What do you reckon about all these gpt-4-level AIs just popping up left, right, and center like gpt-4o, sonnet, llama 3.1, command-r plus, etc., virtually for free? Do you think this is going to pop the whole AI bubble/hype?
Regardless of the benefits of AI, commercially it is a race to the bottom with prices coming close to free and with great open source and free models also contributing to this race to the bottom.
The current winners in this race are Nvidia and especially Meta as they have a large enough cash flow and data moat from it's 1BN+ users to continue investing in releasing powerful open source models and their business doesn't rely solely on AI like OpenAI does.
Other AI labs are relying on more raises with Adept getting sold, Stability AI going nearly bankrupt making it's product as free as possible which isn't sustainable for a startup to do.
IMO, gpt-4 level is the base building block. It's then a race to get this basic thing out for cheaper and more reliable. GPT-4 is someone with IQ 120 or so and people are happy with that. It makes as many mistakes as an above average person. It's a general purpose AI and like a CPU, you can use it for general purpose stuff.
I doubt smarter AI would be better. People will go back to assuming it's omniscient then get disillusioned when it's not.
The race will go back to training. Not models, but training the AI like it's a person. How would you train a junior copywriter who graduated with straight A's? The junior Kotlin developer who memorizes the manual inside out then picks the wrong architecture?
I think what we'll see is services being scaled. Back then, if you wanted to scale, you'd make a product. But then you got Jenny from legal, top 1% in employment law, runs her own blog, charges millions per year but only a few clients at a time. She can train Jenny AI, somewhere top 20% in the country who'd charge less. You'll see Marie AI giving suggestions on how to tidy up your house without needing an actual Marie. Marie wrote a best selling book and has a best selling TV series, but who reads books? The AI version will also be able to give more attention that these individuals can't. On the darker side, we'll also see voice actors being replaced by deepfakes.
Idk if it'll burst the bubble though. It will probably shift it to another group.
I think you have to distinguish between supply-push services and demand-pull services. People will pay big money for demand-pull (I have a problem and you will solve it) and a little money to supply-push (Here are some words that I sell you and that might help you). AI can probably do supply-push well, but will find it a real struggle to address demand-pull. Your customer may only have a 100 IQ, but regarding their specific problems they have a 200 IQ, and your 120 IQ AI attempting to solve their specific problems will end up frustrating, angering and eventually just disgusting the customer.
That's the point, yeah? AI will convert a lot of things from demand-pull to supply-push. That's where the hype is. It's going to lower pay for a ton of people. That's where the hate is. That's why there was a writer's strike last year. That's why artists have been boycotting everything made after 2020.
People aren't disgusted that the AI solves a problem, they're disgusted because they linked their identities to the way a problem is solved. Taxi drivers are still disgusted by Waze, Uber drivers less so.
“Linked their identities to X” is the key behavior to understand here. A lot of creative people have linked their identities to activities that are low on the food chain for AI. The close-minded creatives will resist AI because it threatens the identities they built for themselves, the open-minded ones will embrace AI as a tool to increase their creativity.
I think the margin between OpenAI and the next few best competitors is already fairly slim, but as OP mentioned, the margin between the median competitor and the best is also decreasing.
It suggests that playing catch-up progresses quicker than pushing the state of the art.
I don't see how any of that pops a bubble but I think it could transformative for the leading companies (OpenAI, Anthropic, etc) which may become more rank and file service providers.
AI is not a bubble and already bring real benefit and will bring even more in future. It will spread more and at some point it will became so common that we will just use it and just talk less about it. For example, same happened with mobile phones, at first everybody was talking about them but now everybody just use them and does not talk much about it.
Apps were a bubble. Cloud was a bubble. WWW was a bubble. Houses were a bubble. Bubble tea was literally bubble.
Things can be beneficial and a bubble at the same time. Useful things in short supply are expensive. It's a bubble when someone buys a thing for expensive in a gamble to sell it off to someone else for more expensive. It's not a bubble when a house seller sells a house to someone who wants to live in one. It's a bubble when they're selling to a seller, who sells to another seller, and so on.
Also I think one oddity with AI is that it came in very cheap. So it's hard to make it a bubble. But I guess grifters are stockpiling the shovels without realizing that everyone has one... hence OP's question.
gpt-4 level AI is almost 2 years old by now. When GPT-next comes out at one OOM above, the entire field will shift again as it did with GPT-4, and nobody will care that GPT-4-par models are almost free.
There’s nothing to suggest LLM’s haven’t already fallen victim to diminishing returns.
Just making even bigger models isn’t a near linear jump in quality.
GPT-5 might be massive, expensive, and only a bit better and fall to the same problems as precious models, without a radically different underlying architecture (which is not clear if it exists)
the thing that’ll burst the bubble is when people need to pay the actual retail and not the disruption based pricing.