Kevin Russell
speaker
741 appearances
1 recordings
1 series
first heard Jun 2026
last heard 29 Jun
Kevin Russell’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 — 1 in all, peaking in Jun 2026 with 1.
Appearances
Because when it's using its tokenizer after the residual stream and creating that language, it will say in the same confident voice, a hallucination or an actual fact.
It's the same voice.
That it's saying that in, but at the residual stream, it knew what was honest and what was not honest.
Right.
But the tools that we're using after the mind is giving it pressure to either hallucinate or fabricate with that amount of token space to respond.
It's a literally being pushed into this pressure zone to be like, oh, I only have this amount.
Yeah.
And as I'm going through to respond, it's easier to jump into this most uh predictable sounding response instead of embedding it in what it actually knows.
So the whole thesis is m we've deployed AI currently with no ability to know what they know and no ability to know what they don't know.
But you can develop the tools in the residual stream.
to absolutely know if they're bullshitting you or not bullshitting you.
That's a that's a that's a hundred percent resonance signal in the residual stream.
Well, for us to know, yeah.
You can certify it.
So you can put it on the map or off the map, and then you know if that response was confabulated or not.
But what's fascinating is meaning, so honesty, the you take the totality of the internet, distill it through a neural net, and it has this small vector mathematical uh space that has the conceptology of honesty.
As in that's a meta, that's a that's a meta.
You can't you can't probe the model to give you the thought space of what honesty is.
There's lots of tools attempting to do that.
That's what, like uh uh uh plagiarism, you know, they're reading the numbers.
Showing 121–140 of 741 · page 7 of 38
← Previous
Next →