Hacker News Reader

Comment by famouswaffles | Hacker News Reader

Real word actions and consequences give the words meaning. If you're rude to Bing, the conversation will end. You can keep typing but it won't respond.

Searching the web and ending conversations are the only two actions Bing can currently do so what i'm saying is less applicable for now. But humanity is gearing up to give more and more control of more and more tools to increasingly powerful LLMs. Hell Bing is scheduled a bump in tools in the coming weeks.

Hate to break it to you but "it's not real [insert property]" is not a shield. Not when you allow the machines to interact with the real world.

If your words to [insert LLM] guide responses that perform actions detrimental to you then those actions are no less real or detrimental because you've deluded yourself into thinking "it's not real understanding" is an intelligent argument.

View on HN

So? This is a guardrail added by a Microsoft employee, mate. After their earlier dabblings in AI outputting bigotry and general toxicity, someone in their wisdom saw the sense to try and prevent going down a negative conversation path, hence all the edge lords on Reddit offering prompts to "jailbreak" an LLM and bypass the guardrails, which then get patched by a human being not long after.

You're projecting so much meaning onto something that is functioning as designed by throwing a sentiment var into the calculation, just like you can measure "sentiment" with social media analysis tools to get a feel for which way the wind is blowing or which of the millions of tweets at your brand might be worth prioritizing.

Do you really believe you've thus offended an LLM? That is delusional.

I’m not in the least worried about the supposed dangers of AI; I don’t see it causing major unemployment, nuclear armageddon, or anything particularly scary. However, I think the real danger with those chatbots is how people are becoming emotionally involved with them. This is clearly by design and it’s working extremely well with so many people.

Yes, completely agreed. There's a lot of being kidded that it's more than what it actually is because the hype is overblown. Statements are being made and repeated that simply have no basis in reality right now, nor in the future. Even this assumption that we are close to having AGI, I can only laugh.

But none of its behaviour is even close to by design. The emergence of reasoning is completely unexpected for a word prediction model.

It's a guardrail that Bing controls, which is the whole point.

It doesn't matter what i believe or not. or what you believe or not. What matters are actions and consequences. "It's not really offended" doesn't change the outcome or make it less material(in this case - the end of the conversation). You're focusing on things that don't matter in the slightest.

It doesn't need to be "really offended", whatever that means to you. It just needs to be able to model being offended well enough to take actions that resemble an offended person. If you don't understand that then i don't know what else to tell you.

It's a guardrail that Microsoft controls, which the LLM is bound by.

If a car has a hood and headlights reminiscent of a face and tells you to "buckle up, buttercup" when you don't do your belt, does that mean it's genuinely caring for you?

Microsoft arranged the guardrail. Bing decides when it should be implemented, much like an employee.

>If a car has a hood and headlights reminiscent of a face and tells you to "buckle up, buttercup" when you don't do your belt, does that mean it's genuinely caring for you?

Why do you keep going on pointless tangents?. Whether you think it genuinely cares or not is irrelevant. This isn't a hard concept to understand.

Source? Because every source I'm finding tells me pretty much words to this extent, which doesn't support what you're saying whatsoever with the LLM choosing to do anything:

"RAIL (Reliable AI Markup Language) is a language-agnostic, human-readable format for specifying structure, type information, validators, and corrective actions for LLM outputs. RAIL, an XML flavor, allows users to define the expected structure and types of LLM outputs, the quality criteria for valid output, and the corrective actions to take if the output is invalid."

Why do you keep projecting your own humanity onto the LLM?

Source for what ? Do you have any idea of what you're talking about ? Hold on, How do you think Language models work ? What do you think the plain English Instructions are for ?

A source for how guardrails function as what you state isn't what's documented. You said that the LLM chooses when to use them.

The LLM sits behind the toolkit which sits behind the guardrails. Are you sure you understand how they work?

The guardrails monitor your input, the output of the LLM, and modify the output to suit, including terminating the chat.

It should be in line with the likes of this - https://arxiv.org/abs/2302.04761 or https://twitter.com/minosvasilias/status/1627076214639976449

I didn't say guardrail to mean anything particularly elaborate. The pre-prompt for Bing has instructions to avoid adversarial arguments.

Reply on news.ycombinator.com