News for hackers. Stuff that matters.
Re:GPT4 and the Multi-Modal, Multi-Model, Multi-Everything F... by sebzim4500 on Wednesday March 15, 2023 @03:41PM (#35173645)
I like how the term AGI was invented because people couldn't agree on what AI was and now no-one can agree on what AGI is.

I don't understand why anyone even bothers using the term given in any serious discussion you need to define it up front anyway.

Re:GPT4 and the Multi-Modal, Multi-Model, Multi-Everything F... by z3c0 on Wednesday March 15, 2023 @03:48PM (#35173751)
The distinction between AGI and Narrow AI has existed for almost 30 years.
Re:GPT4 and the Multi-Modal, Multi-Model, Multi-Everything F... by nomel on Wednesday March 15, 2023 @03:52PM (#35173813)
I think a requirement of AGI is a feedback loop to allow continuous execution and “consciousness”.
Re:GPT4 and the Multi-Modal, Multi-Model, Multi-Everything F... by wizofaus on Wednesday March 15, 2023 @04:22PM (#35174202)
Really? How would we even test for that?
Re:GPT4 and the Multi-Modal, Multi-Model, Multi-Everything F... by nradov on Wednesday March 15, 2023 @03:56PM (#35173852)
AGI can pass a rigorous Turing test. By rigorous I mean conducted over at least several hours by a panel of expert evaluators drawn from a wide range of academic disciplines such as computer science, philosophy, psychology, law, history, physics, mathematics, biology, medicine, sociology, and communications. While such evaluations are necessarily subjective, if a panel of experts agrees that a computer system acts like an AGI then it would be tough for the rest of us to dispute.

GPT4 is far from being able to pass such a test, and I don't expect GPT99 to get there either. A fundamentally different approach will be needed.

Re:GPT4 and the Multi-Modal, Multi-Model, Multi-Everything F... by PartiallyTyped on Wednesday March 15, 2023 @04:05PM (#35173997)
> By rigorous I mean conducted over at least several hours by a panel of expert evaluators drawn from a wide range of academic disciplines such as computer science, philosophy, psychology, law, history, physics, mathematics, biology, medicine, sociology, and communications.

Passing such a test is not a property expected of the average human yet the average human is considered a general intelligence, therefore this is not a property of a general intelligence.

Re:GPT4 and the Multi-Modal, Multi-Model, Multi-Everything F... by misnome on Wednesday March 15, 2023 @04:09PM (#35174039)
What if that average human has unlimited time to access and cross reference the entire written internet?
Re:GPT4 and the Multi-Modal, Multi-Model, Multi-Everything F... by PartiallyTyped on Wednesday March 15, 2023 @04:19PM (#35174180)
Humans have mental bounds as well. Having information is not sufficient to synthesize new information that follows logically from the premise, nor is it sufficient to help you understand everything. Even with hundreds of years time there are mental leaps that most of us can't make.
Re:GPT4 and the Multi-Modal, Multi-Model, Multi-Everything F... by nradov on Wednesday March 15, 2023 @04:10PM (#35174056)
Are you even familiar with the Turing test? Passing doesn't require the subject to demonstrate any particular level of knowledge or expertise. The subject only has to convince the evaluators that it is roughly as intelligent as an average human. Average humans are expected to pass such a Turing test.
Re:GPT4 and the Multi-Modal, Multi-Model, Multi-Everything F... by PartiallyTyped on Wednesday March 15, 2023 @04:17PM (#35174143)
Why should a panel exist in the first place if you are not testing knowledge?

Intelligence is independent of one's knowledge. Plenty of intelligent but uneducated people, and plenty of educated people - or even people with perfect recall - that are not particularly intelligent.

If you are testing intelligence you should be able to distill it to particular tests independent of domain expertise.

Any intelligent agent should be capable of in-context learning.

So how would you go about doing that?

