@eosfpodcast It's why I don't use auto correction in messages and it's not even Ai. It made me lazy and not care about spelling because I was expecting system to fix it itself (I'm not English native, though how some natives use there and their wrong, maybe they should disable it too).

Ai chatbots are not much different, you ask them and they spit out some BS. People don't search and investigate things anymore to learn anything along the way, they just expect answer, correct or not.

@eosfpodcast I see we're now in the "of *course* you can't expect this skill from an LLM, that's not what they're good at" stage of the discourse. This sort of remark is also popping up a lot in comments to people posting links to that article about how an old Atari console chess game from almost 50 years ago stomped ChatGPT recently.

This observation is of course totally correct. LLMs are genuinely not good for this sort of thing. They can't play chess very well (certainly not as well as just about any off-the-shelf dedicated chess program from anytime in the last several decades), and they can't actually reason about anything, instead just printing text that *looks* like what a reasoning person might do or think. And, of course, people critical of "generative AI" have been pointing this out for years.

Pointing this out is neither trite nor useless.

When OpenAI literally advertises "Learn something new. Dive into a hobby. Answer complex questions." as a sensible use case for ChatGPT (https://openai.com/chatgpt/overview/), people will expect behavior like this from an LLM.

When Google uses its LLM, Gemini, not just as a search adjunct (where it might be at least *somewhat* useful, since it could plausibly associate your request with actual web pages' text) but as a tool for creating factual summaries of information on the web, people will expect behavior like this from an LLM.

When the entire first *year* of ChatGPT hype claimed things like showing "sparks of artificial general intelligence" (https://www.microsoft.com/en-us/research/publication/sparks-of-artificial-general-intelligence-early-experiments-with-gpt-4/), people will expect behavior like this from an LLM.

When every vendor out there (Anthropic, OpenAI, Microsoft, you name 'em) aggressively markets their models as "reasoning" models, people will expect behavior like this from an LLM.

There are lots of people who have no clue about the limitations of this flavor of "AI", precisely because it has been -- and still is -- hyped so aggressively in counterproductive and misleading ways.

@eosfpodcast

The goal is not to provide answers and information

The goal is to DESTROY information.

@eosfpodcast This is why the EU shouldn't try to keep up with US companies and AI, but should focus on sovereignty,security, and privacy. Taking back control over our own data..

@kimvsparrentak @kimvsparrentak

@eosfpodcast

Yup, that Google Collab that I just wrote using a simple prompt and a math equation to help a coworker understand a very difficult concept that would have probably taken me a whole day to write traditionally is an illusion.

@eosfpodcast

This is clearly the beginning of a new era... Good. I just wait to see all the negative comments that Apple is cooked, lost it and will fade while the AI hypers are the one to be in the dustbin of history. I need a new phone and I might consider Apple now.

@eosfpodcast

"New research reveals that today’s “reasoning” models aren’t thinking at all. They’re just sophisticated pattern-matchers"

No. No, new research does not show that. Everyone who had the least bit of understanding already knew that this was what was going on. But I'm glad to hear that Apple has "realized it"

If anything I would say this is how they frame it, as to not seem like idiots to have bet so much in this horse for so long.