After reading about o3's performance on ARC-AGI, I strongly suspect people will not be so flippantly dismissive of the inherent limits of these technologies by this time next year. I'm genuinely surprised at how myopic HN commentary is on this topic in general. Maybe because the implications are almost unthinkably profound.
Anyway, OpenAI, Anthropic, Meta, and everyone else are well aware of these types of criticisms, and are making significant, measurable progress towards architecturally solving the deficiencies.
People are flippantly dismissive of the inherent limits because there ARE inherent limitations of the technology.
> Maybe because the implications are almost unthinkably profound.
Maybe because the stuff you're pointing to are just benchmarks and the definitions around things like AGI are flawed (and the goalposts are constantly moving, just like the definition of autonomous driving). I use LLMs roughly 20-30x a day - they're an absolutely wonderful tool and work like magic, but they are flawed for some very fundamental reasons.
1. What does inherit limitations mean?
2. How do we know something is an inherit limitation
3. Is it a problem if arguments for a particular inherit limitation also apply to humans?
From what I’ve seen people will often say things like AI can’t be creative because it’s just a statistical machine, but humans are also “just” statistical machines. People might mean something like humans are more grounded because humans react not just to how the world already works but how the world reacts to actions they take, but this difference misunderstands how LLMs are trained. Like humans LLMs get most of their training from observing the world, but LLMs are also trained with re-enforcement learning and this will surely be an active area of research.
One of many, but this is a simple one - LLMs are only limited to knowledge that is publicly available on the internet. This is "inherit" because thats how LLMs are essentially taught the information they retrieve today.
Therefore AI has to be much better than humans at the task to be considered ready to be a replacement.
——
Today robot taxis can only work in fair weather conditions in locations that are planned cities. No autonomous driving system can drive in Nigeria or India or even many european cities that were never designed for cars any time soon .
Working in very specific scenarios is useful , but hardly measure of their intelligence or candidate for replacing humans for the task
But the point still stands that these systems can't be treated as deterministic (i.e. reliable or trustworthy) for the purposes of carrying out tasks that you can't allow "brute forced attempts" for (e.g. anything where the desired outcome is a positive subjective experience for a human).
A new architecture is going to be needed that actually does something closer to our inherently heuristic based learning and reasoning. We'll still have the stochastic problem but we'll be moving further away from the idiot savant problem.
All of this being said, I think there's plenty of usefulness with current LLMs. We're just expecting the wrong things from them and therefore creating suboptimal solutions. (Not everyone is, but the most common solutions are, IMO.)
The best solutions need to be rethinking how we typically use software since software has been hinged upon being able to expect (and therefore test) dertiministic outputs from a limited set of user inputs.
I work for an AI company that's been around for a minute (make our own models and everything). I think we're both in an AI hype bubble while simultaneously underestimating the benefits of current AI capabilities. I think the most interesting and potentially useful solutions are inherently going to be so domain specific that we're all still too new at realizing we need to reimagine how to build with this new tech in mind. It reminds me of the beginning of mobile apps. It took awhile for most us to "get it".
LLMs / Generative Models can have a profound societal and economic impact without being intelligent. The obsession with intelligence only make their use haphazard and dangerous.
It is a good thing court of laws have established precedent that organizations deploying LLM chatbots are responsible for their output (Eg, Air Canada LLM chatbot promising a non-existent discount being responsibility of Air Canada)
Also most automation has been happening without LLMs/Generative Models. Things like better vision systems have had an enormous impact with industrial automation and QA.
It’s in area where we demand correctness and determinism that they will not be suitable.
I think the thrust of this article is hard to see unless you have some experience with formal methods and verification. Or else accept the authors’ explanations as truth.
If I wasn't so slammed with work I have half a mind to go dredge up at least a dozen posts that said the same thing last year, and the year before. Even OpenAI has been moving the goalposts here.
I heard that people still believing in OpenAI hype exist but I haven't met any IRL.
