I would argue that there's a line between telling a model to use a tool vs it autonomously choosing to, but bikeshedding over nomenclature is a minefield.

I would bet a good sum of money that the people mad at that dataset would spontaneously combust if they knew how Twitter gave away their data to any researcher that asked pre-Elon.

Rallying against it in the name of anti-AI when it has a no bots clause in the ToU... feels par for the course.

No way that happened 8 years ago. I still remember it so clearly..

If people flatten my ethics cause I use Qwen Coder locally once in a while, they're entitled to their opinion. There's no real room for nuance on most platforms anymore, cause the simplest interpretations will collect the most likes and bubble to the top. That's probably why the topic feels charged.

I like the 'idea' of LLMs, and I have Ollama running on my machine to help autocomplete code / bounce idea off of, but I hate the industry around it, so I've learned to ignore all the product launches and keep an eye out for models that people are doing actual interesting things with.

All the products driving breathless valuations also need to be judged independently, but most of them are terrible and marketed on falsehoods. There is a tech industry pattern of 'market new tech as the solution to every problem to drive valuation', and after so many of them people get cynical.

The models themselves are the first step in my mind where we can start making value judgements. If someone takes a bunch of MIT code and trains a LLM, I'd defend that as an ethical use. A commercial model using books without a license is worst, but that's a judgement to be made on each model.

I've never heard anyone that understands the tech argue that the underlying architecture is somehow unethical. The people working on HuggingFace curating open datasets to run experiments don't ever seem to be called out for their work, and I'd argue that they're ethically above board.

The discourse around the current AI cycle is always going to a hot mess because there is no separation between the the technology, the models, and the products. AI in most minds is always going to be the thing they type words into to get a result, and this ambiguity muddies the argument.

A bit of pedanticism to start out: the person you quoted is technically correct, cause ChatGPT is just running your question through a prompt to get a query, sending it to Bing (or Shopify??), then feeding the pages that search returns back into the model to summarize.

I've been having more fun than expected with the game. I just wish the prog was the same as CN, where they get currency per match instead of being leveling rewards. ;-;

That last command bunker is about to have a Bad Time