Joe Brockmeier1d ago

RE: https://mastodon.social/@glynmoody/116324703722433365

This is a preview of the future. A number of people have already pointed out the obvious parallel between getting people hooked on drugs and getting hooked on using LLMs... it seems we're seeing it play out faster than I'd have expected.

Show threadLarvitz  1d ago
@jzb AI inference is expensive, both financially and energetically. A lot of platforms are masking that by subsidizing usage to chase growth. But that only works for so long: eventually the VC money slows, investors want revenue, prices rise, and limits get stricter.
Show threadJoe Brockmeier1d ago
@Larvitz Yup. So growing a dependence on those tools is, IMNSHO, unwise even if you discount all of the other (many) ethical, social, cognitive, environmental, legal, and technical objections.
Show threadLarvitz  
@jzb I use a small and open model from France (Ministral) on my own laptop. For surprisingly many repetitive tasks (pattern recognition in files, text redaction), that’s quite good. Without depending on a cloud, without a subscription and without exfiltrating my data.
Mar 31 at 5:53pmWeb
Show threadJoe Brockmeier1d ago
@Larvitz yeah, that’s a different matter; I’m really just talking about being dependent on a service.