Good morning. Today is the kind of day the “Sunday Scaries” phrase was made for—the last day of a long holiday weekend and, for many Americans, a day of headache-inducing travel. But we hope to make it a bit better with a fascinating deep dive into a subject you may have had to explain to your grandma at the Thanksgiving table this week: generative AI.

Yesterday marked the second anniversary of the launch of ChatGPT, and since then, AI has rapidly gone from a sci-fi storyline to an everyday reality, all without initiating the apocalypse (). Here you’ll find all sorts of stories on what’s new in the AI world since OpenAI changed the game two years ago.

We’ll be back tomorrow with the usual newsletter. Hope everyone had a refreshing break before the mad sprint to the end of the year.

 —Dave Lozo, Molly Liebergall, Matty Merritt, Cassandra Cassidy, Adam Epstein

It’s ChatGPT vs. everybody

ChatGPT arrived on the scene two years ago and has emerged as the default generative AI bot for many. It’s not at a level where it has become a verb like “Google,” but ChatGPT is used by 200 million people every week, up from 100 million over the same period a year ago, according to OpenAI.

But as ChatGPT has grown, competitors have emerged. There’s Google’s Gemini, Anthropic’s Claude, Meta’s Llama, Microsoft’s Copilot, and Perplexity’s...Perplexity (they kept it simple), just to name a few. There’s also Grok from xAI, the chatbot with a sense of humor that has access to X (this is meant to be a selling point).

Integrate expectations

Apple was late to the party when it came to integrating generative AI into its products, but the company announced in June that it would add ChatGPT to the iOS, iPadOS, and macOS platforms as part of its Apple Intelligence push. Now when you ask Siri a question, you can get answers drawn from “world knowledge” as opposed to “personal knowledge” of things in your calendar or messages.

• Bloomberg reported in June that Apple could have a similar integration deal with Google this fall, but that date has likely been pushed back to allow OpenAI a longer period of exclusivity on its devices.

Also: Amazon invested an additional $4 billion into Anthropic after putting $2.75 billion into the company in March and another $1.25 billion in 2023. Amazon is striving to improve Alexa and give it a more conversational tone powered by Claude. Microsoft, meanwhile, was forced to delay its rollout of Copilot Plus PCs first in June and then in October due to security concerns, but it’s now expected to release them in December.

Search and destroy

Google’s era of search dominance could soon end following a federal ruling in August that the company maintains an illegal online search monopoly. The DOJ is pressing the company to sell its popular web browser, Chrome, opening the door for disruption in its $2 trillion search empire.

Search party: It used to be that ChatGPT could generate text but couldn’t retrieve information from around the web. But in October, OpenAI released ChatGPT Search (formerly SearchGPT) that can provide conversational answers (with citations to the sources it steals from uses) to questions that usually would go to Google. Meta is reportedly pursuing something similar, while Perplexity’s AI-powered search engine is considered the leader for people searching to shop.

Google’s still on top, for now: Since ChatGPT launched two years ago, Google parent company Alphabet has continuously posted strong earnings and has shown few signs of vulnerability.—DL

AI wants to be your friend

In August, an ad for the wearable AI pendant “Friend” made everyone online shout “BLACK MIRROR!” until they turned blue. The $100 amulet can be worn around your neck and is always listening so it can text you throughout the day. Creator Avi Schiffmann said he made the device to help battle loneliness.

Despite most reactions to the pendant boiling down to “this sucks,” a growing number of artificial intelligence startups are developing AI chatbot companion apps. Most apps are free to download but can charge anywhere from $6 to $16 monthly for a premium subscription with added features like requesting “selfies” from your AI friend.

• Apps like Nomi or Replika let you build your dream friend/mentor/romantic partner Sims-style.
• Chatbot-maker Character.AI raised $150 million last year without any revenue to show before Google poached its founders and staff last year.

But some use apps for...exactly what you think

Dirty talk with the robots. A group of researchers analyzed a million ChatGPT interaction logs and found that the second most popular use was sexual role-playing. Apps like Candy.ai and EVA offer erotic role-playing features, allowing users to build AI lovers who can send raunchy texts and even some cursed AI-generated nudes (we’ve heard).—MM

The AI ‘super users’ who swear by their workflows

Most people have kept doing their jobs the old-fashioned way (sans chatbots), but a growing fraction of the workforce is going Jarvis mode. Professionals who have integrated AI tools like ChatGPT or Microsoft Copilot into their everyday toilings, aka “super users,” told the Washington Post that they’ve shaved up to 15 hours off their to-do lists.

Tiny but mighty. Only 4% of workers use large language models on the job daily, while 67% never do, per a recent Gallup survey. Most of the action happens behind a desk and probably also a Sweetgreen salad. According to the poll:

• White-collar workers are the heaviest users, with 15% using AI weekly. Only 54% reported they never touch the stuff professionally, compared with 81% of people in frontline industries.
• Part of the slow uptake could be due to a lack of top-down guidance—60% of white-collar workers and 70% of workers overall said their workplaces don’t have clear AI policies.