If OpenAI has demonstrated one thing is that they are a hype production machine and they are probably getting ready for next round of investment. I wouldn't be surprised if this model was equally useless as o1 when you factor in performance and price.
At this point they are completely untrustworthy and untill something lands publicly for me to test it's safe to ignore their PR as complete BS.
For most tasks - but not all. I normally paste my prompt in both and while Claude is generally superior in most aspects, there are tasks at which o1 performed slightly better.
”If intelligence lies in the process of acquiring new skills, there is no task X that solving X proves intelligence”
IMO it especially applies to things like solving a new IQ puzzle, especially when the model is pretrained for that particular task type, like was done with ARC-AGI.
For sure, it’s very good research to figure out what kind of tasks are easy for humans and difficult for ML, and then solve them. The jump in accuracy was surprising. But still in practice the models are unbeliavably stupid and lacking in common sense.
My personal (moving) goalpost for ”AGI” is now set to whether a robot can keep my house clean automatically. Its not general intelligence if it can’t do the dishes. And before physical robots, being less of a turd at making working code would be a nice start. I’m not yet convinced general purpose LLMs will lead to cost-effective solutions to either vs humans. A specifically built dish washer however…
The dumbest intern doesn't do that.
Which is the entire point of the article that your comment fail to address.
It still can't think and it won't think.
LANGUAGE models (keyword: language) is a language model, it should be paired with a reasoning engine to translate the inner thought of the machine into human language. It should not be the source of decisions because it sucks at doing so, even though the network can exhibit some intelligence.
We will never have AGI with just a language model. That said, most jobs people do are still at risk, even with chatgpt-3.5 (especially outside of knowledge work, where difficult decisions need to be taken). So we'll see the problems with AGI and the job market way earlier than AGI, as soon as we apply robotics and vision models + chatgpt 3.5 level intelligence. Goodbye baristas, goodbye people working in factories.
Let's start working on a reasoning engine so we can replace those pesky knowledge workers too.
laundering stolen IP from actual human artists and researchers, extinguishing jobs, deflecting responsibility for disasters. yeah, I can't wait for these "profound implications" to come to fruition!
Every critique of AI assumes to some degree that contemporary implementations will not, or cannot, be improved upon.
Lemma: any statement about AI which uses the word "never" to preclude some feature from future realization is false.
There is likely a couple of surprised still left in LLMs but no one should think that any present technology in its current state or architecture will get us to AGI or anything that remotely resembles it.
Expecting them to do non-trivial amounts of technical or mathematical reasoning, or even something as simple as code generation (other than "translate these complex natural-language requirements into a first sketch of viable computer code") is a total dead end; these will always be language systems first and foremost.
If the tokens are bit-for-bit-identical, where does the non-determinism come in?
If the tokens are only roughly-the-same-thing-to-a-human, sure I guess, but convergence on roughly the same output for roughly the same input should be inherently a goal of LLM development.
Whatever "random" seed was used can be reused.
By design, most LLM’s have a randomization factor to their model. Some use the concept of “temperature” which makes them randomly choose the 2nd or 3rd highest ranked next token, the higher the temperature the more often/lower they pick a non-best next token. OpenAI described this in their papers around the GPT-2 timeframe IIRC.
i love how those changes are often just a different seed in the randomness... as just chance.
run some repeated tests with "deeper than surface knowledge" on some niche subjects and got impressed that it gave the right answer... about 20% of the time.
(on earlier openAI models)
If only it was something which we could ontologically map onto existing categories like servants or liars...
1) We're continually learning so we can update our predictions when our pattern matching is wrong
2) We're autonomous - continually interacting with the environment, and learning how it respond to our interaction
3) We have built in biases such as curiosity and boredom that drive us to experiment, gain new knowledge, and succeed in cases where "pre-training to date" would have failed us
For a person to truly understand something they will have a well-refined (as defined by usefulness and correctness), malleable internal model of a system that can be tested against reality, and they must be aware of the limits of the knowledge this model can provide.
