HomeVideos

Chamath: Tokenmaxxing is a bridge to nowhere

Now Playing

Chamath: Tokenmaxxing is a bridge to nowhere

Transcript

44 segments

0:00

Over the last year, I looked at the

0:02

stats this morning, token spend among

0:04

ramp customers has grown by 21 times.

0:07

>> 21 times?

0:07

>> 21 times.

0:09

>> Not 21% we're talking about 21 times.

0:11

>> That's exactly right. So, many CFOs,

0:13

they're often very surprised by the

0:15

bill.

0:15

>> He's saying something so important there

0:17

because

0:18

if your engineers are going off randomly

0:21

in an unguided system and then just

0:24

ripping through million tokens at 56

0:27

bucks,

0:28

what he's talking about is the eventual

0:30

downstream impact to earnings. And that

0:32

eventually a bunch of these public

0:34

market CFOs are going to show up to Wall

0:35

Street and they will have missed

0:37

earnings because their OpEx at some

0:39

point if things are 21Xing every few

0:42

months, somebody's going to miss a

0:44

quarter. Eric would not have released

0:45

this ramp product unless CFOs were like,

0:48

"I can't control the spend."

0:50

>> Yes.

0:50

>> And then he's like, "Well, here, let me

0:52

build it for you." And then if enough

0:54

CFOs essentially turn that feature on

0:56

and start to rate limit how it's spent

0:59

because maybe they're not getting the

1:01

ROI and the engineer doesn't care about

1:03

ROI. The engineer is like, "I want to

1:04

use the latest greatest model."

1:06

>> Yeah. Make me better.

1:07

>> you don't need it. Maybe Mirror's right

1:09

and for 95% of the tasks you should be

1:11

at one level lower especially when it

1:13

costs 1/100 of the cost. Unless you get

1:15

a control of this and you can directly

1:18

say how much money you're making, this

1:19

is a bridge to nowhere. It is a money

1:22

burning furnace.

Interactive Summary

The video discusses the explosive 21x growth in AI token spending among Ramp customers and the resulting concerns for CFOs. The speakers highlight the risk of uncontrolled engineering expenditures impacting company earnings and the necessity for financial oversight and ROI-focused strategies in AI model selection.

Suggested questions

3 ready-made prompts