Issue #11, coming at you!

On today’s quest:

— A correction.
— When the ChatGPT website is better than the mobile app.
— What happens when there are no ads.
— Depressing news.

Baby’s first correction: After I sent out the last newsletter, I noticed in my hallucination experiment chart that actually none of the models got the entire Google Ngram tag right. Perplexity got the part I was looking for right (“eng_us_2019”) but not the part that comes before it.

Tip: Use the website

The mobile version of ChatGPT has been instructed to give shorter answers than the website, leading to at least one hilarious example in which it seemed to object when asked to process a csv file with 15 rows and 8 columns and gave the user a template to fill in themselves.

If you need a long response, use the website.

Thoughts: What happens when there are no ads?

I haven’t written much about Google’s Bard, but I had a great experience using it for summarizing YouTube videos, which is a use they’ve been touting.

I had it list the main points of one of my Grammar Girl podcasts with time stamps, and it did a good job. I had it summarize a podcast I should listen to but don’t because it runs more than an hour, and I felt like I had gotten the gist and was more informed than before.

But then I had it summarize a 20-minute video about a coffee taste test I had seen last week. I had wanted to know which coffee won, but I had given up after watching 5 minutes and then skimming through the rest to try to find the answer. I just didn’t care enough to spend any more time. But with the Bard prompt “Tell me the main points of this YouTube video: [link],” I got the answer: a Kenyan light roast.

YouTube videos are often longer than they need to be because Google incentivized creators to make longer videos. It’s what Google wanted. But I hate having to watch videos to get information I want or need. It’s one of the most inefficient ways to get facts. I love this new summarizing prompt!

As someone who makes a big part of my living from ads on my website and in my podcast, I’ve never hated ads. I usually let the YouTube ads run because I recognize they are how creators get paid. But now, with Bard’s summarizing tool, I haven’t watched any ads. The coffee-taste-test guy didn’t get any ad revenue.*

Personally, I’m not worried about losing podcast listeners from people summarizing my show; I believe people listen because they like the listening experience. If they wanted to just get an answer, they already have a thousand other places to do that.

But I am very worried about websites losing traffic as people turn to AI for answers, and now I am also worried about people who make YouTube videos that are geared toward providing information instead of entertainment.

I’ve been thinking a lot about what it means if a big part of the web disappears, and this morning I spent 30 minutes going through my Pinterest recipe board and saving all my favorite recipes offline. Some of them are already gone — probably not because of AI at this point, but it made me realize I should have already had copies of recipes I wouldn’t want to lose. Running a small website was already a tough business, subject to the whims of the Google search algorithm and just life happening to the owner. And it’s going to get a lot harder.

News

I’ve seen a steady stream of stories recently about how AI is going to eliminate a frightening number of jobs, plus one story about a team that seems to be using it in a way that won’t lead to layoffs.

California governor calls for statewide generative AI training - The story says the move was inspired by a Goldman Sachs report saying AI “is expected to affect 300 million jobs.”

The Short-Term Effects of Generative Artificial Intelligence on Employment: Evidence from an Online Labor Market — Researchers looked at job offerings on freelance work platforms such as Upwork and found a reduction in both jobs and income for copywriters and graphic designers. Surprisingly, they found that highly skilled freelancers were actually more affected than lower level freelancers, which I should point out undermines the advice I gave in the last newsletter about protecting yourself by focusing on higher skilled work.

GPTs are GPTs: An Early Look at the Labor Market Impact Potential of Large Language Models (pdf) — A study by OpenAI found the same thing: “The projected effects span all wage levels, with higher-income jobs potentially facing greater exposure … Our analysis suggests that, with access to an LLM, about 15% of all worker tasks in the US could be completed significantly faster at the same level of quality. When incorporating software and tooling built on top of LLMs, this share increases to between 47 and 56% of all tasks.” [Emphasis added.]

Reshaping the tree: rebuilding organizations for AI — I will end with the one positive story I’ve seen because we sure as hell need one at this point. Ethan Mollick of the fabulous “One Useful Thing” newsletter, described how his small team at Wharton has used AI to get more done and nobody has been fired:

What is AI sidequest?

Using AI isn’t my main job, and it probably isn’t yours either. I’m Mignon Fogarty, and Grammar Girl is my main gig, but I haven’t seen a technology this transformative and potentially disruptive since the development of the internet, and I want to learn about it. I bet you do too.

So here we are! Sidequesting together.

If you like the newsletter, please share it with a friend.

^{* This time he did, from my previous failed attempt to get the answer from his video, but people like him won’t in the future.}

^{Written by a human.}