Alone, our language-oriented mental circuits are a thin, faulty conduit to our mental capacities; we make sense of words as they relate to mutable mental models, and not simply in latent concept-space. These models can exist in dedicated but still mutable circuitry such as the cerebellum, or they can exist as webs of association between sense-objects (which can be of the physical senses or of concepts, sense-objects produced by conscious thought).
So if we are pattern-matching, it is not simply of words, or of their meanings in relation to the whole text, or even of their meanings relative to all language ever produced. We translate words into problems, and match problems to models, and then we evaluate these internal models to produce perhaps competing solutions, and then we are challenged with verbalizing these solutions. If we were only reasoning in latent-space, there would be no significant difficulty in this last task.
AI can only interpolate. We may perceive it as extrapolation, but all LLMs architectures are fundamentally cleverly designed lossy compression
"The distinction between AI and humans often comes down to the concept of understanding. You’re right to point out that both humans and AI engage in pattern matching to some extent, but the depth and nature of that process differ significantly." "AI, like the model you're chatting with, is highly skilled at recognizing patterns in data, generating text, and predicting what comes next in a sequence based on the data it has seen. However, AI lacks a true understanding of the content it processes. Its "knowledge" is a result of statistical relationships between words, phrases, and concepts, not an awareness of their meaning or context"
:)
I didn't downvote, I'm just saying why I think you were downvoted.
We don't care what LLMs have to say, whether you cut back some of it or not it's a low effort wasted of space on the page.
This is a forum for humans.
You regurgitating something you had no contribution in producing, which we can prompt for ourselves, provides no value here, we can all spam LLM slop in the replies if we wanted, but that would make this site worthless.
I use llms as tools to learn about things I don't know and it works quite well in that domain.
But so far I haven't found that it helps advance my understanding of topics I'm an expert in.
I'm sure this will improve over time. But for now, I like that there are forums like HN where I may stumble upon an actual expert saying something insightful.
I think that the value of such forums will be diminished once they get flooded with AI generated texts.
(Fwiw I didn't down vote)
That was the point. If you back up to the comment I was responding to, you can see the claim was: "maybe people are doing the same thing LLMs are doing". Yet, for whatever reason, many users seemed to be able to pick out the LLM comment pretty easily. If I were to guess, I might say those users did not find the LLM output to be human-quality.
That was exactly the topic under discussion. Some folks seem to have expressed their agreement by downvoting. Ok.
Other parts of what we do looks more as a search through the space of possibilities.
And then we act and collaborate and test the ideas that stand against scrutiny.
All of that is in principle doable by machines. The things we currently have and we call LLMs seem to currently mostly address the autocomplete part although they begin to be augmented with various extensions that allow them to take baby steps in other fronts. Will they still be called large language models once they will have so many other mechanisms beyond the mere token prediction?
It’s almost like all the thought leading that proclaimed the death of software eng was nothing but self-promotional noise. Huh, go figure.
Where I see software companies using it most is as a replacement for interns and junior devs. That replacement means we're not training up the next generation to be the senior or expert engineers with real world experience. The industry will feel that badly at some point unless it gets turned around.
That said, combining multiple ais and multiple programs together may mitigate this.
If I'm trying to use some tool that just got released or just got a big update, I wont use AI, if I want to check the syntax of a for loop in a language I don't know I will. Whenever you ask it a question you should have an idea in your mind of how likely you are to get a good answer back.
I saw an interesting example yesterday of type "I have 3 apples, my dad has 2 more than me ..." where of the top 10 predicted tokens, about 1/2 led to the correct answer, and about 1/2 didn't. It wasn't the most confident predictions that lead to the right answer - pretty much random.
