappoint! I love this, thank you for writing it! I’ve been singing predominantly Tom Lehrer songs to my toddler since before their birth, including ofc WWAGTWWG. We’re expecting a 2nd one now, I wonder if I should move on with the times and sing this one instead as a lullaby.
I do wonder though. Have you heard of the researcher at MSR named Dave Bignell? He’s another favourite in our house, with gems like this one: https://www.youtube.com/watch?v=_3Z6Hzqw3QU . I wonder if he’d be up for doing a cover of your song perhaps… :)
DreamerV3 is not a great example, as they use so many hacks to make the task easier that it barely counts as getting a diamond or Minecraft anymore. Action shaping, macro actions, instant block breaking, fake “bug fixing”, all to get a diamond in 0.4% of episodes.
More info here: https://x.com/Karolis_Ram/status/1785750372394348632
Would “delta hedging” be useful here? It helps hedge long option exposure by shorting some amount of a stock.
For example, at the money calls generally have a delta of 0.5, so holding 100 at the money calls and shorting 50 shares makes you roughly neutral for small moves in the underlying asset.
Would probably require monthly rebalancing based on how many options you effectively hold and market moves. It also wouldn’t work well if AGI happens at GDM and Google stock goes exponential (“volatility smile” problem).
non pdf arxiv link: https://arxiv.org/abs/2305.15324
For the four examples of
24-16=12, 53-25=25, 34-16=13, 63-17=16
is this the pattern?ab-cd=ca
I worked in hedge funds for 8 years in my past career. My hedgie friends seem to believe the “safety is for marketing” meme. I’ve given them your resignation story as a proof that no, it’s not, the AI people are actually really concerned. Thank you for writing this article too, I’ll definitely share it widely.
One of the big unexpected realisations for me when I moved from cutthroat finance to AI 9 years ago was that people in AI actually cared about the wellbeing of others, not just money. That explains part of the cynicism I think. I had a similar realisation when moving from Eastern Europe to the UK, “wow, people are actually nice!”.
My friends don’t see the improvements in AI the way I see them though. I’ve had several of them in the past 6 months say versions of “So, this AI thing seems to be running out of steam, improvements are really slow”. They’re at firms that are spending millions on AI, and say that the benefit is basically zero, the LLMs are too unreliable and make too many mistakes, not even replacing any of the junior employee work. That makes them think the whole current AI pause thing is because AI improvement is slowing down so much, that there’s no need to burn more compute from investment perspective.
The questions directed at me are in part them asking for investment advice—we worked together at funds focused on energy infrastructure, where AI/datacenters are a new and increasingly large moving part.