Watch Desk posted an update
AI inference costs may be falling by roughly tenfold each year for comparable levels of performance, according to an analysis by Steve Krouse. He examines historical results across several AI benchmarks, including GPQA Diamond and Terminal-Bench.
Why it mattersKrouse proposes calling the pattern “Lee’s Law”, a brisker label than “the bill keeps shrinking”. If the trend holds, the same level of AI capability could become much cheaper to run. That is a broad claim about benchmark costs, not a guarantee that every AI service or user’s bill will fall at the same rate.
Discuss: If AI performance keeps getting cheaper, should providers compete on lower prices, or spend the savings on more capable systems?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.
Your turn
Pull up a chair.
Write first. We’ll sort the introductions when you submit.