The trouble with LLMs vs humans is that humans learn to predict facts (as reflected in feedback from the environment, and checked by experimentation, etc), whereas LLMs only learn sentence soup (training set) word statistics. It's amazing that LLM outputs are coherent as often as they are, but entirely unsurprising that they are often just "sounds good" flow-based BS.
Your apples question is the same, its not knowledge, it's a calculation, it's intelligence. The only time you're going to get intelligence from AI at the moment is to ask a question that a significantly large number of people have already answered.
To make things worse, I don't think we can even assume that primary facts are always going to be represented in abstract semantic terms independent of source text. The model may have been trained on a fact but still fail to reliably recall/predict it because of "lookup failure" (model fails to reduce query text to necessary abstract lookup key).
They're not that capable. They're just bullshit artists.
LLM = LBM (large bullshit models).
Even with ChatGPT you can ask it to find web citations and if it uses the Python runtime to find answers, you can look at the code.
And to prevent the typical responses - my company uses GSuite so Google already has our IP, NotebookLM is specifically approved by my company and no Google doesn’t train on your documents
How do you, in general, fact check a chain of reasoning?
I can’t tell a search engine to summarize text for a technical audience and then another summary for a non technical audience.
I recently came into the middle of a cloud consulting project where a lot of artifacts, transcripts of discovery sessions, requirement docs, etc had already been created.
I asked NotebookLM all of the questions I would have asked a customer at the beginning of a project.
What it couldn’t answer, I then went back and asked the customer.
I was even able to get it to create a project plan with work streams and epics. Yes it wouldn’t have been effective if I didn’t already know project management, AWS and two decades+ of development experience.
Despite what people think, LLMs can also do a pretty good job at coding when well trained on the APIs. Fortunately, ChatGPT is well trained on the AWS CLI, SDKs in various languages and you can ask it to verify the SDK functions on the web.
I’ve been deep into AWS based development since LLMs have been a thing. My opinion may change if I get back into more traditional development
No, but, as amazing as that is, don't put too much trust in those summaries!
It's not summarizing based on grokking the key points of the text, but rather based on text vs summary examples found in the training set. The summary may pass a surface level comparison to the source material, while failing to capture/emphasize the key points that would come from having actually understood it.
Just like I’m not randomly depending on it to do an Amazon style PRFAQ (I was indoctrinated as an Amazon employee for 3.5 years), create a project plan, etc, without being a subject matter expert in the areas. It’s a tool for an experienced writer, halfway decent project manager, AWS cloud application architect and developer.
There is an entire “reproducibility crisis” with research.
Try training an LLM.
Right noe there's no incentive though. People keep paying good money to use these tools despite their hallucinations, aka lies/gas lighting/fake information. As long as users don't stop paying and LLM companies don't have business pressure to lean on accuracy as a market differentiator, no one is going to bother fixing it.
It's inherit to transformers that they predict the next most likely token, its not possible to change that behavior without making them useless at generalizing tasks (overfitting).
LLMs run on statistics, not logic. There is no fact checking, period. There is just the next most likely token based on the context provided.
I wouldn't expect them to add an additional LLM layer unless hallucinations from the underlying LLM aren't acceptable, and in this case that means it is unacceptable enough to cost them users and money.
Adding a check/audit layer, even if it would work, is expensive both financially and computationally. I'm not sold that it would actually work, but I just don't think they've had enough reason to really give it a solid effort yet either.
Edit: as far as fact checking, I'm not sure why it would be impossible. An LLM wouldn't likely be able to run a check against a pre-trained model of "truth," but that isn't the only option. An LLM should be able to mimic what a human would do, interpret the response and search a live dataset of sources considered believable. Throw a budget of resources at processing the search results and have the LLM decide if the original response isn't backed up, or contradicts the source entirely.
I can ask the computer "hey I know this thing exists in your training data, tell me what it is and cite your sources." This is awesome. Seriously.
But what that means is you can ask it for sample code, or to answer a legal question, but fundamentally you're getting a search engine reading something back to you. It is not a programmer and it is not a lawyer.
