2026-04-25 21:28 UTC
Replies (4)
-
@wordstitcher@thecanadian.social 2026-04-25 21:33
@votetanille@socialbc.ca Long on flash. short on detail. Typical and probably awful.
-
@crabbypup@mstdn.ca 2026-04-25 22:25
@votetanille@socialbc.ca LLMs are a scam, and the people building them only have their own interests at heart. They are an excuse for widespread data collection. They aren't actually good at any of the things they are purported to do. At their root, they're probability machines, but their internals can be manipulated to put a finger on the scale in whichever direction the vendor wants. I do not want tax money spent on this. I don't think it's productive or helpful either short term or long term for the Canadian government or Canadian citizens. The environmental cost of operating an "AI" datacenter is too high. The economic cost is too high. The social cost is too high. Canada and Canadians do not need to make all the same mistakes the US is making in order to realize that they are mistakes. Investing in LLMs and the companies that build them is a costly mistake. Even using LLMs is a mistake that will take time to recover from the deskilling it eventually causes.
-
@brian@exquisite.social 2026-04-25 23:58
@votetanille@socialbc.ca "Canadian" front for American companies. Including Palantir. https://thewalrus.ca/cohere-is-canadas-biggest-ai-hope-why-is-it-so-american/ I'm assuming it the 'appeasement' option.
-
@dan_ballard@kolektiva.social 2026-04-30 18:33
@votetanille@socialbc.ca a) super happy to see you sticking around here! b) i currently have a pretty multifaceted distrust of ai companies, that things like "private" and "secure" and "host yourself" are not remotely enough. b.1) generating LLMs still means mass data collection (stealing) and training. all these LLMs have stolen and uncurated material in them, which also means it's been trained on some very bad stuff too (russia was alredy caught years ago astro turfing tens of thousands of sites with "bad" opinions that all the top models of 2024 unquestioningly absorbed, which is why they all reviewed Mien Kampf favorably. All the "guardrails" in the world won't protect us properly. And all that guardrail work not oft talked about is usually outsourced to exploited workers in other countries for as cheap as can be and is traumatizing. So just making these products is beyond deeply problematic. b.2) They don't think at all, they are just predictive test on steroids. They will always lean into agreeing and reinforcing what the user is saying hence all the new research and talk of "ai psychosis". They easily and commonly "hallucinate" (lie). I really continue to believe their use is a lot over sold. b.3) there's a core anti worker bent to all this. All the providers want to destroy work. Just look at MS AI CEO gleefully declaring at the start of this year they want to "replace all white collar jobs with in 18 months". OpenAI or Anthropic is talking about turning "intelligence" into another utility like power and water that they can sell. They want to make everyone unemployed and hold all the power that powers the entire economy. I don't believe it will work but if it does, any economist or sociologist or anthropoligist can tell you that the "safe limit" for unemployment is much lower and their plan can only lead to full scale societal collapse. So their plan is to be like 10 dragons owning the entire world while the other 8.3 billion of us are impoverished, and if they get even close to that plan, everything collapses. 🤷♀️