By Alex Eule
I was about to start writing this column on Thursday afternoon when Facebook parent Meta Platforms launched a new chatbot. It wasn't procrastination, I swear. As a tech editor, I have to be up on this stuff. So, I spent a few minutes trying Meta's new generative AI tool. The column could wait. Sorry, editors.
But a few minutes turned into more. I dare you to go to Meta.ai without spending the next hour texting friends and family all of your artistic creations. I started by asking the bot to imagine cows in a field, and then I layered on turtles, all inside a baseball stadium. I have no idea why. The machine added each element in real time, as I typed.
My pièce de résistance was a series of cartoon giraffes bathing in a toilet. I suggested to my wife that it would be good for our bathroom wall. (She was less enthused.)
This is amazing technology -- and a total waste of time.
Meta's new chatbot, which is based on its latest open-source large-language model, called Llama 3, has all of the required features of an artificial-intelligence chatbot in 2024. I was able to get a solid list of album recommendations and a bulleted list of comparative data on sport-utility vehicles. But I'm not ready to update my view on the current state of generative AI; as an everyday consumer tool, I remain unimpressed. (My colleague Eric J. Savitz has a different take.)
Still, I'm trying to keep an open mind. The stock market has added trillions in new value on the back of AI. That has to count for something. I'll get to the "Wall Street" portion of this column, but first more on my quest for AI appreciation.
Just after my Meta.ai lark, I spoke to Dr. Peter Noseworthy, a professor of cardiology and medical director for business development at the Mayo Clinic. I had reached out to Mayo hoping to get beyond the chatbot chatter.
It didn't take long. Within a few minutes, Noseworthy was talking about the lifesaving implications of AI projects and clinical trials that Mayo is working on. He combined tech, business, and medicine in a way I hadn't heard before.
"One of the problems in healthcare," Noseworthy told me, "is that it's delivered as a service, and as a result, it's not scalable. There are probably aspects of medicine that, with large-language models and AI, can convert from a service to a product. And once that happens, it's scalable, it's portable, and people can carry these things in their pocket."
He says Mayo's team is using AI to find disease markers with a commonly performed electrocardiogram: "We found that it's actually a really powerful marker of all kinds of diseases that even people who read ECGs for a living don't rely on an ECG for."
Ultimately, those ECG tools can be used with a device like an Apple Watch, he says. "And then, the consumer carries, on his wrist, a clinical-grade ECG tool to deploy our AI."
We talked about phones as an omnipresent form factor. "We can detect all kinds of signals for health and disease in a patient's voice, particularly when it's trended over time," says Noseworthy. Similar signals could come from a phone's step counter and by measuring changes in a user's gait.
He compared AI and phones to a mother who quickly spots illness on a child's face. "That sort of clinical gestalt that comes from an astute observer is probably replicable with AI. And we're making dozens, if not hundreds, of recordings of our faces every day. There may be opportunities to leverage things like that."
Our phones have already taken over lots of roles in our lives, but the doting mother would be a new one.
Noseworthy also gave me a different framework for considering chatbots. "The closest we have to a patient-doctor interaction is probably a generative AI LLM solution in the form of a medical chatbot," he says. "If you'd asked me six months ago, I would have thought we were a few years out for something I trusted. Now, I think we might be six months from something that's clinically useful."
Mayo is actively working on these chatbots, Noseworthy says. It will still take a few more years, he adds, "to fully emulate clinical acumen."
In the short term, Noseworthy says a chatbot could help patients adjust medications and ask questions about drug interactions.
"The most mature version of that technology will allow patients to converse in natural language with a chatbot to discuss their symptoms or concerns and get high-quality, vetted medical information presented in an easily understood and natural feeling conversation," he told me.
This kind of personalized tool takes AI to another level.
I spoke to the founder and CEO of a company called Limitless, which says it's harnessing AI to give us memory superpowers. I had already tested out an earlier product from the company called Rewind.ai, which uses character and audio recognition to record everything that comes through a Mac screen and speaker, including calls, texts, and emails. Then, using a large-language model, the app can answer questions about our lives: "Tell me about my last conversation with Mom. What time is our lunch?"
