Joe @twitskeptic@qoto.org

@ct_bergstrom I find them to be quite useful in certain domains, like code generation. The scary part is the rush to integrate them into consumer products (like search) where users are forced to use them and may not even be aware that they are using them.

@masonporter GPT-4 is trickier.

@ct_bergstrom Lots of privacy problems with LLMs, but I'm having a hard time understanding why this is one of them given that anyone who knows how to click on the History button can get the same info.

@wc_ratcliff@ecoevo.social If only - LLMs will obligingly insert typos if requested.

@ct_bergstrom

Bing's version of GPT-4 seems so inferior to OpenAI's that I'm wondering if Microsoft screwed up the fine tuning in their rush to get it out. Or maybe they are using an earlier version. But I do get a reasonable answer with your question with GPT-4.

GPT-4 does give me brand recommendations but, so far at least, only when I ask. Pretty obvious how this will play out in the long term.

Even GPT-4 thinks that Thomas Friedman's NYTimes opinion piece is kind of ridiculous.

"The article is a mix of measured analysis and some hype. It rightly highlights the potential transformative impact of AI technologies like ChatGPT, the benefits they can bring, and the need for collaborative efforts to govern and regulate them. However, the excitement and awe expressed in the piece can be perceived as somewhat hyped, particularly when it comes to the comparison with nuclear energy and the characterization of the current era as a "Promethean moment."

While AI technologies are indeed impressive and transformative, it is essential to remain cautious and critical when discussing their capabilities and potential consequences. A more measured approach would involve acknowledging both the potential benefits and the limitations of AI technologies, as well as the challenges that come with governing and regulating them."

https://www.nytimes.com/2023/03/21/opinion/artificial-intelligence-chatgpt.html

@talyarkoni@sigmoid.social Perhaps some of this criticism is driven by the inappropriate rush to deploy these models in, for example, medical applications. https://news.nuance.com/2023-03-20-Nuance-and-Microsoft-Announce-the-First-Fully-AI-Automated-Clinical-Documentation-Application-for-Healthcare

@Riedl That's why it's so reassuring that Microsoft is already deploying GPT-4 in medical transcription applications. https://news.nuance.com/2023-03-20-Nuance-and-Microsoft-Announce-the-First-Fully-AI-Automated-Clinical-Documentation-Application-for-Healthcare

Deploying LLMs in medical applications is a really irresponsible move by Microsoft. https://newsie.social/@STAT/110062564971170267

@talyarkoni@sigmoid.social I'm guessing that available attention is already saturated, so I think more AI generated media will just lower the value of most media in general. And there will be a huge knockoff problem if it's cheap and fast to create a non-identical but similar version of a big media hit. So I'd guess your 2nd choice will account for the majority of successes.

But who knows? These big technological shifts are always so unpredictable.

@talyarkoni@sigmoid.social Regarding 3): we're already there with random "influencer" success on YT or TT. The probability of success is very low, though, and will be even lower if LLMs increase the pool of people who can implement "good ideas".

Gentle introduction to LLM prompt engineering without all of the confusing formalism that many of papers resort to. https://lilianweng.github.io/posts/2023-03-15-prompt-engineering/

@digiphile @jeffjarvis My mistake - I didn't realize the linked article was over 10 years old.

@digiphile @jeffjarvis Why would anyone believe any claims that Twitter management makes since Musk took over?

@devezer One of the problems I had in my attempt to learn jazz piano (I was trained classically) is that I never was trained to play by ear, so it's difficult to improvise. I'd guess that's a consideration for klezmer as well.