Rob Wiblin
speaker
1,787 appearances
3 recordings
1 series
first heard Sep 2024
last heard May 2025
Rob Wiblin’s voice in public audio — every appearance, attributed to the second.
Trend
recordings per month · last 12 monthsNo recordings in the last 12 months.Older appearances are listed below; set an alert to hear about the next one.
Appearances
Uh but first, uh, as I just said, uh you lead model training at Anthropic.
What's something that people get wrong or kind of misunderstand about uh AI uh AI model training?
I imagine there could be quite a few things.
This whole vision that scaling is just gonna keep going.
Uh you know, we just make just throw in more data, throw in more compute, uh the models are gonna become more powerful.
That was kind of the part of the it feels like a very anthropic idea, or it was it was part of the founding vision uh that uh that that Dario had, right?
That hypothesis.
What do you think is kind of generating that skepticism that many people have?
I mean, I th like people who are skeptical of scaling laws, there's there's some pretty smart people who are kind of involved in ML are certain certainly certainly have their technical chops.
Um why do you think they are generating this prediction that you that you disagree with?
I mean the argument for why you might expect these scaling laws, you know so so sc scaling laws it's that you uh you you increase the amount of compute and data by some particular uh proportion and then you get a a similar i improvement each time in uh in in the accuracy of the model.
That's that's kind of the the rule of thumb here.
And and the
The argument that I've heard for why you might expect that trend to break and perhaps the improvements to become smaller and smaller for a given uh given given scale up.
i it's something along the lines of, you know, as you're approaching human level, the model can learn by just copying existing state of the art of what humans are already doing in the in in the in the training set.
But then if you're trying to exceed human level, if you're trying to do, you know, I write better essays than any human has ever written, uh then that's maybe a different regime and you might expect like more gradual process like more more gradual improvements uh once you're trying to what you what you try to I get get to a superhuman level.
Do you think that that argument kinda holds up?
Hm.
And is the argument for how it could speed up that at the point that you're near human level, then you can use the AIs in the process of doing the doing the work?
Or is it something else?
Showing 1121–1140 of 1,787 · page 57 of 90
← Previous
Next →