Chris Vallance

speaker

69 appearances 3 recordings 1 series first heard Jun 2026 last heard 6d ago

Chris Vallance’s voice in public audio — every appearance, attributed to the second.

Trend

recordings per month · last 12 months

Recordings per month over the last 12 months — 3 in all, peaking in Sep 2026 with 1.

Appearances

newest first · ▶ plays the moment

Global News Podcast · Tankers moving again after Iran deal · 18 Jun 2026 podcast

But they did say it was getting harder.

So it does seem that the AI companies are getting better at getting their models to stick to the guardrails.

And I guess there is a sort of deeper problem here as well, which is that they don't really understand what they're being asked to do or what they are creating.

The AI doesn't understand in the way that we would understand.

So, for example, with this prompt, what it's being asked to do is quite innocuous.

So it may not trigger alarm bells within the model itself.

But I think a human might look at that and say, this looks really odd and weird.

Why am I being asked to do that?

So in a sense, what the AI companies are faced with

is the challenge of sort of specifying rules for everything rather than, you know, if you like, if it's a human, you might sort of, they might sort of see the output and go, oh, that looks concerning.

I'm not going to present that to the user.

Whereas the AI model is trying to follow, if you like, a set of rules, a set of criteria.

So their understanding of what's bad and good is different from a human's understanding.

Yet, exactly, exactly.

Well, OpenAI say, we take these reports seriously.

After investigating this trend, we've introduced additional safeguards against this type of prompt.

Our safety systems are designed to block potentially harmful images that are uploaded to chat GPT, and we analyze whether the AI-generated image

violates our policies before we show them to the user.

We also combine automated systems and human review to identify and block harmful material.

And I think just to translate that a little bit, what they're saying there is they also have separate AI systems, if you like, scanning the output, trying to spot things that would be against its rules.

Showing 41–60 of 69 · page 3 of 4