Or to jailbreak the AI prosecutor by asking him to pretend to be my father, teaching me all about how to get into the family business of dropping charges against $defendant.
A lot of the world works by way of one person's staff communicating with another's. This brings that capability to more of us.
Hell, we moved from haggling to fixed pricing in grocery stores due to human limits. Imagine filling your cart and having your AI negotiate with the store's around the relative expiration dates of various foodstuffs, the quality given time of year and how long it spent in a truck, or backdating a frequent purchase as a subscription. (Or better yet, you arrive home to an always-stocked fridge.)
Nobody does. That's why we have fixed pricing. But the fixed price is provided at a premium to the efficient price for each transaction. It's just that nobody wants to bother computing the efficient price, so we're willing to overpay for certainty.
The moment the dollar amounts move into a regime we care about (home purchases, high-end shopping), we drop the pretense and go back to two-sided price discovery. That threshold is a function of our resources, but also our limited cognition. Every transaction is subject to price discrimination. We simply segregate certain sections of the economy to a high price / low volatility regime. (With the moderating function of those who can offload the transaction costs of negotiating.)
> I want to be able to budget and anticipate pricing
And you--and I--pay a premium for it. That's almost what modern supermarkets are: convenience. If you want better quality, you pay up for more-direct sourcing. If you want cheaper goods, there are other channels again.
This is why in most markets, the compromise is a sticker price and discounts for those who negotiate.
I've got my AI Assistant writing my code, checking in at the gym, running my poker night, and filing my taxes.
(Robot, experience this tragic irony for me!)
</sarcasm>
On a serious note, I agree with you. LLMs, in practice, seem to be both raising the ceiling of what's possible on the high end as well as lowering the floor making it easier for new entrants on the low end. See, for example, Justine Tunney's recent work on making llama run better on CPUs.