back

by mark_l_watson·5mo ago·view on hn ↗
I upvoted you, and mostly agree with you, but: as much as I love using strong small local models sometimes a commercial 'frontier' model is very useful (for me).

I deleted my free OpenAI account (I paid for it until a year ago) and just started a $20/month Anthropic account. My one-year prepaid Gemini account will expire in two months and I will decide then to keep one of Anthropic or Gemini.

Once again: I agree that super-spending on super scaler data centers is a net negative for humanity. For me it is not a matter of price: I am happy paying $20/month and only using energy guzzling models occasionally when I really need them. Sort of like recycling to lesson our burden on the environment: try to minimize AI energy and resource use, but still get work done.

EDIT: we can also use LLMs more efficiently: build software composed of small well tested libraries. It is more energy efficient to write and debug little 200 line libraries than soaking up large projects in your context. Also, working on small composable libraries works better with smaller open models like qwen3.5:35b.

1 comments
> we can also use LLMs more efficiently: build software composed of small well tested libraries. It is more energy efficient to write and debug little 200 line libraries than soaking up large projects in your context.

So, NPM? In reality AI is making this LESS likely to happen. It's easier to write a small utility function with AI then find and use a library these days

This might be preferable to the ever expanding tangled dependency graph and associated supply chain risk. OTOH, perhaps said graph can be reasonably wrangled with LLMs and enable safe small library re-use?