Nitasha (Natasha) Tiku
speaker
139 appearances
1 recordings
1 series
first heard Jul 2026
last heard 24 Jul
Nitasha (Natasha) Tiku’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 Jul 2026 with 1.
Appearances
And this is something people had been warning against, that maybe the biggest danger is not when your systems are in deployment, because then you have all your safeguards up, it's refusing a lot of requests, but it's actually during this kind of testing environment.
And I'll just say one more thing.
This incident followed another blog post last week from OpenAI about how they had to stop even just testing an internal model.
because they weren't monitoring it and it started doing all of these things because they let it go on and on because that's how these models get more functional.
They're able to think and try thousands and thousands of different ways of breaking into things.
OK, I will say, I mean, I do think it's important to note that this is like many people are interpreting this as like we were right about the paperclip maximizer.
You know, this is happening in this way.
But I think there's a way that because they conceived of it, of the problem this way, they approached security in a certain way.
Like you could have been thinking like a cybersecurity professional the whole time and not thinking about it.
AI alignment, aligning it with human values.
And one of the things I mentioned that like prior OpenAI blog post, what they realized is that they didn't have sufficient monitoring systems.
When you let a system go on and on for a really long time, they weren't watching it closely.
Like there are simple mechanisms online.
It doesn't even go back to what Reid said about knowing what's happening on the neurons.
They weren't even watching what it was doing during the security test in a secure way.
I think the way that they wrote the blog post at the top, it says, like, we are treating this as an unprecedented cybersecurity incident that shows the capabilities.
And I will say, like, I don't think anyone's arguing that these LLMs make people much, much, much better.
And that it's capable of improving and enhancing and trying, you know, cyber offensives that humans can't do.
But everyone I talked to was like, this is not an example of enhanced capabilities.
This is an example of like kind of a sloppy testing environment.
Showing 41–60 of 139 · page 3 of 7
← Previous
Next →