Tech
TOPSHOT-US-TECHNOLOGY-AI-MICROSOFT-COMPUTERS
OpenAI CEO Sam Altman (Jason Redmond/AFP via Getty Images)
fabrications

AI used to transcribe medical visits is having wild and sometimes spooky “hallucinations”

The OpenAI tool sometimes adds made-up, unsettling details.

Jon Keegan

“ChatGPT can make mistakes. Check important info,” reads a warning at the bottom of OpenAI’s ChatGPT 4o chat interface.

That message is there because many of today’s AI tools are prone to “hallucinations,” where incorrect or “imagined” facts are included in a response to the user.

That may not matter much if it invents an extra ingredient for the chocolate-chip cookie recipe you asked for (even if it recommends adding glue to your pizza), but if you’re using an AI tool to help with more critical tasks — such as transcribing medical interviews — the results could be disastrous.

A new investigation by the AP found that OpenAI’s Whisper text-transcription tool is being widely used for medical transcription, despite the fact that it’s been found to hallucinate “racial commentary, violent rhetoric, and even imagined medical treatments,” according to the report. And considering how many medical professionals are using the tool, researchers found a troubling rate of hallucinations.

The problems persist even in well-recorded, short audio samples. A recent study by computer scientists uncovered 187 hallucinations in more than 13,000 clear audio snippets they examined.

That trend would lead to tens of thousands of faulty transcriptions over millions of recordings, researchers said.

OpenAI warns against using Whisper in “high-risk domains,” but that has not stopped its use in healthcare software like Nabla, which has been used to transcribe an estimated 7 million medical visits, according to the report, which included the following example:

A speaker said, “He, the boy, was going to, I’m not sure exactly, take the umbrella.”

But the transcription software added: “He took a big piece of a cross, a teeny, small piece ... I’m sure he didn’t have a terror knife so he killed a number of people.”

Adding to the problem is that Nabla deletes the original recordings of the interviews for “data safety reasons,” leaving no mechanism for verifying the transcriptions.

More Tech

See all Tech
tech

Apple to pay Google $1 billion a year for access to AI model for Siri

Apple plans to pay Google about $1 billion a year to use the search giant’s AI model for Siri, Bloomberg reports. Google’s model — at 1.2 trillion parameters — is way bigger than Apple’s current models.

The deal aims to help the iPhone maker improve its lagging AI efforts, powering a new Siri slated to come out this spring.

Apple had previously been considering using OpenAI’s ChatGPT and Anthropic’s Claude, but decided in the end to go with Google as it works toward improving its own internal models. Google, which makes a much less widely sold phone, the Pixel, has succeeded in bringing consumer AI to smartphone users where Apple has failed.

Google’s antitrust ruling in September helped safeguard the two companies’ partnerships — including the more than $20 billion Google pays Apple each year to be the default search engine on its devices — as long as they aren’t exclusive.

Apple had previously been considering using OpenAI’s ChatGPT and Anthropic’s Claude, but decided in the end to go with Google as it works toward improving its own internal models. Google, which makes a much less widely sold phone, the Pixel, has succeeded in bringing consumer AI to smartphone users where Apple has failed.

Google’s antitrust ruling in September helped safeguard the two companies’ partnerships — including the more than $20 billion Google pays Apple each year to be the default search engine on its devices — as long as they aren’t exclusive.

tech

Netflix creates new made-up metric for advertisers

It’s not quite WeWork’s community-adjusted EBITDA, but it’s also not quite a real number: Netflix announced today that it has 190 million “monthly active viewers” for its lower-cost ad-supported tiers. The company came up with the metric by measuring the number of subscribers who’ve watched “at least 1 minute of ads on Netflix per month” and multiplying that by what its research assumes is the number of people in that household.

It builds on Netflix’s previous attempt at measuring ad viewership with monthly active users, which is the number of profiles that have watched ads (94 million as of May). The MAV measurement, of course, is a lot bigger, and bigger numbers are more attractive to advertisers, who are spending more and more on streaming platforms.

“After speaking to our partners, we know that what they want most is an accurate, clear, and transparent representation of who their ads are reaching,” Netflix President of Advertising Amy Reinhard explained in a press release. “Our move to viewers means we can give a more comprehensive count of how many people are actually on the couch, enjoying our can’t-miss series, films, games, and live events with friends and family.”

Netflix last reported its long-followed and more easily understood paid membership numbers at the beginning of the year, when it crossed 300 million.

It builds on Netflix’s previous attempt at measuring ad viewership with monthly active users, which is the number of profiles that have watched ads (94 million as of May). The MAV measurement, of course, is a lot bigger, and bigger numbers are more attractive to advertisers, who are spending more and more on streaming platforms.

