If you’re more of a determinist, then human brains are basically like organic LLMs and we’re just unique models trained on our own life’s datasets - with other traits/data like preferences and biases also being inherited via genetics, of course.
Are we not simply employing statistics when we deduce that yes, if we release our grip, the smartphone will fall to the floor?
You wouldn't have to ever see a phone drop in order to know that, and neither would you have to study or know of the terminology of gravity.
It comes from the statistical knowledge that all things fall when not blocked.
2. Rule of inherent knowledge: Some understanding or knowledge, like objects falling when not supported, can be known without explicit study or exposure to the specific terminology (e.g., gravity).
3. Rule of generalization: Observations or experiences with one type of object (e.g., a smartphone) can be generalized to other objects or situations, as long as they share similar characteristics (e.g., not being supported).
4. Rule of causality: There is an implied cause-and-effect relationship between an action (releasing the grip) and an outcome (the smartphone falling to the floor).
5. Rule of experiential learning: Knowledge can be gained through direct experiences, even if the specific terms or scientific concepts are not known.
I believe GPT-3 doesn't have the abstraction for arithmetic.
If the machine acts like it has emotions, runs forever(this kind of agency is already possible to implement though expensive) and can use tools, then treat it like it doesn't have emotions at your own peril. When the machine can "hit you back" (not necessarily physically of course), you'll learn manners pretty quickly. You can see glimpses of this with bing.
Oh please. The human gives the verbose output meaning by projecting onto it. You're clearly letting your emotions run wild. It's a fucking LLM.
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.
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.
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.
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?
>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.
"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?
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.
I didn't say guardrail to mean anything particularly elaborate. The pre-prompt for Bing has instructions to avoid adversarial arguments.
This is a tired assertion, easily disproved for current llms (read some of Lecun's stuff) and if some variation is going to be claimed, it needs to be an affirmative defence. "Maybe x is true" is a meaningless statement.
How so? If we don't know what "real intelligence" consists of, how can we know that LLMs can't have "real intelligence"?
I think it absolutely matters if by only quacking we also want to know is it possibly wanting, feeling, experiencing, having an ego etc.
You say LLMs are able to reason, but https://youtu.be/QO2BvVq_fWU?t=663 here is a proponent of the more simplistic explanation of how they work. On one hand I saw pretty convincing arguments for it having the ability to reason, on the other hand how do we even conclude anything if we can't really know what's in the training data.
ELIZA: How do you do. Please tell me your problem.
YOU: I'm worried about shmerple
ELIZA: Do you believe it is normal to be worried about shmerple ?
The fact that it can insert your made up word into patterns has no significance in itself.> I'm sorry to hear that you're worried about someone named "shmerple." Can you please provide me with more information about who shmerple is and why you're worried about them?
First question would be: is Shmerple a person or a concept, and assume it's a person, since most people don't start conversations about worrying about concepts.
Therefore: who is Shmerple and why are you worried about them?
Even if this kind of intelligence is just statistics, does it really matter? If it quacks like a duck, it's a duck. Maybe our own brains are nothing more than overrated statistical machines?
The only issue I see with the hype is when people attempt to anthromorphize it.