Cal Newport
👤 SpeakerAppearances Over Time
Podcast Appearances
Like, here we go.
And they tried it, and they said, it's not much better.
And this was this big brick wall surprise for the industry.
Like, wait, it didn't get better.
Everyone else tried as well, right?
Grok, they tried this with Grok as well, with the Colossus data center.
Like, we're going to have 200,000 GPU data center.
No one's ever built anything this big.
And it was like a little bit better.
Meta tried this.
They had a model called BMF.
Like we built the biggest data is bigger than anyone we've had before.
They didn't release it because it was marginally better than the last model that they had.
And so this was a huge issue, right?
You couldn't just make the models bigger and train them bigger.
So what they did was they switched to
What are other ways we can get performance increases?
And can we get more narrow by what we mean with performance?
And this is when we began to get all the alphabet soup models.
Well, it's GPT-O, O3-mini slash whatever.