正在加载视频...
视频加载失败
Perplexity’s CEO Aravind Srinivas: how a single power user can now consume as much compute as an entire small team. "There are real engineers at Meta and other companies spending around $10 million a year per engineer on these coding tools. There are users in Perplexity Computer, who spends... show more
12,764 次观看 • 5 天前 •via X (Twitter)
19 条评论

Full video

That is how it should be of course, but I'm skeptical that other than burn tokens they're achieving any kind of efficiency or reliability.

@rohanpaul_ai The $400 billion question is how much of that is actually being used to create AI models, and not just build hype around them 🤑💸

I too could spend $10M a year on tokens. The hard part is doing work with that that someone else will buy from me for over $10M.

One power user consuming a whole team’s compute is a wild shift.

This is a fascinating shift in how we think about software usage and productivity.

my api spend keeps going up every month and i thought that was bad until i saw the 10k/month numbers. not even close apparently

岗位职责:独立烧光全组预算。

It seems the real shift is that compute is the new headcount. Have you guys tried running agent flows solo? One person can burn a team's budget and still ship faster.

oh, they should shift to typesafe jev then

Cheaper tokens were supposed to shrink the bill. They raised it. Ten thousand a month per user is someone running agent loops around the clock, which nobody could afford when a call cost real money. The unit price fell and the total spend went up.

$10k/month as a badge of being a power user is a very specific kind of flex. it says nothing about output, only about willingness to spend. same instinct as bragging about gym hours instead of the lift number.

Per-seat SaaS pricing is dead, usage-based agent orchestration takes over.

Maybe he should be focusing on finding a moat for his company than going on podcasts. All I have ever seen is him going to podcasts or sponsoring them, or do Elon style bitching on Twitter

$10M a year per engineer only makes sense if the output scales linearly with spend, and it doesn't. At some point you're paying for compute to compensate for a workflow that's still broken.

一年一千万美金砸在编码工具上,比工资贵几十倍了吧😂

$10M/year on tools for one engineer is also the number that makes replacing that engineer financially obvious.

Vi un tipo en un café en Chacao con tres laptops corriendo prompts en bucle mientras tomaba café. Solo. ☕️🔄

The interesting shift is that one person can now drive the workload of an entire team.
