logoalt Hacker News

postalcodertoday at 1:46 AM4 repliesview on HN

I wonder if it's related that that OpenAI has found a way to cut inference costs by half, according to The Information.

https://www.theinformation.com/newsletters/ai-agenda/openai-...


Replies

layla5alivetoday at 2:39 AM

https://archive.ph/NEwVz

"However, these inference optimizations, which rival Anthropic refers to as “compute multipliers,” are a big focus for all the labs. Anthropic CEO Dario Amodei has been publicly talking about the concept since at least mid-2023, when he said on a podcast that the company limits “the number of people who are aware of a given compute multiplier” because it could give other AI labs a leg up if they were to be able to replicate them. (Compute multipliers can also refer to efficiency optimizations in the model-training phase.)"

Yes, on a world with finite resources where your industry is singlehandedly siphoning ALL THE RESOURCES - hoard general efficiency optimizations and treat them as trade secrets - winning is all that matters, normal people and other species and the planet be damned.

Everything I hear about Dario these days makes me like him less and less. He sure did seem to speed run the 'tech leader with scruples' to 'tech villain' path! I guess all the cycles are compressing as we approach the singularity..

show 11 replies
torginustoday at 10:52 AM

I wonder if AI labs are actively manipulating the narrative (and thus investor sentiment) by airing problems, and then solving them weeks to months later. I wouldn't be surprised if they have a lot of stuff figured out that is not included in the current version, just to make a steady product cycle with years of tangible improvements from one version to another (this is a common practice in the industry).

For example, if inference isn't too expensive, but they figure out how to cut costs, then price goes down. After all, why pay OpenAI when a smaller datacenter can give you similar models?

But, if they make a huge issue about how inference is too expensive, they engineer a crisis of their own creation - then, once they deploy the solution (which they might already have), then they're back on top.

minimaxirtoday at 2:51 AM

Semi-related, has anyone noticed their GPT 5.5 usage in Codex being cut in half as of a couple days ago? I got a lot more mileage out of my session usage yesterday for the same workload.

show 2 replies
drivebyhootingtoday at 1:50 AM

What’s the technique? And did they buy it from thinking machines?

show 1 reply