The hype train really wants to exaggerate this to "we're going to steal all the jobs" because that makes the stock price go up.
They would be far less excited about that if they read a little history.
It won't steal them all, but it will have a major impact by stealing the lower level jobs which are more routine in nature -- but the problem is that those lower level jobs are necessary to gain the experience needed to get to the higher level jobs.
It also won't eliminate jobs completely, but it will greatly reduce the number of people needed for a particular job. So the impact that it will have on certain trades -- translators, paralegals, journalists, etc. -- is significant.
My workflow typically starts with asking ChatGPT to analyze a webpage where I need to authenticate. I guide it to identify the username and password fields, and it accurately detects the credential inputs. I then inform it about the presence of a session cookie that maintains login persistence. Next, I show it an example page with links—often paginated with numbered navigation at the bottom—and ask it to recognize the pattern for traversing pages. It does so effectively.
I further highlight the layout pattern of the content, such as magnet links or other relevant data presented by the CMS. From there, I instruct it to generate a Python script that spiders through each page sequentially, navigates to every item on those pages, and pushes magnet links directly into Transmission. I can also specify filters, such as only targeting items with specific media content, by providing a sample page for the AI to analyze before generating the script.
This process demonstrates how effortlessly AI enables coding without requiring prior knowledge of libraries like beautifulsoup4 or transmission_rpc. It not only builds the algorithm but also allows for rapid iteration. Through this exercise, I assume the role of a manager, focusing solely on explaining my requirements to the AI and conducting a code review.
LLM-as-search is essentially the hand-tuned expert systems AI vs deep learning AI battle all over again.
Between natural language understanding and multiple correlations, it's going to scale a lot further than previous search approaches.
> idiot savants with vast and unreliable intelligence.
Remember, intelligence !== knowledge. These LLMs indeed have vast and unreliable knowledge banks.
EDITED My ASCI art pyramid did not work. So imagine a pyramid with DATA at the bottom, INFORMATION on top of the data, and KNOWLEDGE sitting on top of the INFORMATION, with WISDOM at the top.
And then trying top guess where AI is? Some people say that Information is the knowing, what, knowledge the how, and Wisdom the why.
If we want to get pedantic, I would point out that “knowledge” is formally defined as “justified true belief”, and I doubt we want to get into the quagmire of whether LLM’s actually have beliefs.
I took OP’s point in the casual meaning, i.e. that LLMs are like what I would call an “intelligent coworker”, or how one might call a Jeopardy game show contestant as intelligent.
I would say "knowledge" rather than "intelligence"
The key feature of LLMs is the vast amounts of information and data they have access to, and their ability to quickly process and summarize, using well-written prose, that information based on pattern matching.
LLMs will likely never get us to 100% solutions on a large class of problems.
But! A lot of problems can be converted into versions with a subcomponent that LLMs can solve 100% of.
And the fusion of LLMs doing 100% of that subportion + humans doing the remainder = increased productivity.
Re-engineering problems to be LLM-tolerant, then using LLMs to automate that portion of the problem, is the winning approach.
That's quite the embarassment if you actually mean it.
> A group of individuals adept with use of such an idiot savant enhanced environment would be incredibly capable. They'd be a force unseen in human civilization before today
More than the people who landed someone on the moon?
One of the obvious uses for current LLMs is as a smarter search tool against static knowledge collections.
Turns out, this is a real world problem in a lot of "fuzzy decision" scenarios. E.g. insurance claim adjudication
Status quo is to train a person over enough years that they can make these decisions reliably. (Because they've internalized all the documentation)
Pretty close to the idea of human brainstorming and has worked. Could it do orbital math? Maybe not today but the approach seems as feasible as the work Mattingly did for Apollo 13.
I recommend reading his interview with Matteo Wong, where he proposes the opposite: https://www.theatlantic.com/technology/archive/2024/10/teren...
