Its the forever problem of AI - its like a one time reaping of high quality data and then information goes stale because few people will provide information for free, fewer people will be motivated to build any content online, most things will get protected and locked down, open source is saturated with agents.
This is 100% the "glory" days of AI. The price is highly subsidized by investors, govt but thats going to ramp up and the product quality will dilute as they need to drive revenue for the markets. Its the same long lived strategy of every great tech company that has VC returns to feed. If OpenAI can stay private for a longer period of time they could ostensibly continue to build a better product and not be handcuffed to wallstreet in the near term.
Cost of AI will get subsidized for 3-4 years while corporate america gets addicted, then pricing will ramp up and corporate ameria will be dependent. This has been said millions of times over.
I think using AI in 10 years will feel like using StackOverflow today. SO is kind of stuck in time, where every answer there is how you’d do things in 2015. They structurally can’t seem to let go of old answers and ingest new information, so they are dying in the past.
The same might happen to AI if it stops having access to fresh information. If people move to walled gardens that can’t be consumed by AI then there is a chance that these products eventually get stuck in the past, too. I don’t know what would be a walled enough garden for this, but it might happen.
1.) The web and software code get stuck in the past, as everything gets built with outdated styles, a leftover from the open source era.
2.) A few companies that can, like Vercel, will build new UI or new APIs and backend structures and gate access to them with their own proprietary models. They will make their docs internal or not readable by AI bots, like what Tailwind did.
The same was said about Uber and Lyft in other words that once Venture Capital subsidies ended the prices would be passed on to consumers. While it's true on inflation adjusted basis is not that extreme from its low point maybe 30% higher.
And with Rideshare you have a fundamental cost of a car and a person whereas with AI costs can quickly be reduced across multiple dimensions of optimizations Hardware stack architecture region energy costs latency and other trade-offs. Then you have open models competing to bring costs down. So I just don't know that we'll ever see the cost go up
Rideshare is interesting because uber et al might actually be cheaper per ride than waymo et al. Yes, they pay a driver a very low wage which is some overhead, but they don't own any cars. The driver pays for the car, its maintenance, registration and insurance, fueling, and storage. It is dubious if it even pencils out for anyone but the power users in the gig economy. Waymo has massive overhead with its fleet requirements, on the other hand, real estate requirements as well.
AI could be similar. The costs of the bleeding edge are probably enormous. But imagine if your competitor was some cheap model, with a layer of southeast asians between the user and the model getting the apparent quality of the model far higher leveraging some cheap expert human prompters. There's some odds there that they might route you at the bleeding edge. Plus also, plenty of users probably aren't even that sensitive to a low quality vs high quality response for their purposes.
reply