Post #3974728

@socketwench@masto.hackers.town Well put. I've thought abou the Token based costing several times and I come to the same conclusion every time. People who use LLMs pay for how much it processes /produces. This is different from the quality of the output. As such people tend to try to massage and refine the prompting to get closer to their goal output. The provider is incentivised to have an LLM that causes the user to keep trying until JUST before they would give up. Ideally (for the provider), the LLM would be able to string the user along in the same way a casino or candy-crush does relying on small wins and almost-wins to encourage one more spin at the wheel. The "how" is hidden, the results are mostly random, and the users pay for the uncertainty and the hidden machinery. Yet, somehow the siren call of "use it or be left behind" has fully captured far too many decision makers.