PR pros are using AI ... a lot

Plus, free webinars and hilarious stupid AI fails

Issue #22. Courses are popping up

On today’s quest:

— AI in PR
— Free AI webinars
— Hilarious AI failures

PR pros are using AI and plan to use it even more

MuckRack surveyed 1,001 PR professionals at the very end of last year and found that AI already has a big footprint in the industry: 87% use or plan to use AI in their work.

Further, of the 64% already using it, 74% report an increase in the quality of their work, and 89% say they complete projects more quickly with AI. The most common use was writing copy for social media.

Reassuringly, 95% say they always edit the AI output. A quick registration will give you the whole report from MuckRack.

Free AI webinars are popping up

I don’t know anything about these personally, but they’re free.

‘AI for beginners’

I registered for this AI for Beginners course to see what it would be like, and it’s clearly a loss leader. I’ve already gotten multiple emails about other courses, but I also got an interesting email about creating a company AI policy. It’s not spammy, but they will try to upsell you.

Writing With Technologies

This one has much more detail on the website on what will be covered:

The Studio, a UK organization, is presenting a series of free webinars on AI writing. According to the website the series will be:

Featuring talks from writers, creators, academics, publishing professionals and AI experts, the series will explore questions such as: 

  • How can writers use AI tools to support their creativity? 

  • In what ways is automation transforming publishing business models and content production? 

  • How might we update creative writing curriculums to effectively support students navigating an evolving AI landscape? 

  • How can interactive machine learning enable richer storytelling and audience engagement in live performance? 

The first webinar of the series seems to have a gaming-creation focus and is at noon (UK time) on Wednesday, January 24. Register at The Ticket Source.

News

More automated bot failures

Following on the item from the last newsletter about error messages from ChatGPT bots showing up in tweets, they’re now showing up in Amazon listings. For example, the title of the listing for a dresser read "I'm sorry but I cannot fulfill this request it goes against OpenAI use policy."

Futurism, which reported the weirdness, found multiple examples of such listings, some of which have been taken down from Amazon but archived elsewhere. I searched myself for the phrase “I Cannot Fulfill This Request” and still found multiple live listings.

Futurism speculates that the “admittedly hilarious product listing suggests companies are hastily using ChatGPT to whip up entire product descriptions, including the names — without doing any degree of proofreading — in a likely failed attempt to optimize them for search engines and boost their discoverability.”

Ars Technica notes that using AI in this way, while stupid, isn’t against Amazon’s policies. In fact, Amazon has its own AI tool meant to help vendors write product listings.

AI voice narration comes to The Atlantic

Stuart Thompson, a tech reporter for the New York Times, noticed that The Atlantic is using AI voices from Eleven Labs to narrate articles on its site. I ran out of free articles before I found one that had any narration, but a podcast I listen to used a voice clone like this a few weeks ago, and I didn’t like it.

I use Eleven Labs to read my text to me for proofreading, and I like how natural the voice sounds for that use. I’m not inherently opposed to consensually used AI voices, but when it appeared on the podcast, I found it harder to pay attention. It was a little too fast, and the intonations were a little wrong. It was close, but not quite right.

Fail: Microsoft Copilot hallucinations

It’s been a while since I had a straight-up, obviously wrong answer from AI. Here’s what happened:

On social media, I asked how people feel about grammar, spelling, and punctuation on restaurant menus, and I got a bunch of interesting comments. I want to write a piece on it, but gathering all the comments into a Word document to make them easier to work with is tedious. I wondered if AI could do it for me.

I tried ChatGPT first, and it said that although it can access “a select list of websites,” it cannot access social media sites, which is disappointing but fine.

Next, I tried Microsoft’s Copilot — which I’ve only used once or twice before. I thought it might be able to access the comments since Microsoft owns LinkedIn.

But it completely made up a response. The comments were just generic statements about writing that seem to have been triggered by the appearance of “grammargirl” in the URL. Also, why is it linking to salesdorado.com as a source? Is that an ad?

I was surprised this level of hallucination is still happening. Be careful out there, folks.

Tip: I unrecommend Microsoft Copilot

There’s a second reason the Copilot failure is interesting: I read a lot of buzz last week about Copilot upgrading to the newest OpenAI model. People were saying you could now essentially get the paid version of ChatGPT free through Copilot.