> With o1, you can kind of do this. I gave it a problem I knew how to solve, and I tried to guide the model. First I gave it a hint, and it ignored the hint and did something else, which didn’t work. When I explained this, it apologized and said, “Okay, I’ll do it your way.” And then it carried out my instructions reasonably well, and then it got stuck again, and I had to correct it again. The model never figured out the most clever steps. It could do all the routine things, but it was very unimaginative.
I agree with his overall vision, but transformer-based chatbots will not be the AI algorithm that supports it. Highly-automated proof assistants like Isabelle's Sledgehammer are closer (and even those are really, really crude, compared to what we could have).
1. LLMs have little value, are totally unreliable, will never amount to much because they don’t learn and grow and mature like people do., so they cannot replace a person like me who is well advanced in a career.
2. LLMs are incredible useful and will change the world because they excel at entry level work and can replace swaths of relatively undifferentiated information workers. LLM flaws are not that different from those workers’ flaws.
I’m in camp 2, but I appreciate and agree with the articulation of why they will not replace every information worker.
Basically this. They already have vastly better-than-human ability at finding syntax errors within code, which on its own is quite useful; think of how many people have probably dropped out of CS as a major after staying up all night and failing to find a missing semicolon.
These will do things like highlight places where you're trying to call a method that isn't defined on the object, but they don't understand the intent of what you're trying to do. The latter is actually important in terms of being able to point you toward the correct solution.
I started out in Cursor, but I quickly realized Claude's erudite knowledge of AWS would not help me here, but what I needed was to refactor the code quickly and often, so that I'd finally find the perfect structure.
For that, IDE tools were much more appropriate than AI wizardry.
The foundation of a computer science education is a rigorous understanding of what the steps of an algorithm mean. If the students don't develop that, then I don't think they're doing computer science anymore.
I mean if the alternative is quitting entirely because they can't see that they've mixed tabs with spaces, then yes, it's very very helpful to their development.
I dropped out of cs half because I didn’t enjoy the coding because they dropped us into c++ and I found the error messages so confusing.
I discovered python five years later and discovered I loved coding.
( the other half of the reason is we spent two weeks designing an atm machine at a very abstract level and I thought the whole profession would be that boring.)
I think both editors have their flaws, I just know the native vi keybindings better at the moment.
This was about a million years ago. I had just installed a pirated copy of Windows XP (FCKGW-RHQQ2...) and was in the first quarter of my physics degree, taking a class in C. Different times....
... like a dozen? And in 100% cases it's their teacher's fault.
That being said they are very useful. I mostly use them as a far superior alternative to web search and as a kind of junior research assistant. Anything they find must be checked of course.
I think we have invented the sci-fi trope of the AI librarian of the galactic archive. It can’t solve problems but it can rifle through the totality of human knowledge and rapidly find things.
It’s a search engine.
There is a well known phenomenon known as the AI effect: when something works we start calling it something else, not AI. Heuristics and complex reasoning trees were once called AI. Fuzzy logic with control systems was once called AI. Clustering was once called AI. And so on…
This certainly has one root in human or carbon-based life cheuvanism but I think there’s something essential happening too. With each innovation we see its limits and it causes us to go back and realize that what we colloquially call intelligence was more than we thought it was.
Intelligence predicts, but is prediction intelligence?
Again, here by intelligence I mean what complex living organisms and humans do.
I still believe there are things going on here not modeled by any CS system and not well understood. Not magic, just not solved yet. We are reverse engineering billions of years of evolution folks. We won’t figure it all out in a few decades.
So I don’t think there needs to be a semantic debate over where in the process intelligence started. The early responses to stimulus is a form of prediction, but not one that requires thinking.
There can be much disagreement that prediction is at the core of intelligence, or if optimizing ability to predict leads to intelligence. But from the established facts, it is the case the higher forms of life were bootstrapped from the lower ones, and also our biochemistry does have reward functions. Successfully triggering those rewards will generally hinge on making successful predictions. Take from that what you will.
