Bill Gates feels Generative AI has plateaued, says GPT-5 will not be any better::The billionaire philanthropist in an interview with German newspaper Handelsblatt, shared his thoughts on Artificial general intelligence, climate change, and the scope of AI in the future.
Cool, Bill Gates has opinions. I think he’s being hasty and speaking out of turn and only partially correct. From my understanding, the “big innovation” of GPT-4 was adding more parameters and scaling up compute. The core algorithms are generally agreed to be mostly the same from earlier versions (not that we know for sure since OpenAI has only released a technical report). Based on that, the real limit on this technology is compute and number of parameters (as boring as that is), and so he’s right that the algorithm design may have plateaued. However, we really don’t know what will happen if truly monster rigs with tens-of-trillions of parameters are used when trained on the entirety of human written knowledge (morality of that notwithstanding), and that’s where he’s wrong.
You got it the wrong way around. We already have a ton of compute and what this kind of AI can do is pretty cool.
But adding more compute power and parameters won’t solve the inherent problems.
No matter what you do, it’s still just a text generator guessing the next best word. It doesn’t do real math or logic, it gets basic things wrong and hallucinates new fake facts.
Sure, it will get slightly better still, but not much. You can throw a million times the power at it and it will still fuck up in just the same ways.
This is short-sighted.
The jump to GPT 3.5 was preceded by the same general misunderstanding (we’ve reached the limit of what generative pre-trained transformers can do, we’ve reached diminishing returns, ECT.) and then a relatively small change (AFAIK it was a couple additional layers of transforms and a refinement of the training protocol) and suddenly it was displaying behaviors none of the experts expected.
Small changes will compound when factored over billions of nodes, that’s just how it goes. It’s just that nobody knows which changes will have that scale of impact, and what emergent qualities happen as a result.
It’s ok to say “we don’t know why this works” and also “there’s no reason to expect anything more from this methodology”. But I wouldn’t dismiss further improvements as a forgone possibility.
Another way to think of this is feedback from humans will refine results. If enough people tell it that Toronto is not the capital of Canada it will start biasing toward Ottawa, for example. I have a feeling this is behind the search engine roll out.
ChatGPT doesn’t learn like that though, does it? I thought it was “static” with its training data.
You can finetune LLMs using smaller datasets, or with RLHF (reinforcement learning from human feedback) wherein people can give ratings to responses and the model can be either “rewarded” or “penalized” based off of the ratings for a given output. This retrains the LLM to produce outputs that people prefer.
Active Learning Models. Though public exposure can eaily fuck it up, without adult supervision. With proper supervision though, there’s promise.
So it will always have the biases of the supervisors
I was speculating about how you can overcome hallucinations, etc., by supplying additional training data. Not specific to ChatGPT or even LLMs…
Toronto is Canadian New York. It wants to be the capital and probably should be but it doesn’t speak enough French.
I mean, that’s more-or-less what I said. We don’t know the theoretical limits of how good that text generation is when throwing more compute at it and adding parameters for the context window. Can it generate a whole book that is fairly convincing, write legal briefs off of the sum of human legal knowledge, etc.? Ultimately, the algorithm is the same, so like you said, the same problems persist, and the definition of “better” is wishy-washy.
It will obviously get even better, but you’ll never be able to rely on it. Sure, 99.9% of that generated legal document will look perfect, till you overlook one sentence where the AI hallucinated. There is no fact checking in there, that’s the issue.
Yeah and I think he may be scaling to like true AGI. Very possible LLMs just don’t become AGI, you need some extra juice we haven’t come up with yet, in addition to computational power no one can afford yet.
Except that scaling alone won’t lead to AGI. It may generate better, more convincing text, but the core algorithm is the same. That “special juice” is almost certainly going to come from algorithmic development rather than just throwing more compute at the problem.
See my reply to the person you replied to. I think you’re right that there will need to be more algorithmic development (like some awareness of its own confidence so that the network can say IDK instead of hallucinating its best guess). Fundamentally though, llm’s don’t have the same dimensions of awareness that a person does, and I think that that’s the main bottleneck of human-like understanding.
My hypothesis is that that “extra juice” is going to be some kind of body. More senses than text-input, and more ways to manipulate itself and the environment than text-output. Basically, right now llm’s can kind of understand things in terms of text descriptions, but will never be able to understand it the way a human can until it has all of the senses (and arguably physical capabilities) that a human does. Thought experiment: Presumably you “understand” your dog - can you describe your dog without sensory details, directly or indirectly? Behavior had to be observed somehow. Time is a sense too. EDIT: before someone says it, as for feelings I’m not really sure, I’m not a biology guy. But my guess is we sense our own hormones as well
I’m not sure I’d say it’s plateaued today but I definitely think machine learning is going to hit a wall soon. Some tech keeps improving until physical limits stop progress but I see generative AI as being more like self-driving cars where the “easy” parts end up solved but the last 10% is insanely hard.
