Garrison Lovely
speaker
631 appearances
3 recordings
1 series
first heard Jul 2026
last heard 15 Sep
Garrison Lovely’s voice in public audio — every appearance, attributed to the second.
Trend
recordings per month · last 12 monthsRecordings per month over the last 12 months — 3 in all, peaking in Sep 2026 with 2.
Appearances
Like I I've been covering this, you know, for three and a half years full-time, and it's just like
a really worrying level of of uh leap in performance, at least from those numbers.
You know, grains of salt are that benchmarks systematically overstate how capable AIs are because they can be automatically graded and anything like that, you can just use reinforcement learning and and make the AI get much, much better at it.
And so we'll have to see, you know, how it is to use this.
And like my test of is it AGI is like, can it plan a wedding?
Um
don't know.
Like it maybe it can now because it can do like days and days of work at a time apparently and result in like useful answers without any kind of guidance, you know, i in between.
I think that like, yeah.
And and OpenAI says like this new technique they're using.
It's like, we're just doing a little bit of it.
So it's fine.
We can still see the chain of thought.
But it's the first step on a slippery slope and if it does create performance increases, people will do it and it will just like beat this race to the bottom.
And there are very few things where the industry was successfully self-governing.
But this was one of them.
And it's very concerning that uh they're doing this, especially in light of just like the catastrophic series of security failures, the complete uh lack of desire to share the information with the public, uh, covering up, you know, important stuff that's happening and that they're responsible for.
And, you know, a lot of this stuff is criminal.
Like,
The actions of these AIs would be criminal if a human did it, but because a human did it, it's not a crime.
Showing 421–440 of 631 · page 22 of 32
← Previous
Next →