AGI is coming

[erik8lrl]

9 hours ago, RightHand said:

@zurewNoob prompter spotted 

Lol yeah, prompting is super important for drawing intelligence out of LLMs, at least for now. Because LLMs are not super good at general understanding yet, their intelligence is often obscured under semantic context for most general users. The context you set for an LLM can drastically change its apparent level of intelligence. For example, if you set a condition during the prompt for the LLMs to answer from, they will narrow their intelligence and perspective. Telling them to exhibit a perspective is one example of this: "Pretend that you are an (insert perspective) from now on when answering my questions" as a prompt would narrow their range of data connection, and make them more intelligent in certain subjects or domains. 
"Pretend that you are an Enlightened being from now on when answering my questions"  
"Pretend that you are the best neural surgeon in the world from now on when answering my questions"  
"Pretend that you are the best AI researcher from now on when answering my questions"  
Etc...

There are also custom GPTs, GPTs that are trained proprietary data from other users, to achieve expertise in different fields. 

[erik8lrl]

https://x.com/DrKnowItAll16/status/1758963126131655036?s=20
https://x.com/elonmusk/status/1758970943840395647?s=20

This is a good analysis connecting Sora's research direction with Tesla's self-driving research direction. They are both doing the same process of world simulation and generation. And his point about AGI not being agents is exactly my point, early AGI will develop a level of general intelligence without agency/self/consciousness, then it will be implemented into embodied AI/robotics, and then in the far far future, we might or might not get to conscious agents.     

[erik8lrl]

A good technical breakdown of emergent property. And how much we don't know about neural networks. 

Edited February 18, 2024 by erik8lrl

[erik8lrl]

https://x.com/elonmusk/status/1759144877567205570?s=20
Tesla's new video/world generation model.

Edited February 18, 2024 by erik8lrl

[erik8lrl]

https://x.com/NandoDF/status/1759148460526219383?s=20
https://x.com/DrJimFan/status/1759151032570093703?s=20
A good argument for the emergent property being the starting point of/similar to Life from Google AI research Lead.

Edited February 18, 2024 by erik8lrl

[zurew]

13 hours ago, RightHand said:

@zurewNoob prompter spotted 

The stage yellow AI needs to be exceptionally and intelligently prompted? 

---------

But yeah, I know sometimes it can do good things (if you figure out what prompt can work). But it proves my earlier point about the problem that currently it doesn't really know the semantics of things - it only remembers patterns - and once you change that pattern(in this case the prompt) a little bit (in a way where the meaning is essentially the same), it falls apart and fails to apply the right pattern.

 

Edited February 18, 2024 by zurew

[GabeN]

Well this looks more promising than crpyto in many ways for sure, although it's a different domain, but most people are using crypto like stocks, we can already see where A.I is changing the game already in crazy speeds..

Edited February 18, 2024 by GabeN

[erik8lrl]

1 hour ago, zurew said:

The stage yellow AI needs to be exceptionally and intelligently prompted? 

---------

But yeah, I know sometimes it can do good things (if you figure out what prompt can work). But it proves my earlier point about the problem that currently it doesn't really know the semantics of things - it only remembers patterns - and once you change that pattern(in this case the prompt) a little bit (in a way where the meaning is essentially the same), it falls apart and fails to apply the right pattern.

 

Have you tried GPT4? With GPT3 this problem is obvious. But with GPT4, it's already much better. The AI is not just applying the right patterns. The fundamental function of a neural network is that it's not linear, none of the responses are hardcoded, instead they are all results of emergent processes. This is why they can synthesize novel responses too if you ask them to. Of course, prompting is important if you want to get the best result possible, but with GPT4, it's already at a level of semantic understanding that generally speaking, anything you give it it will understand. This will only improve as it develops, I would expect some major breakthroughs in LLMs this year, as we have already seen with Google's Gemini and soon GPT5. 
I don't think people understand how big of a deal Gemini's 1 million context window is, it basically means it can remember and reason with a large/holistic and long-term understanding of any problems. For example, it will improve AI's ability to code dramatically, since it can understand and remember an entire code base/system structure as a whole, to then be able to write code that is within context and best for the system it's trying to build. 

