Subject: "AI" and price models. Screenshot: Companies waiting to be <strike>milked</strike> serviced by "AI" firms.
@berkes : You said, "When the daily bill for tokens is higher than the common monthly income in your country, it's inaccessible."
@cstross : You responded, "When that happens, companies will start hiring human developers again, including newbies, and be forced to pay wages competitive with agentic development and priced against developer scarcity. Writing code will be returned to the domain of human skills."
The token-based "AI" price model is already a failure. Here's what it looks like from the perspective of a code and data architect [who's been in the sectors for 48 years].
My feeling is that LLM and GenAI are highly useful tools in some contexts but the current mania is deranged. To the tune of 17 times more deranged than the dot-com mania was. That is quite a large serving of fruitcake.
I use LLMs every day now as a starting point for analyses of different types. I asked one of them to check this post. But I'd never agree to let an LLM write code or text for me unless I was allowed to do a complete rewrite.
In the case of code, that would be impossible except for small pieces. That is all that I feel is safe in the context of software.
I agree, too, that the wholesale swallowing of the Web has been illegal [and not Fair Use]. In theory, "AI" firms should be compelled to destroy what they have and start over. In practice, as billionaires are in charge, that isn't going to happen, but laws that are actually enforced need to be put into place.
Regarding #AI price models:
Services that replace semi-skilled human workers entirely with "AI" services [fast food order takers being one example] typically use non-token SaaS price models. It works out to paying a simulated person less than minimum wage.
One of my past employers is a national chain [in the U.S.] that is experimenting with this. My former VP of IT tells me that customer satisfaction is at only 80% but they hope to get it to 90%. My guess is that most corporations will say "80% is fine" and expand the use of "AI" at this level.
Full-scale replacement isn't going to work in sectors where highly skilled humans are involved. I'm referring to contexts such as software developers and paralegals.
Use of "AI" at that level is going to collapse regardless of pricing. LLMs can't replace people in those contexts. It's already proving to be a disaster for some legal firms. Improvements in models aren't going to fix this.
We know that there are problems inherent in the entire concept of LLM as it stands now that are mathematically impossible to fix. "Scale" as much as you wish. Drain the supply of electricity for a dozen cities. It'll make no difference.
I asked an #LLM to comment on this part. The LLM responded as follows: [edited for length]
"Legal firms attempting to replace paralegals [with AIs] have encountered critical failures ranging from hallucinated case law to compliance breaches. No amount of parameter scaling or energy consumption can bridge the gap between probabilistic text synthesis and legal or computational correctness. [AI] tools are effective as assistants for drafting preliminary text or analyzing data, but treating them as autonomous domain experts invites structural collapse in professional services."
Companies that bulldoze through the idea that "AI" can be used to replace skilled employees are going to fail. This leaves the grey area of retaining [some] skilled employees and telling them to use "AI" to "work smarter" and that profits had better skyrocket because "AI" is magic.
That is where "AI" token usage figures into things the most.
Until now, companies have forced employees to make heavy use of "AI" tokens. One developer posted that he'd responded by asking "AI" to play games for him and thereby burn tokens. He was praised at his next review.
However, companies are likely to decrease token usage from here. "AI" output isn't going to be good enough to justify the costs. That is before the large price increases that "AI" firms are going to need to impose.
"AI" firms believe that regular firms are locked into "AI" usage now like cows locked into stalls and waiting to be milked. The fact is, not really.
