Louis Ingenthron @LouisIngenthron@social.cleverlibre.org

@jeffjarvis They're describing "failing fast" as a good thing.

Which, it can be, but not if you're not talking about *literal rocket science*.

@mloxton Has vaping been around long enough to kill people? Lung cancer is a slow death and that's a relatively new technology.

I was raised on country music, so I say this as a lifelong fan: Can we please stop making country songs celebrating murderous vigilantes and lynch mobs? I'd really appreciate it, thanks.

It's unbelievable to me that any modern #programming language in 2023 could still struggle with circular imports. wtf...

@willoremus If they've been fine with it being used to collect creepy amounts of data to predict their buying habits, I struggle to imagine they'll have a problem with this.

And if they do, OpenAI can just offer them a $5 Walmart gift card to shut up.

@micahflee Oh boy, you did not include the relevant context...

(A) It starts like this: "I did some research on AI’s water footprint using state-of-the-art **estimation** methodology" (emphasis added)

and (B) it doesn't give a timeframe for that usage. Is 700,000 liters of water a day or over the entire life of the technology?

@hnapel @micahflee @semioticstandard @x51@tilde.zone Like any tool, it depends on how it's used. As @semioticstandard mentioned, it's great as a learning tool, minimizing frustration and friction to help you pack your own brain with more knowledge as efficiently as possible.
I learned how to code in a whole new language over the weekend using an AI assistant to help. I don't need the assistant to program in that language for me; the skill is mine. The AI just helped me learn it.

@quotesofnote@zirk.us Yes, they can. Summarizing is recognizing the patterns of what's important and what's not (often by detecting repetition or placement). And all these algorithms really are is advanced pattern recognition tools.

Summarization is a damn-near perfect use case for what that technology is actually good at and meant to achieve.

You can try it yourself with ChatGPT now. Paste an article and ask it to summarize. In many cases, it produces output that's more succinct and easier to read than how a human would summarize it.

@brad_frost Funny, I had the exact opposite reaction when smartphones became popular...

@quotesofnote@zirk.us The point of the summary is that it's summarizing text it's given in full. The facts will be as good as the source piece. It's not making anything up; it's just remixing it. That's what makes it so good at that task.

@quotesofnote@zirk.us Correct. Anyone using LLMs to generate facts are using them wrong. That's like asking a calculator to write you a story.

But there are areas where they truly excel and will have massive business impact. One example is that they're phenomenal at taking a large block of text and succinctly summarizing it. That doesn't really require logic, it turns out... it's just patterns.

If we gear the tool to what it's good at instead of trying to shoehorn it into roles it's not good at, we can achieve some truly striking results.

@Paulatics @bingofuel I'm not looking for endless letters begging for donations, but I keep getting those from causes I support. 🤷‍♂️ I would treat a receipt of key points to incorporate from their message as a similarly robotic gesture that can easily be dismissed.

But asking you to repeat a whole speech verbatim? Beyond the pale.

There's a really interesting #study to be done about whether #LLMs are more helpful or not based on how you treat them...
Based on what training data is used, it's entirely possible...

I hope somebody is researching this.

Wow, Venus is super bright tonight.

@slightlyoff Isn't that the point of metadata?