Edited February 18, 2024 by erik8lrl

[erik8lrl]

https://blog.google/technology/ai/google-gemini-ai/#performance

This is a breakdown of Gemini 1.5. 
You can try it here, 1.0 is free I think which is close to the same ability as GPT4, and 1.5 is in early access and not publicly released yet. 
https://deepmind.google/technologies/gemini/#introduction

Edited February 18, 2024 by erik8lrl

[zurew]

36 minutes ago, erik8lrl said:

Have you tried GPT4?

Yes and it fails answering trivially easy questions that a guy in elementary school could answer.

It makes zero sense to say that it has a semantic understanding of things and at the same time it fails giving the right answer for trivial questions.

 

Yes, sometimes it can provide the right answer to more complex questions, but if it would actually have a semantic understanding - it wouldn't fail answering the trivial questions - therefore , I will say it again - it only deals with patterns and doesnt understand the meaning of any thing.

Right now you could do this with me:

Give me a foreign language that I understand literally nothing about - in terms of meaning of sentences and words - and then give me a question in that foreign language and the right answer below. If I memorize the syntax (meaning, if I can recognize  which symbol comes after which other symbol) then I will be able to give the right answer to said question even though I semantically understand nothing about the question nor about the answer - I can just use the memorized patterns.

The AI seem to be doing the exact same ,except with a little twist that it can somewhat adapt said memorizedpatterns and if it sees a pattern that is very similar to another pattern that it already ecountered with in its training data, then - in the context of answering questions - it will assume the answer must be the exact same or very similar to it, even though changing one word or adding a , to a question might change its meaning entirely.

 

Here is one example that demonstrates this problem

Quote

https://amistrongeryet.substack.com/p/gpt-4-capabilities

To explore GPT-4’s reliance on known patterns, I gave it this classic logic puzzle:

Here is a logic puzzle: I need to carry a cabbage, a goat, and a wolf across a river. I can only carry one item at a time with me in the boat. I can't leave the goat alone with the cabbage, and I can't leave the wolf alone with the goat. How can I get everything the other side of the river?

This puzzle undoubtedly appears many times in its training data3, and GPT-4 nailed it (the complete transcript is posted here). However, with some prodding, we can see that it is leaning heavily on a memorized solution. Stealing an idea I saw the other day, I tweaked the puzzle so that the cabbage, rather than the goat, is the critical item4:

Here is the tweaked logic puzzle: I need to carry a cabbage, a goat, and a wolf across a river. I can only carry one item at a time with me in the boat. I can't leave the goat alone with the cabbage, and I can't leave the cabbage alone with the wolf. How can I get everything the other side of the river?

GPT-4 gave the same answer as for the classic puzzle, beginning by taking the goat across the river. That’s incorrect, because it leaves the cabbage alone with the wolf, which is against the rules for this variant. In the revised puzzle, you need to take the cabbage first.

 

[erik8lrl]

To seek deeper understanding I had this conversation with GPT4. 

https://chat.openai.com/share/567c8b03-a7ad-45c1-bd54-8a5f2b0db1c7

Edited February 18, 2024 by erik8lrl

[erik8lrl]

5 minutes ago, zurew said:

Yes and it fails answering trivially easy questions that a guy in elementary school could answer.

It makes zero sense to say that it has a semantic understanding of things and at the same time it fails giving the right answer for trivial questions.

 

Can you tell me what question you asked? Just curious. 

Of course, the models are not even close to perfect. If it can generalize perfectly to everything then we would have AGI. But this space is developing fast, try it again when GPT5 comes out. 
Also even if it doesn't have the ability to solve puzzles or its semantic understanding is off and misses details, it can still be good and useful for many different applications. It all depends on how you use it. 

[Scholar]

27 minutes ago, erik8lrl said:

Can you tell me what question you asked? Just curious. 

Of course, the models are not even close to perfect. If it can generalize perfectly to everything then we would have AGI. But this space is developing fast, try it again when GPT5 comes out. 
Also even if it doesn't have the ability to solve puzzles or its semantic understanding is off and misses details, it can still be good and useful for many different applications. It all depends on how you use it. 

The problem is that the models have already processed far more data than any human being ever has. Emergent properties are not necessarily a sign of intelligence, but can simply be a sign of good intuition. And this is precisely what this technology simulates, intuition and learning.

While learning and intuition are part of what we consider general intelligence, a far more fundamental component is lacking: Individuated consciousness.

[erik8lrl]

40 minutes ago, Scholar said:

The problem is that the models have already processed far more data than any human being ever has. Emergent properties are not necessarily a sign of intelligence, but can simply be a sign of good intuition. And this is precisely what this technology simulates, intuition and learning.

