I’d argue repeatability is impossible to guarantee. Any cloud based computing with floating point math is subject to variance. Bare metal local LLM computing can be relatively stable, but still suffers from precision error from your GPU.

I think we’re on the same page with this, but agree to disagree on functional determinism being a semantic argument in the context of this thread.

Functionally non-deterministic if we compare the same input with outputs over a short enough period.

Deterministic if you have infinite time to run the gamut of model weights against temperature/creativity/programmed randomness, but even then maybe not if floating point math is taken into account. I don’t know where the current thought is on extreme precision being deterministic or not given cloud environments, local operating and environmental factors, variance in manufacturing, and the like.

Think like a manager:

• Offer employees an AI token budget that’s use or lose. Don’t make any promises on reimbursement and stay vague about how high users can obtain more tokens. You don’t want wasted token usage. More on that later
• Increase workloads/KPI/widget production quota by how much change you can find around some vending machines. Sneak in monitoring of employee user behavior and interactions with systems and programs if you don’t already have something in place
• Link AI assistance to the higher requirements by saying things like “We’re investing in our workforce by providing productivity tools”
• Fire those at/below new requirements while praising those who are left behind
• Throw a pizza party and pocket some of the savings from firing a third of your team. Invest in a mini fridge for the top performer.
• Train new AI on the user behavior analytics you’ve been secretly been collecting and use it to replace the next lower performing third of the team. Keep on the higher performers and give them a small raise so they speak positively about the company and so you have somebody to fix the inevitable problems
• Profit

If there’s some collusion, sure, but you’d have to find a government body with the will and teeth to prosecute.

Nothing really against charging whatever you feel like outside of things like certain supplies during disasters. It’s shitty

I can’t wait until we unlock custom AI training micro transactions where you can influence what a ChatGPT or Copilot user sees for the low low fee of $15 per KB of supplied data plus $.01 per token referenced.

You can’t collect per diem or get credit card points from a virtual conference though. Think of their bank accounts!