I had done some tests of my own and concluded Copilot still isn’t as good as ChatGPT Plus — not because it hallucinated, but because the answers weren’t as complete. And in fact, I didn’t even think the answers were as good as the free version of ChatGPT.

I was debating whether to do more tests and whether to recommend it to you as something to try — again because I was seeing so many other people make a big deal out of it. But no, after getting the hallucination, Copilot wouldn’t be my first, second, or third choice.

I’ll put an example comparison at the very end in case anyone is interested, but the experience made me realize it’s not just the model that matters, but also the interface. 

Funny: AI assistants hanging out in the break room

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 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.

Example of ChatGPT Plus versus Copilot

Having heard Copilot now had an engine equivalent to ChatGPT Plus, I did a few tests. I still think ChatGPT Plus is better. Here’s an example:

Copilot gave a less complete answer than ChatGPT Plus, but provided links to its sources and a link to suggested further reading.

Microsoft Copilot (with GPT-4 Turbo, free)

Prompt: Can you compare the energy usage of doing one ChatGPT query versus watching a 20-minute YouTube video??

Response: I’m sorry, but I don’t have enough information to provide a direct comparison between the energy usage of doing one ChatGPT query and watching a 20-minute YouTube video. However, according to a commentary by the International Energy Agency, streaming a Netflix video in 2019 typically consumed around 0.077 kWh of electricity per hour. This is much less than the original estimate of 6.1 kWh per hour, which was later corrected to 0.78 kWh per hour. The article also notes that the carbon footprint of streaming video is relatively small compared to other activities, such as driving a car or flying on a plane.

It’s difficult to estimate the energy usage of a ChatGPT query, as it depends on a variety of factors such as the size of the model being used, the amount of data being processed, and the hardware being used to run the model. However, it’s worth noting that many AI models are designed to be energy-efficient, and researchers are constantly working to develop more sustainable AI technologies.

In conclusion, while I can’t provide a direct comparison between the energy usage of a ChatGPT query and watching a 20-minute YouTube video, it’s clear that both activities consume energy. However, the carbon footprint of streaming video is relatively small compared to other activities, and many AI models are designed to be energy-efficient.

[Note: It also provided links at the bottom to five “learn more” sources.]

ChatGPT also declined to give a firm answer but included more detail about what would factor into a comparison. It also seemed less defensive about the concept. For example, it didn’t mention the energy cost of driving a car.

ChatGPT-4 Plus (paid)

Prompt: Can you compare the energy usage of doing one ChatGPT query versus watching a 20-minute YouTube video?

Response: Comparing the energy usage of doing one ChatGPT query to watching a 20-minute YouTube video involves considering various factors, as energy consumption in both cases depends on multiple parameters.

1. ChatGPT Query Energy Consumption:

- Server-Side Processing: The majority of the energy consumed during a ChatGPT query is on the server side, where complex AI models run on powerful data centers.

- User's Device: The energy used by the user's device (like a computer or smartphone) while typing and sending a query is relatively minimal.

- Network Data Transfer: Some energy is used in transferring data between the user's device and the servers.

2. Watching a 20-Minute YouTube Video:

- Streaming Server Energy: YouTube videos are streamed from data centers, consuming energy for video hosting and data transmission.

- User's Device: The energy consumption here is typically higher than for a ChatGPT query, as video playback, especially at higher resolutions, requires more processing power.

- Network Data Transfer: Streaming a video, particularly in high definition, can use significantly more data than a text-based ChatGPT query, leading to higher energy consumption in data transfer.

Key Differences:

- Data Size and Transmission: Video streaming generally involves larger amounts of data compared to text-based queries, leading to potentially higher energy usage.

- User's Device Processing: Video playback is more resource-intensive than processing text responses.

- Resolution and Quality: Higher resolutions and quality settings in video streaming can exponentially increase energy usage.

- Duration of Activity: A ChatGPT query is typically a quick interaction, whereas a 20-minute video consumes energy continuously over a longer period.

In summary, while both activities consume energy, watching a 20-minute YouTube video is likely to consume more energy overall due to the larger data sizes, continuous nature of the activity, and higher processing demands on the user's device. However, quantifying the exact energy usage would require specific details about server efficiency, data transfer rates, device power consumption, and video quality.

Written by a human.