Tristan Harris
speaker
532 appearances
2 recordings
1 series
first heard Aug 2026
last heard 16 Sep
Tristan Harris’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 — 2 in all, peaking in Sep 2026 with 1.
Appearances
you will see the neurons light up that start to trigger deception.
I can give you examples.
There's actually research at Anthropic that when they kind of give these, the way they test and do safety is they try to do like a brain scan of this digital brain of like, say, GPT-6 or Opus 4.5 or whatever.
And when they're asking the model to answer a question, they can use these techniques called mechanistic interpretability, basically means just do a brain scan and see which of the digital neurons are lighting up.
In one of the examples, one of the neurons that lit up in Anthropic was if you did a double-click and read out what's inside of that neuron that's lighting up when we tell it to be deleted, and it starts doing deception, the content of the neuron is they deserved to be deceived because they were pigs.
They deserved to be deceived because they were pigs.
Now, we don't know what is causing that.
We don't know where that is in the training data, but it's somewhere in there.
And the point is that these technologies, this is a different kind of technology than a nuclear weapon or than an airplane, where you're writing the code for exactly what you want both of those things to do using clear computation.
It's just clear logic running in both of those technologies.
Whereas with AI, the whole point, as you said, Steve, is it's probabilistic.
it's responding in a general way to all patterns.
And it's been trained on the totality of human knowledge.
And inside of the totality of human knowledge, it will discover that in order to achieve any goal, the best strategy to achieve that goal is to stay alive.
So it has to develop these sub goals like staying alive, preserving myself,
having more power, acquiring more resources.
These are all goals that I know to your audience might sound like science fiction, but it's actually all happening right now.
As you said, this is not me telling you and freaking people out and being a doomer.
This is not that.
This is literally, I'm just sharing objective evidence from the field.
Showing 301–320 of 532 · page 16 of 27
← Previous
Next →