Issue #44

On today’s quest:

— OpenAI releases much anticipated new model
— Useful search tips for both getting AI and avoiding AI
— A new high school has mostly AI teachers
— AI-generated news anchors popping up
— Which AI uses are OK for journalists?
— If AI could have a mental breakdown, this could do it

OpenAI releases much anticipated new model

OpenAI surprised the AI world by launching its much anticipated new model last week. Code named Project Strawberry, OpenAI o1 is dramatically better at math, science, and coding than previous models,* but is actually a bit worse at writing and no better at editing, according to benchmarks released by the company.

Christopher Penn speculates we may end up with more models that are optimized for different tasks: logic versus creativity, for example.

The new model achieves its gains by using chain-of-thought reasoning, which is similar to how people have been prompting chatbots to “think a problem through step by step” to get better results.

A good example of what “reasoning” means comes from Ethan Mollick who has had access to the model for a while: the new model can solve hard crossword puzzles that stumped all previous models. Whereas previous models would get the first word wrong and become stuck, OpenAI o1 can work its way out of the loop.

The new model is currently available to only ChatGPT Plus and Team subscribers.

Useful Search Tips

I like to control whether I’m getting AI or not, and I’ve seen a few tips about how to do that.

🔎 Try adding “before 2022” in your searches. Darrly Mott on Bluesky says, “Every artist I know has started adding ‘Before 2022’ in every Google Image Search” to avoid getting AI images. This came after a discussion about how Google Image searches are returning images of anatomically incorrect animals, such as animals with antlers when they shouldn’t have them. In the comments, someone said they also check dates like this when they look for books on Amazon.

🔎 You can also turn off Google’s AI Overviews in Chrome and Firefox.

🔎 You can also get old-school “ten blue links” results from Google by creating web settings that use an “&udm=14” tag (instructions) or by starting your search at udm14.com.

🔎 Google has added a quick way to access Gemini straight from a browser. Just type @Gemini in the address bar.

A new high school with mostly AI teachers

David Game College in London is offering a program for students age 15 to 17 that will be taught almost exclusively by AI with minor “learning coach” support from humans. It appears to be similar to what we would call an alternative high school in the United States, targeting homeschoolers and students who “have faced challenges in maintaining a presence within a conventional school setting.” The school says learning coaches will be in the room to monitor behavior and teach classes AI can’t do well such as art and sex ed. — David Game College, Sky News

AI-generated news anchors popping up

The story of AI news anchors broke out in the mainstream U.S. news last week when a news organization in Hawaii started using cringey AI-generated news anchors. But Reuters Institute reported a few weeks ago that organizations around the world are rolling out AI news anchors, which most often take the form of hot young women.

Some avatars have both professional and personal social media accounts as companies try to turn them into trusted influencers. Fifty-seven percent of people who say they get news on TikTok say they pay the most attention to online personalities and influencers, whereas only 34% say they pay attention to journalists and news media.

I also found it interesting that the reaction is different on different platforms: “Most of the negative feedback he receives is on Facebook, he says, while on LinkedIn most responses are positive and on Instagram there’s a mix.” I see a huge difference in how people respond to AI stories on social media too, with Bluesky being the most negative, and LinkedIn and Threads being the most positive.

Which AI uses are OK?

The Reuters Institute article also included this fascinating chart about what AI activities people find acceptable for journalists. Editing the spelling and grammar of an article, translating an article, and making charts and infographics top the list, and creating artificial presenters comes in last. Writing an article — which I think is over-represented in discussions of AI, at least outside the context of school essays — has a slightly negative perception.

If AI could have a mental breakdown, this could do it

This was posted before the release of OpenAI o1. I wonder how it would do.

Do you think it would eventually go all Matrix on him and reply “There are no R’s.”

Maybe Claude will give ChatGPT a shoulder to cry on

Riley’s follow-up post: “As many have pointed out, there are much better ways to prompt this. Also, it’s silly to use an LLM here, as there’s little hope of making them 100% reliable at counting.

What’s notable IMO isn’t that it can’t count, but that it doesn’t see it can’t count …”

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.

^{* People have pointed out problems in the past with using standardized tests to measure AI quality, but FWIW, OpenAI says that whereas GPT 4o could solve only 12% of the questions on a math test designed for the brightest students, OpenAI 1o solved 74% to 93%, which would have placed it in the top 500 math students in the United States.}

^{Written by a human.}