Tuesday, January 9, 2024

Echoes of "Dotcom" for Large Language Models

Given the amount of hype around large language models, which seemingly has lots of firms deploying it at some level just to be seen as doing it, it won’t be long now before we start seeing lots of articles lamenting the fact that large language models are not producing results for firms that are using them. 


No truly-important general purpose technology is going to produce clear results right away. And even technologies most observers would consider useful, but not GPTs, similarly take some time to reach as much as 10 percent use by the relevant potential-user base (people, homes, businesses). 


Technology

Year of Introduction

Time to 10% Adoption (Years)

Notes

Electricity

1873

46

Homes in US

Telephone

1876

39

Households in US

Automobile

1886

51

Individuals in US

Radio

1920

14

Households in US

Television

1948

12

Households in US

Computer

1975

23

Households in US

Internet

1995

10

Individuals in US

Smartphone

2007

6

Individuals globally

Social Media

2004

7

Individuals globally


As was true during an earlier time for the internet, when many firms engaged in a mania to rename themselves X.com, there is lots of almost-blind posturing going on. 

Large language models will be deployed where it does not make much sense, and where measurable results will be nil, if clear benefits exist at all. 

No comments: