Emad Mostaque
speaker
1,351 appearances
1 recordings
1 series
first heard Dec 2024
last heard Dec 2024
Emad Mostaque’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
We overindexed on chefs versus cooks.
So you know, everyone's on the spectrum, Wait But Why had a great post from being a cook that follows recipes to making recipes.
How many recipes do you really need to make?
Where AI is actually really good now with function calling large context windows is following recipes.
You give it a handbook and it's basically replaces a SAS or it replaces an employee or something like that.
And we wanna have more and more employees that follow the DAM instructions.
effect.
How often do you need to come up with something brand new for the vast majority of people?
For that, there is the O one type models.
There are all these other kind of things.
And again, this is where you have the AGI race away.
But if you're trading, say for instance, on a million H one hundreds like Elon's about to, or a million TPUs like Demis is about to, or a million tradium, where God forbid Daria's about to, poor guy, you know, like how big is that model actually gonna be for inference side?
You kind of went straight through the consumer side and now you're like, Well, am I gonna build a multi trillion parameter model that needs a Cerebrus wafer or a Google TPU pod to run?
You know, because that's not really suitable for consumers at the side of things, but it might be more intelligent.
But it might only be five percent more intelligent.
Whereas it requires orders of magnitude less compute to have the lower ones.
And this is where it becomes very interesting because classically most of computation has been sequential as opposed to like parallelized.
And we're seeing this now again with test time compute and others.
Like, who's gonna win if you need to have millions and millions of GPUs running in parallel, running O one type things to check something and cross check it?
Probably more distributed elements.
Showing 461–480 of 1,351 · page 24 of 68
← Previous
Next →