LessWrong 2.0 Reader

• Fine-tuning for GPT-4 is in an experimental access program since at least November, and OpenAI has written about fine-tuning GPT-4 for a telecom company.
• Anthropic says "Our API does not currently offer fine-tuning, but please ask your Anthropic contact if you are interested in exploring this option."
• You can apparently fine-tune Gemini 1.0 Pro.

Maybe setting up custom fine-tuning is hard and labs often only set it up during deployment...

(Separately, it would be nice if OpenAI and Anthropic let some safety researchers do fine-tuning now.)

He isn’t in charge there. He simply offers research directions and probably a link to academia.

(Yeah, the Altman statement seems to be emailed to journalists, despite being short. Not sure why he's not just tweeting like previously.)

Guessing his lawyers wrote this. And probably to match the medium of ScarJo’s initial statement to show respect to her.

The LessWrong Review [? · GW] runs every year to select the posts that have most stood the test of time. This post is not yet eligible for review, but will be at the end of 2025. The top fifty or so posts are featured prominently on the site throughout the year. Will this post make the top fifty?

I think Frontier Red Team is about eliciting model capabilities and Alignment Stress Testing is about "red-team[ing] Anthropic’s alignment techniques and evaluations, empirically demonstrating ways in which Anthropic’s alignment strategies could fail."

I was just reading about this, and apparently subvocalizing refers to small but physically detectable movement of the vocal cords. I don't know whether / how often I do this (I am not at all aware of it). But it is literally impossible for me to read (or write) without hearing the words in my inner ear, and I'm not dyslexic (my spelling is quite good and almost none of what's described in OP sounds familiar, so I doubt it's that I'm just undiagnosed). I thought this was more common than not, so I'm kind of shocked that the reacts on this comment's grandparent indicate only about 1/3 (of respondents to the "poll") subvocalize. The voice I hear is quite featureless, and I can read maybe 300 words per minute, which I think is actually faster than average, though needing to "hear" the words does impose an upper bound on reading speed.

That could be.

My recollection from Zuckerberg was that he was thinking of AGI as a fairly far-away goal, more like 8 to 20 years++ (and I'd assume "transformative AI" would be further), and that overall, he just hasn't talked much about it. 

I wasn't thinking of all of Yann LeCun's statements, because he makes radical/nonsensible-to-me statements all over the place. It's not clear to me how much most of his views represent Meta, though I realize he is technically in charge of AI there.

An unintended consequence: While writing this post, I realized that adopting this heuristic for my dating life led to it bleeding into other parts of my life. In conversations with friends, family, bosses, I found myself asking what an emotionally healthy person would do in this context and it improved those interactions, as well.

From the RSP Evals report:

As a rough attempt at quantifying the elicitation gap, teams informally estimated that, given an additional three months of elicitation improvements and no additional pretraining, there is a roughly 30% chance that the model passes our current ARA Yellow Line, a 30% chance it passes at least one of our CBRN Yellow Lines, and a 5% chance it crosses cyber Yellow Lines. That said, we are currently iterating on our threat models and Yellow Lines so these exact thresholds are likely to change the next time we update our Responsible Scaling Policy.

What's the minimum X% that could replace 30% and would be treated the same as passing the yellow line immediately, if any? If you think that there's an X% chance that with 3 more months of elicitation, a yellow line will be crossed, what's the decision-making process for determining whether you should treat it as already being crossed?

In the RSP it says "It is important that we are evaluating models with close to our best capabilities elicitation techniques, to avoid underestimating the capabilities it would be possible for a malicious actor to elicit if the model were stolen" so it seems like folding in some forecasted elicited capabilities into the current evaluation would be reasonable (though they should definitely be discounted the further out they are).

(I'm not particularly concerned about catastrophic risk from the Claude 3 model family, but I am interested in the general policy here and the reasoning behind it)

The LessWrong Review [? · GW] runs every year to select the posts that have most stood the test of time. This post is not yet eligible for review, but will be at the end of 2025. The top fifty or so posts are featured prominently on the site throughout the year. Will this post make the top fifty?