“If you really want an "AI strategy" for your country, it should be this: wait for the bubble to burst, then buy hardware and talent at fire-sale prices in the wave of ensuing bankruptcies, and use them to extract more performance from free, open source models.” -- @pluralistic pluralistic.net/2026/07/22/tab… ❤️👍
Pluralistic: Trump’s America can’t even win a rigged game (22 Jul 2026) – Pluralistic: Daily links from Cory Doctorow
Pluralistic: Trump's America can't even win a rigged game (22 Jul 2026)Cory Doctorow (Pluralistic: Daily links from Cory Doctorow)
Jonathan Lamothe likes this.
reshared this
Resuna
in reply to zerok • • •Cory Doctorow
in reply to Resuna • • •A model that I run on my laptop that doesn't even trigger my fan when it executes is definitionally not "a waste of energy." I consume more CPU when I do a gaussian blur in the Gimp.
Jacob Ibáñez 🔒
in reply to Cory Doctorow • • •Cory Doctorow
in reply to Jacob Ibáñez 🔒 • • •Cory Doctorow
in reply to Cory Doctorow • • •Cory Doctorow
in reply to Cory Doctorow • • •Cory Doctorow
in reply to Cory Doctorow • • •Ovidi Nawer
in reply to Cory Doctorow • • •Using neural networks is, with 1 unit of exception in cancer localization which still can't be truly verified and is rather shaky, an invitation to go back to the point we're now: religious glorification that allows further bullshitry to happen, without any reason besides a "maybe it will be better"
Sadly can't be a fun experiment anymore
Cory Doctorow
in reply to Ovidi Nawer • • •Running Whisper or an Ollama-installed chatbot on my own computer doesn't require anyone to improve anything - no more than my piping the output of a shell command through grep puts pressure on the grep maintainers to release a new version.
Ovidi Nawer
in reply to Cory Doctorow • • •Sophie of Ratsnake Games
in reply to Cory Doctorow • • •Sophie of Ratsnake Games
in reply to Cory Doctorow • • •this is "flying doesn't have CO2 emissions because the plane would fly anyway if i wasn't on it" logic and i'm disappointed to see an acclaimed public thinker like you not understand the connection between supply and demand.
Models are getting trained because there is a demand for them from users.
Resuna
in reply to Cory Doctorow • • •Cory Doctorow
in reply to Resuna • • •@resuna Spotting paragraphs in a document where I forgot to put a period at the end doesn't make me less creative. It doesn't damage my reasoning ability.
This is a very weird thing to believe.
Can you explain to me how you think using a spellchecker makes me a worse writer?
Is that only true if the spellchecker uses this form of statistical inference? Or have all the spellcheckers I've used to write 30+ books also make me worse at writing?
Resuna
in reply to Cory Doctorow • • •Cory Doctorow
in reply to Resuna • • •@resuna
1. How is running a FLOSS model on my laptop "expensive" or "convoluted?" I just invoke the model from a terminal, type "find typos" and hit ctrl-V-CR
2. The model finds other typos besides dangling sentence. Here's a before, after and diff:
craphound.com/before.txt
craphound.com/after.txt
craphound.com/diff.txt
Resuna
in reply to Cory Doctorow • • •Jonathan Lamothe
in reply to Cory Doctorow • •@Cory Doctorow @Resuna @zerok It's more computationally expensive than writing a simple script to do the same thing. You just don't notice it because modern computers are so stupidly powerful.
There is a big caveat to this statement though: the latter requires the user to know how to write such a script, or to have one convieniently handy. Plus, you have to weigh the effort of writing such a script against the amount of work it's going to save you.
Personally, I wouodn't trust an LLM not to hallucinate and subtly corrupt my work in a non-obvious way, but that's a judgement call and you might choose differently.
Nicole Parsons likes this.
Cory Doctorow
in reply to Jonathan Lamothe • • •@me @resuna Getting a list of proposed correx can't "subtly corrupt" a work - only the writer accepting those correx without exercising judgment or discernment can do that.
I've been overriding bad spellchecker suggestions for almost 40 years. I've been overriding misguided copyedits for more than 30 books.
I'm perfectly capable of deciding which correx I want applied to my work and which ones to reject, irrespective of how those correx are generated.
Nicole Parsons reshared this.
Jonathan Lamothe
in reply to Cory Doctorow • •Nicole Parsons likes this.
JBrown, Esq
in reply to zerok • • •Ciourte Piaille
in reply to zerok • • •