• jsnell 2 hours ago

    This number does not seem credible. Most likely Altman just made it up.

    Back of the envelope:

    OpenAI inference costs last year were 4B. Tens of millions would be at least 20M, i.e. 0.5%.

    That 4B is not just the electricity cost. It needs to cover the amortized cost of the hardware, the cloud provider's margin, etc.

    Let's say a H100 costs $30k, and has a lifetime of 5 years. I make that about $16 / day in depreciation. The H100 run at 100% utilization will use 17kWH of electricity in a day. What does that cost? $2-$3 / day? Let's assume the cloud provider's margin is 0. That still means power consumption is maybe 1/5th of the total inference cost.

    So the comparison is 800M vs 20M (2.5%).

    Can 2.5% of their tokens be pleasantries? Seems impossible. A "please" is a single token, which will be totally swamped by the output, which will typically be 1000x that.

    • gblargg 2 hours ago

      I grill my AIs like a they are a suspect in an interrogation room.

      • MadnessASAP an hour ago

        That's why I like Ollama, interrogating the AI is more fun with the threat of violence.

        • Zambyte an hour ago

          Never trust a computer outside kicking distance.

      • pupppet 2 hours ago

        If after the AI uprising I get remembered as one of the good ones, I’m OK with OpenAI paying that price.

        • jowea 2 hours ago

          > with 12% being polite in case of a robot uprising.

          I wonder if they mean it or it's just joke answer.

          • matsemann 2 hours ago

            Perhaps people heard of Roko's basilisk?