The company's newer product, also called Limitless, offers apps for the Mac, Windows, and the web. This past week, it launched a wearable pendant with a microphone that users can attach to their clothing to record life's real conversations -- the ones that don't go through a device.
This all enables huge possibilities and, of course, serious privacy concerns. Dan Siroker, Limitless' founder, says the two aren't contradictory. "It starts with the fundamental premise -- that privacy should not come at the cost of convenience," he says. "Facebook and all ad-driven companies did the world a huge disservice by kind of making you think that should be a choice -- that you should have to compromise your privacy for convenience."
The new pendant offers a consent mode that will track only voices that have given permission to be recorded. The company says it has already received 10,000 orders for the $99 pendant, which will ship in August. Limitless says all of the data brought into its cloud are encrypted with a secret key that no one else can access, including the company. Data sent through large-language models for summarizing and transcribing is deleted after 30 days.
Limitless has raised $33 million from venture-capital firms like Andreessen Horowitz and New Enterprise Associates, as well as from OpenAI founder Sam Altman.
The key to Limitless, Siroker says, is that it's offering context to AI models. It's like providing a prompt to a chatbot that says, "I like the colors blue and green. Pick me the best Benjamin Moore paint color."
With Limitless, the context is more than just liking blue and green, though. It's potentially sending every conversation you've ever had. Ultimately, that's how an AI model trained on generic internet data will bring personalized answers to our lives.
So, what does all this mean for stocks? The promise of AI has already driven a massive market rally. The irony is that Apple, the company with more context about our lives than any other, has been left behind.
Consider the Apple Health app on my phone. That app has years worth of data about my daily step count, my heart rate, the number of times I stand in a day, my exposure to loud sounds, and my blood oxygen level. A few rows over on my home screen, I have thousands of photos with location and time stamps. And then there are my texts. With my permission, Apple could conceivably feed all of this into a large-language model to create an absurdly detailed journal of my life. It would be a self-help book on steroids.
Yes, there are privacy issues, but we've already given all of these data to Apple. I'm fine with the company putting it to better personalized use.
The current knock on Apple is that unlike Big Tech rivals Meta and Alphabet, or the start-up OpenAI, it has yet to create its own large-language model.
I have no proprietary information on whether it will. CEO Tim Cook has said that Apple is putting a lot of time into AI, with details coming later this year. The company will probably have major announcements in June at its annual developers conference. For now, though, investors have all but given up on Apple and AI. And that feels like a major oversight.
Late last month, Melius Research analyst Ben Reitzes published a note about the Cupertino, Calif., company's AI efforts titled, "Apple May Still Have the Last Laugh."
Reitzes has covered Apple for more than two decades, as an analyst and banker. "The reason we wrote the note was to say, remember, these guys are market makers. And maybe we should hear them out," Reitzes told me. "Even going back to the original example of when Steve Jobs was trying to build iTunes. He didn't make the music. He made enjoying the music better."
Reitzes says Apple won't have to make a world-beating AI model to be a winner in this new world. "They don't need to be Microsoft. They don't need to be OpenAI or Gemini. Apple needs to be Apple, and that is providing an interface that allows you to enjoy where apps are going in this new world. We think they can do it, and they are going to start to lay the groundwork for that carefully, potentially with a splash in June."
For now, Apple is down 14% this year, versus a 4% gain for the S&P 500 index and a 56% gain for AI leader Nvidia.
Melius says there are legitimate concerns about Apple's iPhone sales in China, where the company seems to be quickly losing market share. And he understands why some investors are worried about the company's AI efforts thus far.
But Apple has been here before. From September 2012 to September 2013, between the release of the iPhone 5 and the iPhone 5S, Apple shares fell 33% versus a 18% gain for the S&P 500. At the time, Samsung Electronics was gaining steam with its larger-screen phones.
(MORE TO FOLLOW) Dow Jones Newswires
April 19, 2024 20:09 ET (00:09 GMT)
Copyright (c) 2024 Dow Jones & Company, Inc.
Comments