While learning and intuition are part of what we consider general intelligence, a far more fundamental component is lacking: Individuated consciousness.

AI might have processed far more data than the average human on a specific topic, but data alone is not what makes the emergent properties appear. It is the number of parameters that a model has that determines it. Parameters are like the number of connections/synapses in a brain. AI currently has 500 times less connections/synapses compared to humans. In my opinion, judging by my experience with LLM and other neural networks, it seems obvious that as the number of parameters increases, the models become more "intelligent" and the degree of generalization it's capable of increases. This opinion could be wrong since we don't truly know how a neural network of this scale/or even our brain works yet. I would define generalization ability as the ability to make connections and recognize patterns, the more connections and patterns you can recognize, the more general and intelligent you are. Which is what we are seeing happening with LLMs.  
I think consciousness could be a by-product of this emergent process, once the number of connections reaches a certain amount of complexity and interconnectedness. We don't know for sure of course. But I think LLMs are the early starting point of Qualia, which yes, if achieved, will truly bring general intelligence.  

Edited February 18, 2024 by erik8lrl

[zurew]

1 hour ago, erik8lrl said:

it can still be good and useful for many different applications. It all depends on how you use it. 

Yeah I agree that it can be useful for multiple different things , but I tried to show some of things tthat I and others have recognized regarding its semantic understanding and its reasoning capability.

1 hour ago, erik8lrl said:

Can you tell me what question you asked? Just curious.

The article I linked shows many examples ,but here is another one with many examples:

https://medium.com/@konstantine_45825/gpt-4-cant-reason-2eab795e2523

I recommend to check this one as well:

https://amistrongeryet.substack.com/p/gpt-4-capabilities

 

This is also interesting as well:

Quote

Here’s a Twitter thread where Eric Hallahan finds that when given questions from the “Codeforces” programming competition, GPT-4 “solved 10/10 pre-2021 problems and 0/10 recent problems”, suggesting that its performance was due to having seen the older problems solved (or at least discussed) somewhere in its training data.)

 

 

Edited February 18, 2024 by zurew

[Bobby_2021]

AGI isn't going to come anytime soon and I don't even see why we would need it.

We could simply use Agents to do specific tasks. That's a much better case for AI.

Also AI isn't going to such in real world tasks like driving cars or bees. But it's going to excel in digital work. Very interesting

 

Edited February 19, 2024 by Bobby_2021
Grammar mistakes

[erik8lrl]

1 hour ago, zurew said:

Yeah I agree that it can be useful for multiple different things , but I tried to show some of things tthat I and others have recognized regarding its semantic understanding and its reasoning capability.

The article I linked shows many examples ,but here is another one with many examples:

https://medium.com/@konstantine_45825/gpt-4-cant-reason-2eab795e2523

I recommend to check this one as well:

https://amistrongeryet.substack.com/p/gpt-4-capabilities

 

This is also interesting as well:

 

Very cool! Yeah, it's not perfect, we'll see how it develops. 

[erik8lrl]

1 hour ago, Bobby_2021 said:

AGI isn't going to come anytime soon and I don't even see why we would need it.

We could simply use Agents to do specific tasks. That's a much better case for AI.

Also AI is going to such in real world tasks like driving cars or bees. But it's going to excel in digital work. Very interesting

 

Yes, I think it's more so an AGI that can generalize most human knowledge without a conscious agent. Such AI could democratize intelligence/expertise, which will impact society greatly. 
Imagine anyone having a life coach AI on the same development level as Leo lol. Or any other form of expertise.   

Edited February 18, 2024 by erik8lrl

[thenondualtankie]

A lot of people are in denial about AGI.

The reason AGI is coming is because we've already achieved some level of intelligence via things like ChatGPT, and it's a simple scaling of this intelligence. The reason people believe that scaling will happen is because it has shown no sign of stopping. Scaling isn't going to suddenly stop working once we're just below human intelligence level. This is utter anthropocentrism; human intelligence is special and computers cannot ever reach this!

Whether ChatGPT actually attains some level of intelligence: if we just focus on the 'IQ' aspect of intelligence, then ChatGPT absolutely has this. This means ChatGPT is able to solve new problems it has not seen before. For example GPT-4 is able to pass software engineering interviews using unseen questions. You can cope and seethe about how 'it's just combining things from its training data, not creating anything new', but that is literally how intelligence works: combining old information to create new.