There’s also the economic reality of scaling. Maybe the “hard” problems could, in theory, be easily solved with enough compute power. We’ll eventually solve those problems but it’s going to be on Nvidia’s timeline, not OpenAI’s.
Generative ai is a bit different from self driving cars in the sense that they’re tolerant to failures. This may give more room for improvements when compared to other applications.
Maybe, but I am sure the tools the AIs can use will improve making the AIs jobs easier and thus the AI more efficient. I hope he is right tbh.
Eww, as a long time Linux user I need to take a shower now. I feel dirty.
The next big steps coming right now are AI trained on generative data and agents that act more automatically (rather than waiting for a prompt, take an action like searching the web and act on that to better complete the goal for example), and better indexed data so generated data can be informed by and cite sources in the moment.
I’ve been saying this for years!
I hope so. Theyve already got scary implications for creative parts of the economy.
That said, we’re in the Cambrian explosion of the tech. As it plateaus, the next step will be enhanced tooling and convenience around it. Better inputs than just text, better, more applications in new spaces, etc.
What does Bill Gates know about GenAI? Is he an expert on the subject?
deleted by creator
I think he could be right about generative AI, but that’s not a serious problem given we’re moving beyond generative AI and into virtual intelligence territory.
Generative ai right now requires someone (or something) to initiate it with a prompt, but according to some of the latest research papers in OpenAI as well as the drama that happened recently surrounding the leadership, it appears that we’re moving beyond the ‘generative’ phase into the ‘virtual intelligence’ phase.
It’s not going to be ‘smart’ it will be knowledgeable (and accurate, hopefully). That is to say VI’s will be useful as data retrieval or organization but not necessarily data creation (although IIRC the way to get around this would be to develop a VI that specifically only works on creating ideas but we’d be moving into AGI territory and I don’t expect we’ll have serious contenders for AGI for another decade at least).
The rumours abound surrounding the OpenAI drama, the key one being the potential for accidentally developing AGI internally (I doubt this heavily). The more likely reason is that the board of directors had a financial stake in Nvidia and when they found out altman was working on chips specifically for AI that were faster, lower cost, and lower power consumption than current nvidia trash (by literally tens of thousands of dollars), they fired him to try and force the company onto their preferred track (and profit in the process, which IMO, kind of ironic that a non-profit board of directors has so many ‘closed door’ discussions with nvidia staff…)
This is just the thoughts of a comp-sci student with a focus on artificial intelligence systems.
If interested in further reading:
https://www.ibm.com/blog/understanding-the-different-types-of-artificial-intelligence/
https://digitalreality.ieee.org/publications/virtual-intelligence-vs-artificial-intelligence
Keep in mind that because it’s still early days in this field that a lot of terms haven’t reached an established consensus across academia yet, so you’ll notice variations in how each organization explains what “x” type of intelligence is.
Sure bill.
I mean he she should know. Windows plateaued about 20 years ago.
Bill Gates views on AI are about as insightful as Gordon Ramsey’s on orbital mechanics.
Since I started using AI in March, it has not stop getting better and better with it seemingly accelerating not decelerating. If we are heading to a plateau, there are no signs of it yet. I am sure there will be plateaus, and maybe we have even had one earlier this year, but with the speed of AI development, it may mean a plateau of months versus the normal years. We will see. Bill has been saying a lot of things these past years that make no sense, unless you look at it from a money/market manipulation perspective.
And the Wright Brothers said heavier than air flight would only ever be an amusement for the rich, and never commercially viable.
Even taking Gates’ qualifications at face value doesn’t mean he’s actually right.
Source on that quote from the Wright brothers? Because they never said that as far as I’m aware.
They didn’t, AFAIK. It was a NYT article that quoted someone who made a similar prediction:
Once the Wright Brothers proved flight was possible, some assumed it was just a pointless rich play thing. Famed astronomer William H. Pickering said, “The expense would be prohibitive to any but the capitalist who could use his own yacht.”
You didn’t read the Wright brothers or this article did you? Gates isn’t at all damning AI tech. All he said is that gpt 5 is unlikely to be very different from 4. He’s probably correct. A next best word algorithm can only go so far. Thats only a part of how language and cognition works. Until some sort of adjunct algorithm gets tagged on I don’t think we will see big leaps either.
deleted by creator
And 640k ought to be enough for anybody.
He just has money, which gives him and too many others the idea that he has expertise.
He has money because he was a damn fine computer programmer, had some great ideas, and got pretty good at selling and monopolizing his product. He doesn’t “just have money”. He was skilled and intelligent. He may or may not be wrong about gpt, but he has a hell of a lot more knowledge about the subject and insight into gpt’s inner workings than probably anyone else on Lemmy.
Has he written a line of code in the past 30 years?
And a voice that should be heard
Isn’t this the same guy that said we won’t ever need more than 64K of RAM?
640k (the conventional limit in DOS), and he didn’t actually say it.
If it comes from Billyboy, he could be right in the sense that it won’t get any better for the filthy commoners
Bill gates can fuck off.