“After speaking to our partners, we know that what they want most is an accurate, clear, and transparent representation of who their ads are reaching,” Netflix President of Advertising Amy Reinhard explained in a press release. “Our move to viewers means we can give a more comprehensive count of how many people are actually on the couch, enjoying our can’t-miss series, films, games, and live events with friends and family.”

Netflix last reported its long-followed and more easily understood paid membership numbers at the beginning of the year, when it crossed 300 million.

tech

Ahead of Musk’s pay package vote, Tesla’s board says they can’t make him work there full time

Ahead of Tesla’s CEO compensation vote at its annual shareholder meeting tomorrow, The Wall Street Journal did a deep dive into how Elon Musk, who stands to gain $1 trillion if he stays at Tesla and hits a number of milestones, spends his time.

Like a similar piece from The New York Times in September, this one has a lot of fun details. Read it all, but here are some to tide you over:

  • Musk spent so much time at xAI this summer that he held meetings there with Tesla employees.

  • He personally oversaw the design of a sexy chatbot named Ani, who sports pigtails and skimpy clothes and for whom “employees were compelled to turn over their biometric data” to train.

  • The chatbot, which users can ask to “change into lingerie or fantasize about a romantic encounter with them,” has helped boost user numbers, which are still way lower than ChatGPT’s.

  • Executives and board members have told top investors in the past few weeks that they can’t make Musk work at Tesla full time. Board Chair Robyn Denholm explained that in his free time, Musk “likes to create companies, and they’re not necessarily Tesla companies.”

Like a similar piece from The New York Times in September, this one has a lot of fun details. Read it all, but here are some to tide you over:

  • Musk spent so much time at xAI this summer that he held meetings there with Tesla employees.

  • He personally oversaw the design of a sexy chatbot named Ani, who sports pigtails and skimpy clothes and for whom “employees were compelled to turn over their biometric data” to train.

  • The chatbot, which users can ask to “change into lingerie or fantasize about a romantic encounter with them,” has helped boost user numbers, which are still way lower than ChatGPT’s.

  • Executives and board members have told top investors in the past few weeks that they can’t make Musk work at Tesla full time. Board Chair Robyn Denholm explained that in his free time, Musk “likes to create companies, and they’re not necessarily Tesla companies.”

tech

Motion Picture Association to Meta: Stop saying Instagram teen content is “PG-13”

In October, Meta announced that its updated Instagram Teen Accounts would by default limit content to the “PG-13” rating.

The Motion Picture Association, which created the film rating standard, was not happy about Meta’s use of the rating, and sent the company a cease and desist letter, according to a report from The Wall Street Journal.

The letter from MPA’s law firm reportedly said the organization worked for decades to earn the public’s trust in the rating system, and it does not want Meta’s AI-powered content moderation failures to blow back on its work:

“Any dissatisfaction with Meta’s automated classification will inevitably cause the public to question the integrity of the MPA’s rating system.”

Meta told the WSJ that it never claimed or implied the content on Instagram Teen Accounts would be certified by the MPA.

The letter from MPA’s law firm reportedly said the organization worked for decades to earn the public’s trust in the rating system, and it does not want Meta’s AI-powered content moderation failures to blow back on its work:

“Any dissatisfaction with Meta’s automated classification will inevitably cause the public to question the integrity of the MPA’s rating system.”

Meta told the WSJ that it never claimed or implied the content on Instagram Teen Accounts would be certified by the MPA.

tech

Dan Ives expects “overwhelming shareholder approval” of Tesla CEO pay package

Wedbush Securities analyst Dan Ives, like prediction markets, thinks Tesla CEO Elon Musk’s $1 trillion pay package will receive “overwhelming shareholder approval” at the company’s annual shareholder meeting Thursday afternoon. The Tesla bull, like the Tesla board, has maintained that approval of the performance-based pay package is integral to keeping Musk at the helm of the company, which in turn is integral to the success of the company. Ives is also confident that investors will back the proposal allowing Tesla to invest in another of Musk’s companies, xAI.

“We expect shareholders to show overwhelming support tomorrow for Musk and the xAI stake further turning Tesla into an AI juggernaut with the autonomous and robotics future on the horizon,” Ives wrote in a note this morning.

The compensation package has received pushback, including from Tesla’s sixth-biggest institutional investor, Norway’s Norges Bank Investment Management, and from proxy adviser Institutional Shareholder Services.

Latest Stories

Sherwood Media, LLC produces fresh and unique perspectives on topical financial news and is a fully owned subsidiary of Robinhood Markets, Inc., and any views expressed here do not necessarily reflect the views of any other Robinhood affiliate, including Robinhood Markets, Inc., Robinhood Financial LLC, Robinhood Securities, LLC, Robinhood Crypto, LLC, or Robinhood Money, LLC.