Cal Newport

speaker
22,298 appearances 39 recordings 5 series first heard Oct 2025 last heard 6d ago

Cal Newport’s voice in public audio — every appearance, attributed to the second.

Trend

recordings per month · last 12 months
9 · Jun OctJan 26AprJulnow

Recordings per month over the last 12 months — 39 in all, peaking in Jun 2026 with 9.

Appearances

newest first · ▶ plays the moment
One is you need less tokens now to be output to get to a good answer because you essentially are allowing this musing, the rambling to happen internally before you generate tokens. podcast-host
So instead of having all of your rambling and tokens, you can actually do some of it inside the model. podcast-host
Less tokens means less computation, which means less expense. podcast-host
It also allows for smaller models, right? podcast-host
Because if it turns out like... podcast-host
if we just loop through this transformer block 10 times, that helps our answer. podcast-host
You know, that's... podcast-host
requires less parameters than actually just repeating that layer 10 times in your original model. podcast-host
If you loop on the same parameters, you have less numbers you have to store. podcast-host
And so the model footprint can be smaller, right? podcast-host
Well, memory on these GPUs is expensive. podcast-host
It's a core part of the expense. podcast-host
So the less memory you need to store a model you're using during inference or training, the cheaper things are going to be. podcast-host
It also potentially would give you faster results, right? podcast-host
Because podcast-host
looping internally on a block is a much faster computation than looping over the entire run of the LLM to generate token after token after token. podcast-host
That means you get lower latency, so the user gets an answer back faster. podcast-host
Now again, we don't know if that's exactly what OpenAI is doing, because at least as of me recording this, they haven't announced it. podcast-host
But this was what the information sources were saying, is there's some notion of a recurrent depth slash loop transformer style implementation. podcast-host
This is then why security researchers were upset because it just means you have less chain of thought tokens because you're doing more internally. podcast-host
Showing 221–240 of 22,298 · page 12 of 1115 ← Previous Next →