Post #3957144

2026-07-20 09:54 UTC

My prediction, we will massively overbuild AI infrastructure only to realize that most people in the end would like to run things locally on their own (or maybe limited rented) hardware. https://huggingface.co/blog/security-incident-july-2026

Replies (4)

@TallSimon@mstdn.ca 2026-07-20 11:32

@koen_hufkens@mastodon.social (I had to estimate because of bad resource use reporting in the original paper, but...) The last AI application I reimplemented with my own math ran on about 2.3 million times less resources and required ~3% of the data of the AI (not even LLM) original. Same data, same application, and I got better results. Then I realized I had a small oversight and could make my version at least 3 times faster. As Ed Zitron says, let AI burn.

Open ##3958403
@ErikJonker@mastodon.social 2026-07-20 09:57

@koen_hufkens@mastodon.social ...maybe that prediction is already true for many large companies, running (chinese) open weight models on their own infrastructure, still those local models also need a lot of compute so it will be more redirection of available resource. Ofcourse companies like Anthropic and OpenAI could suffer

Open ##4225347
@nick_appleyard@mastodonapp.uk 2026-07-20 10:15

@koen_hufkens@mastodon.social Or not at all.

Open ##4225350
@coreysnipes@hachyderm.io 2026-07-20 10:53

@koen_hufkens@mastodon.social Agreed; this is the mostly likely post-correction reality. Ed Zitron says he's watching for data center / scaler capital expenditure pullbacks as the first signs of correction.

Open ##4225355