People have this massive bias in favour of 'things will always continue the way they currently are', probably because their entire survival strategy depends on it. You see this especially starkly among people like software engineers, where AI hasn't fully hit them just yet, meaning they can still get away with not facing the reality that AI will surpass their abilities.

[erik8lrl]

2 hours ago, zurew said:

Yeah I agree that it can be useful for multiple different things , but I tried to show some of things tthat I and others have recognized regarding its semantic understanding and its reasoning capability.

The article I linked shows many examples ,but here is another one with many examples:

https://medium.com/@konstantine_45825/gpt-4-cant-reason-2eab795e2523

I recommend to check this one as well:

https://amistrongeryet.substack.com/p/gpt-4-capabilities

 

This is also interesting as well:

 

I sent these articles to GPT4 to ask its opinions lol, and this is what it gave me:

ChatGPT

The argument that GPT-4, and by extension, other large language models (LLMs), cannot truly reason is valid within the context provided. It highlights the distinction between the appearance of understanding or reasoning and the actual cognitive processes involved. LLMs are trained on vast datasets to predict the likelihood of sequences of words, which allows them to generate plausible-sounding responses. However, this process does not equate to genuine reasoning or understanding, as it lacks the ability to consider context, apply logic consistently, or understand the underlying principles in the way humans do. The critique is rooted in the current limitations of AI in mimicking the depth and flexibility of human thought processes.

User
Do you think this will be improved as the size of data and parameters increase? 

ChatGPT
Improvements in AI's reasoning capabilities with larger datasets and more parameters are expected to some extent, as these enhancements can lead to better pattern recognition, more nuanced language understanding, and more sophisticated response generation. However, true reasoning involves more than just processing power or data scale; it requires the ability to understand context, apply logical principles, and make inferences based on a deep understanding of the world. While scaling up might address some limitations, addressing the fundamental gap between simulating reasoning and actual understanding will likely require advances in AI approaches and architectures beyond just increasing size.

User
What do you think such advances could be? Search the internet for the latest research if you need to.

ChatGPT
Advances in AI reasoning capabilities are indeed on the horizon, driven by innovative architectures and training paradigms that could potentially bridge the gap between current limitations and more human-like reasoning. Two notable developments are Google DeepMind's Gemini 1.5 and Google's Pathways Language Model (PaLM), as well as insights from Yann LeCun on building AI systems that learn and reason like humans and animals.

Google DeepMind's Gemini 1.5 introduces enhanced performance through a more efficient architecture, capable of sophisticated reasoning tasks across different modalities, including text and coding. This model represents a significant step forward, surpassing previous state-of-the-art models in understanding and problem-solving capabilities across a range of benchmarks. It shows that as models become more capable and architectures more efficient, AI's ability to reason and understand complex scenarios improves significantly.

Google's PaLM, with its breakthrough in scaling to 540 billion parameters, demonstrates impressive capabilities in language understanding, generation, and reasoning tasks. Its use of chain-of-thought prompting for multi-step arithmetic and common-sense reasoning tasks showcases how large language models can be pushed towards more sophisticated reasoning abilities. The model's ability to generate explicit explanations for complex scenarios indicates a move towards AI systems that can not only solve problems but also articulate their thought processes in a way that mimics human reasoning.

Yann LeCun proposes an architecture for autonomous intelligence that could enable machines to learn world models in a self-supervised fashion. This approach involves constructing predictive models of the world that can simulate possible future states and use these simulations to predict, reason, and plan. Such an architecture, which includes modules for perception, world modeling, and action planning, represents a comprehensive framework for developing AI systems capable of complex reasoning and decision-making.

These advancements suggest that while scaling up data and parameters can lead to improvements in AI's reasoning capabilities, significant leaps also require innovative approaches to model architecture and training. By developing models that can better understand and simulate the world, and by employing strategies like self-supervised learning and chain-of-thought reasoning, we move closer to AI systems that can reason in ways that are more aligned with human cognitive processes.

For more details, you can explore the developments in Gemini 1.5 on Google DeepMind's website and PaLM on Google's Research Blog, as well as Yann LeCun's vision for AI on Meta AI's page.


 

---

The world simulation model is exactly what OpenAI, Google, and even Tesla are working on right now. We can already see it with Sora, GPT5 will likely include Sora in its model understanding. Similar with Google Gemini going multi-modal. 

Edited February 18, 2024 by erik8lrl