Re:GPT4 and the Multi-Modal, Multi-Model, Multi-Everything F... by nradov on Wednesday March 15, 2023 @06:40PM (#35175699)
The (notional) panel exists to evaluate intelligence, not test knowledge or educational achievement. While some basic level of general knowledge would be necessary to even start a test the main focus would be on ability to learn and reason, including metacognition.
Re:GPT4 and the Multi-Modal, Multi-Model, Multi-Everything F... by logifail on Wednesday March 15, 2023 @05:01PM (#35174654)
> Passing such a test is not a property expected of the average human [..]

I'm thinking of the rabbit hole that is the hurdle to get citizenship in a particular country. "Outsiders" have to apply, pay, take and pass the test. "Insiders" don't have to take the test, they pass by default. Even if they don't know the answers to the test.

Re:GPT4 and the Multi-Modal, Multi-Model, Multi-Everything F... by catoc on Wednesday March 15, 2023 @03:56PM (#35173854)
AGI clearly has a connotation of sentience and intent, of general intelligence - as opposed to a tool, however intricate. Read Nick Bostrom (yes I know, as a person he's cancelled, probably even rightly so, still scientifically relevant reference; and totally not a discussion I want to get into further here, pls)
Re:GPT4 and the Multi-Modal, Multi-Model, Multi-Everything F... by antibasilisk on Wednesday March 15, 2023 @04:31PM (#35174312)
But LLMs have been proven to demonstrate intent, and sentience is a philosophical question that doesn't really make any meaningful difference outside of questions of morality.
Re:GPT4 and the Multi-Modal, Multi-Model, Multi-Everything F... by catoc on Wednesday March 15, 2023 @04:53PM (#35174566)
I would say that the strings that are output by LLMs may fluently describe intent and purpose, but that actual intent requires a self awareness that utterly absent.

Just like the constant annoying 'apologies' do not in any way signify the model actually feels sorry.

Re:GPT4 and the Multi-Modal, Multi-Model, Multi-Everything F... by pixl97 on Wednesday March 15, 2023 @05:26PM (#35174950)
>Just like the constant annoying 'apologies' do not in any way signify the model actually feels sorry.

So your average corporation?

>I would say that the strings that are output by LLMs may fluently describe intent and purpose, but that actual intent requires a self awareness that utterly absent

The only reason I only partially/somewhat agree with this is current LLMs have no learning feedback loop as a means of short term actions to long term improvement. The moment that changes I'll take your statement and throw it in the trash bin of history. LLMs output have no (at least self-)purpose because they are just text output. The moment that text output goes to tooling, that tooling performs actions, and those actions are fed back into training the model then purpose is given. The self awareness loop is complete.

Re:GPT4 and the Multi-Modal, Multi-Model, Multi-Everything F... by antibasilisk on Wednesday March 15, 2023 @09:33PM (#35177066)
I don't agree that intent requires self-awareness.
Re:GPT4 and the Multi-Modal, Multi-Model, Multi-Everything F... by logifail on Wednesday March 15, 2023 @04:56PM (#35174606)
> LLMs have been proven to demonstrate intent [..]

That's quite the claim. Can you provide references?

Re:GPT4 and the Multi-Modal, Multi-Model, Multi-Everything F... by antibasilisk on Wednesday March 15, 2023 @09:31PM (#35177048)
There are earlier papers, but page 15 of the GPT-4 paper demonstrates this.
Re:GPT4 and the Multi-Modal, Multi-Model, Multi-Everything F... by bootsmann on Wednesday March 15, 2023 @04:57PM (#35174614)
>But LLMs have been proven to demonstrate intent

I'd like to see this proof. Everything we've seen so far is just the AI reproducing what humans do, which happens to be writing with intent.

Re:GPT4 and the Multi-Modal, Multi-Model, Multi-Everything F... by antibasilisk on Wednesday March 15, 2023 @09:32PM (#35177056)
The two are not meaningfully distinguishable on a materialistic level.