Natasha (Nitasha) Tiku

speaker
130 appearances 1 recordings 1 series first heard Jul 2026 last heard 10 Jul

Natasha (Nitasha) Tiku’s voice in public audio — every appearance, attributed to the second.

Trend

recordings per month · last 12 months
1 · Jul OctJan 26AprJulnow

Recordings per month over the last 12 months — 1 in all, peaking in Jul 2026 with 1.

Appearances

newest first · ▶ plays the moment
topic really interesting is because I think they do enormous amounts of work to make them sound more human.
Like if you if you look at like GPT-2, for example, and you said a man walks into a bar, it would just like finish the sentence.
You couldn't you couldn't ask.
You
To follow what you wanted there was not you know there wasn't that like reinforcement learning from human feedback so that like oh people like it when you phrase something in this way and I just find it wild that anthropic has you know has said on some of these blog posts like we can't.
there's no way we could possibly make this human assistant less human.
You have like a multi-thousand word constitution where you're talking to it about morality and its potential consciousness.
Like, and you're apologizing to Claude about, you know, its moral personhood in case it has one.
Like you are doing a lot to make it sound human and then to kind of, yeah, I guess then do tests where you're just going-
What do you mean?
That's different.
They're saying they tweak the responses to those answers.
But in terms of how human it sounds, where you can very easily like start to get annoyed with it or start to like.
You know, a lot of the research is, you know, which they caveat, you know, they say we're not really sure how to interpret this results.
But a lot of it involves applying the same techniques you would use for human psychology.
A lot of it just involves asking the chatbot questions about itself.
And that's where I find it super interesting that this is happening in parallel because they had anthropic, for example, researchers.
In in one of its first like model welfare reports, which they include in like mythos to there.
It's like here's a cyber weapon and here's its internal feelings.
But it had it let two chatbots talk to themselves.
Showing 61–80 of 130 · page 4 of 7 ← Previous Next →