Intelligence is also very good at pattern recognition. Did people once argue for pattern recognition maximalism?
Biological (including human) intelligence is clearly multi-modal and I strongly believe there are aspects that are barely understood if at all.
The history of CS and AI is a history of us learning how to make machines that are unbelievably good at some useful but strictly bounded subset of what intelligence can do: logic, math, pattern recognition, and now prediction.
I think we may still be far from general intelligence and I’m not even sure we can define the problem.
To you & I that's true. But especially for the masses that's not true. It seems like at least once to day that I either talk to someone or hear someone via tv/radio/etc who does not understand this.
An example that amused me recently was a radio talk show host who had a long segment describing how he & a colleague had a long argument with ChatGPT to correct a factual inaccuracy about their radio show. And that they finally convinced ChatGPT that they were correct due to their careful use of evidence & reasoning. And the part they were most happy about was how it had now learned, and going forward ChatGPT would not spread these inaccuracies.
That anecdote is how the public at large sees these tools.
Well, there's the first problem.
> were most happy about was how it had now learned
on tomorrow's episode, those same hosts learn that once their chat session ended, the same conversion gets to start all over from the beginning.
The latter kind of fear mongering hype has been exploited by companies like ClosedAI in a bid for regulatory capture.
Categorically ruling out intelligence because "it's just a token predictor" puts us at the opposite of the spectrum, and that's not necessarily a better place to be.
I'm sorry but your comment is a good example of the logical shell game many people play with AI when applying it to general problem solving. Your LLM AI is both an idiot and an expert somehow? Where is this expertise derived from and why should you trust it? If LLMs were truly as revolutionary as all the grifters would have you believe then why do we not see "forces unseed in human civilization before today" by humans that employ armies of interns? That these supposed ubermensch do not presently exist is firm evidence in support of current AI being a dead end in my opinion.
Humans are infinitely more capable than current AI, the limiting factor is time and money. Not capability!
Maybe you are unfamiliar with the term idiot savant?
Over the past couple of years of educating myself a bit, whilst I am no expert I have been anticipating a dead end. You can throw as much training at these things as you like, but all you'll get is more of the same with diminishing returns. Indeed in some research the quality of responses gets worse as you train it with more data.
I am yet to see anything transformative out of LLMs other than demos which have prompt engineers working night and day to do something impressive with. Those Sora videos took forever to put together, and cost huge amounts of compute. No one is going to make a whole production quality movie with an LLM and disrupt Hollywood.
I agree, an LLM is like an idiot savant, and whilst it's fantastic for everyone to have access to a savant, it doesn't change the world like the internet, or internal combustion engine did.
OpenAI is heading toward some difficult decisions, they either admit their consumer business model is dead and go into competing with Amazon for API business (good luck), become a research lab (give up on being a billion dollar company), or get acquired and move on.
These LLM AIs need to be treated and handled as what they are: idiot savants with vast and unreliable intelligence.
What does any advanced organization do when they hire a new PhD, let them loose in the company or pair them with experienced staff? When paired with experienced staff, they use the new person for their knowledge but do not let them change things on their own until much later, when confidence is established and the new staffer has been exposed to how things work "around here".
The big difference with LLM AIs is they never graduate to an experienced staffer, they are always the idiot savant that is really dang smart but also clueless and needs to be observed. That means the path forward with this current state of LLM AIs is to pair them with people, personalized to their needs, and treat them as very smart idiot savants great for strategy and problem solving discussion, where the human users are driving the situation, using the LLM AIs like a smart assistant that requires validation - just like a real new hire.
There is an interactive state that can be achieved with these LLM AIs, like being in a conversation with experts, where they advise, they augment and amplify individual persons. A group of individuals adept with use of such an idiot savant enhanced environment would be incredibly capable. They'd be a force unseen in human civilization before today.