Gen Z loves AI: The rare super user is tapping chatbots to analyze data sets, transcribe audio and shorthand notes, teach them Excel formulas, build trend reports, check grammar, caption videos, roleplay meetings, and assist with more tedium. The youngest generation of white collars especially stan AI, according to a Google Workspace survey published last week:

• Ninety-three percent of Gen Z respondents said they use at least two AI tools, like DALL-E or Otter.ai, each week, compared with 79% of millennials.
• Eighty-eight percent of Gen Z respondents reported tapping AI to start overwhelming projects.

Cautious embrace: AI experts warn against sharing sensitive info with tools lacking enterprise-grade security. And to tamp down the risk that bots will spit out errors or bias, some professionals ask large language models more clarifying questions before producing results.—ML

Thank you for this plate, data centers

Microplastics might be inside our food, but artificial intelligence is all over it . With farms, restaurants, and grocery stores planting more AI in their operations, our meals are increasingly intertwining with server farms.

Out in the fields…farmers and ranchers are getting government incentives to optimize their operations ASAP. Humanity will need 70% more food by 2050, when the world population hits 10 billion, according to the American Farm Bureau Federation trade group. Every efficiency helps:

• Crop-surveying drones, GPS tools, and hands-free combine steering are some of the most common food-raising AI tech, according to the BBC—87% of agricultural businesses already used some type of AI by late 2021.
• Other recent innovations include driverless tractors and precision sensors that can reduce herbicide spraying by up to two-thirds, according to John Deere, which has been AI-ifying equipment for decades.
• In September, the world’s first large-scale indoor vertical berry farm opened in Virginia, where AI-driven growth analysis is supposed to help produce 4+ million pounds of strawberries annually.
• European scientists even made an algorithm this year for interpreting oinks, designed to keep farm pigs as happy as possible.

At fast-food/casual chains…Yum Brands—parent to Taco Bell, Pizza Hut, and KFC—is scaling voice-activated drive-thru tech, AI-enabled shift-scheduling, digital menu boards, and AI-based marketing. Chipotle recently invested in the AI platform Lumachain, and over the summer, Wendy’s used Palantir AI tools to stay stocked on Frosty supplies during its $1 deal.

In supermarket aisles…grocery companies including Albertsons and Family Dollar are ramping up their behind-the-scenes AI use for merchandise management and demand forecasts, according to Grocery Dive. On the dystopian front, Kroger denied allegations from members of Congress that its new electronic shelf labels could lead to facial recognition-based surge pricing.

Looking ahead…grocers are planning to quadruple their AI spending by 2025, generating $113 billion in value, according to a recent report from food industry groups.—ML

The year in AI fails (and 1 win)

No matter how you view AI—as an invention that will change our lives for the better or bring our civilization to its knees forever—we can all agree this young technology is far from perfect as it stands now.

With many more chatbots available in 2024, there were so many more fails compared to last year. Here are the most notable ones (along with one AI win).

The fails

Willy’s Chocolate Experience: AI-generated images made it seem like Wonka’s factory would come to life for families in Scotland willing to pay $44. Instead, they arrived at a nearly empty warehouse that resulted in the police being called (and sad Oompa Loompa memes).

The “Goodbye Meta AI” hoax: Sometimes the fail is by the humans, not the AI. More than 600,000 people—including Ted 2 star Tom Brady—posted a message on IG saying they did not give Meta permission to use their images to train its AI. That message, shockingly, does not prevent Meta from legally using a person’s data or likeness.

Google AI Overview mistakes: There are AI hallucinations and then there are the responses AI Overview spit out this year. Some of the unhelpful answers in search results included putting glue on pizza, eating rocks for nutrients, and eating boogers to boost the immune system.

Honorable mentions: A parcel delivery service chatbot swore at customers; an NYC chatbot said it was legal for businesses to commit illegal acts; Grok AI accused NBA star Klay Thompson of vandalizing homes when it misunderstood what “shooting bricks” meant in basketball parlance.

There were serious issues, too: The National Highway Traffic Safety Administration announced that its investigation into Tesla found that the company’s Autopilot feature was involved in 13 fatal accidents dating back to 2021. A woman is suing Character.AI, saying her 14-year-old son committed suicide with encouragement from the company’s chatbot. And pornographic deepfakes of Taylor Swift resulted in Congress calling for laws to make the creation of such images illegal.

One win: A painting of Alan Turing created by an AI-powered humanoid robot sold for almost $1.1 million at auction—which could also be looked at as a fail, considering a human taped a banana to a wall and got $6.2 million for it.—DL

✳︎ A Note From Pacaso

This is a paid advertisement for Pacaso’s Regulation A offering. Please read the offering circular at invest.pacaso.com.