Chris_Leong

Karma: 7,016

Chris_Leong 14 Nov 2024 13:48 UTC
LW: 10 AF: 4
−2
AF
in reply to: DanielFilan’s comment on: DanielFilan’s Shortform Feed
I agree that we probably want most theory to be towards the applied end these days due to short timelines. Empirical work needs theory in order to direct it, theory needs empirics in order to remain grounded.

Chris_Leong 14 Nov 2024 13:38 UTC
LW: 6 AF: 4
2
AF
in reply to: DanielFilan’s comment on: DanielFilan’s Shortform Feed
Thanks for writing this. I think it is a useful model. However, there is one thing I want to push back against:
Looking at behaviour is conceptually straightforward, and valuable, and being done
I agree with Apollo Research that evals isn’t really a science yet. It mostly seems to be conducted according to vibes. Model internals could help with this, but things like building experience or auditing models using different schemes and comparing them could help make this more scientific.
Similarly, a lot of work with Model Organisms of Alignment requires a lot of careful thought to get right.

Chris_Leong 13 Nov 2024 11:29 UTC
2 points
0
on: AI Craftsmanship
Remember back in 2013 when the talk of the town was how vector representations of words learned by neural networks represent rich semantic information? So you could do cool things like take the [male] vector, subtract the [female] vector, add the [king] vector, and get out something close to the [queen] vector? That was cool! Where’s the stuff like that these days?

Activation vectors are a thing. So it’s totally happening.

Chris_Leong 13 Nov 2024 6:11 UTC
LW: 4 AF: 2
0
AF
in reply to: Buck’s comment on: Thomas Kwa’s Shortform
“How can we get more evidence on whether scheming is plausible?”—What if we ran experiments where we included some pressure towards scheming (either RL or fine-tuning) and we attempted to determine the minimum such pressure required to cause scheming? We could further attempt to see how this interacts with scaling.

Chris_Leong 12 Nov 2024 11:42 UTC
4 points
0
in reply to: AnthonyC’s comment on: Linkpost: “Imagining and building wise machines: The centrality of AI metacognition” by Johnson, Karimi, Bengio, et al.
I guess I was thinking about this in terms of getting maximal value out of wise AI advisers. The notion that comparisons might be unfair didn’t even enter my mind, even though that isn’t too many reasoning steps away from where I was.

Chris_Leong 12 Nov 2024 4:56 UTC
2 points
0
in reply to: AnthonyC’s comment on: Linkpost: “Imagining and building wise machines: The centrality of AI metacognition” by Johnson, Karimi, Bengio, et al.
That’s a fascinating perspective.

Chris_Leong 31 Oct 2024 5:38 UTC
0 points
−4
on: What TMS is like
Fascinating. Sounds related to the Yoga concept of kryias.

Chris_Leong 31 Oct 2024 0:41 UTC
0 points
0
in reply to: ChristianKl’s comment on: Why I quit effective altruism, and why Timothy Telleen-Lawton is staying (for now)
I would suggest adopting a different method of interpretation, one more grounded in what was actually said. Anyway, I think it’s probably best that we leave this thread here.

Chris_Leong 29 Oct 2024 22:51 UTC
0 points
0
in reply to: ChristianKl’s comment on: Why I quit effective altruism, and why Timothy Telleen-Lawton is staying (for now)
Sadly, cause-neutral was an even more confusing term, so this is better than the comparative. I also think that the two notions of principles-first are less disconnected than you think, but through somewhat indirect effects.

Chris_Leong 29 Oct 2024 16:45 UTC
2 points
0
in reply to: lincolnquirk’s comment on: Why I quit effective altruism, and why Timothy Telleen-Lawton is staying (for now)
We’re mostly working on stuff to stay afloat rather than high level navigation.
Why do you think that this is the case?

Chris_Leong 29 Oct 2024 16:38 UTC
−1 points
−4
in reply to: ChristianKl’s comment on: Why I quit effective altruism, and why Timothy Telleen-Lawton is staying (for now)
I recommend rereading his post. I believe his use of the term makes sense.

Chris_Leong 29 Oct 2024 16:30 UTC
2 points
0
on: The Alignment Trap: AI Safety as Path to Power
I don’t think I agree with this post, but I thought it provided a fascinating alternative perspective.

Chris_Leong 28 Oct 2024 19:41 UTC
4 points
0
on: Winners of the Essay competition on the Automation of Wisdom and Philosophy
Just wanted to mention that if anyone liked my submissions (3rd prize, An Overview of “Obvious” Approaches to Training Wise AI Advisors, Some Preliminary Notes on the Promise of a Wisdom Explosion),
I’ll be running a project related to this work as part of AI Safety Camp. Join me if you want to help innovate a new paradigm in AI safety.

Chris_Leong 27 Oct 2024 15:36 UTC
2 points
0
in reply to: avturchin’s comment on: avturchin’s Shortform
What’s ABBYY?

Chris_Leong 26 Oct 2024 16:42 UTC
5 points
2
on: Brief analysis of OP Technical AI Safety Funding
That’s useful analysis. Focusing so heavily on evals seems like a mistake given how AI Safety Institutes are focused on evals.

Chris_Leong 25 Oct 2024 11:01 UTC
3 points
−1
on: Linkpost: Memorandum on Advancing the United States’ Leadership in Artificial Intelligence
I guess Leopold was right^[1]. AI arms race it is.
1. ^
  I suppose it is possible that it was a self-fulfilling prophecy, but I’m skeptical given how fast it’s happened.

Chris_Leong 24 Oct 2024 23:54 UTC
4 points
2
on: Big tech transitions are slow (with implications for AI)
The problem is that accepting this argument involves ignoring how AI keeps on blitzing past supposed barrier after barrier. At some point, a rational observer needs to be willing to accept that their max likelihood model is wrong and consider other possible ways the world could be instead.

Chris_Leong 24 Oct 2024 16:07 UTC
3 points
−1
in reply to: ChristianKl’s comment on: Why I quit effective altruism, and why Timothy Telleen-Lawton is staying (for now)
I thought that this post on strategy and this talk were well done. Obviously, I’ll have to see how this translates into practise.

Chris_Leong 24 Oct 2024 6:55 UTC
2 points
0
on: Introducing Transluce — A Letter from the Founders
One thing I would love to know is how it’ll work on Claude 3.5 Sonnet or GPT 4o given that these models aren’t open-weights. Is it that you have access to some reduced level of capabilities for these?

Chris_Leong 23 Oct 2024 16:06 UTC
2 points
−5
on: Why I quit effective altruism, and why Timothy Telleen-Lawton is staying (for now)
That was an interesting conversation.
I do have some worries about the EA community.
At the same, I’m excited to see that Zach Robison has taken the reins as CEA and I’m looking forward to seeing how things develop under his leadership. The early signs have been promising.