Biz & IT – Page 16

Trump plans to dismantle Biden AI safeguards after victory

AI, AI regulation, Biden, Biz & IT, Donald Trump, Joe Biden, large language models, machine learning, Policy, US government, White House / DJ Henderson / November 6, 2024

That’s not the only uncertainty at play. Just last week, House Speaker Mike Johnson—a staunch Trump supporter—said that Republicans “probably will” repeal the bipartisan CHIPS and Science Act, which is a Biden initiative to spur domestic semiconductor chip production, among other aims. Trump has previously spoken out against the bill. After getting some pushback on his comments from Democrats, Johnson said he would like to “streamline” the CHIPS Act instead, according to The Associated Press.

Then there’s the Elon Musk factor. The tech billionaire spent tens of millions through a political action committee supporting Trump’s campaign and has been angling for regulatory influence in the new administration. His AI company, xAI, which makes the Grok-2 language model, stands alongside his other ventures—Tesla, SpaceX, Starlink, Neuralink, and X (formerly Twitter)—as businesses that could see regulatory changes in his favor under a new administration.

What might take its place

If Trump strips away federal regulation of AI, state governments may step in to fill any federal regulatory gaps. For example, in March, Tennessee enacted protections against AI voice cloning, and in May, Colorado created a tiered system for AI deployment oversight. In September, California passed multiple AI safety bills, one requiring companies to publish details about their AI training methods and a contentious anti-deepfake bill aimed at protecting the likenesses of actors.

So far, it’s unclear what Trump’s policies on AI might represent besides “deregulate whenever possible.” During his campaign, Trump promised to support AI development centered on “free speech and human flourishing,” though he provided few specifics. He has called AI “very dangerous” and spoken about its high energy requirements.

Trump allies at the America First Policy Institute have previously stated they want to “Make America First in AI” with a new Trump executive order, which still only exists as a speculative draft, to reduce regulations on AI and promote a series of “Manhattan Projects” to advance military AI capabilities.

During his previous administration, Trump signed AI executive orders that focused on research institutes and directing federal agencies to prioritize AI development while mandating that federal agencies “protect civil liberties, privacy, and American values.”

But with a different AI environment these days in the wake of ChatGPT and media-reality-warping image synthesis models, those earlier orders don’t likely point the way to future positions on the topic. For more details, we’ll have to wait and see what unfolds.

Trump plans to dismantle Biden AI safeguards after victory Read More »

Corning faces antitrust actions for its Gorilla Glass dominance

antitrust, Apple, Biz & IT, Corning, European Commission, Gorilla Glass, IPhone, Tech / Shannon Garcia / November 6, 2024

The European Commission (EC) has opened an antitrust investigation into US-based glass-maker Corning, claiming that its Gorilla Glass has dominated the mobile phone screen market due to restrictive deals and licensing.

Corning’s shatter-resistant alkali-aluminosilicate glass keeps its place atop the market, according to the EC’s announcement, because it both demands, and rewards with rebates, device makers that agree to “source all or nearly all of their (Gorilla Glass) demand from Corning.” Corning also allegedly required device makers to report competitive offers to the glass maker. The company is accused of exerting a similar pressure on “finishers,” or those firms that turn raw glass into finished phone screen protectors, as well as demanding finishers not pursue patent challenges against Corning.

“[T]he agreements that Corning put in place with OEMs and finishers may have excluded rival glass producers from large segments of the market, thereby reducing customer choice, increasing prices, and stifling innovation to the detriment of consumers worldwide,” the Commission wrote.

Ars has reached out to Corning for comment and will update this post with response.

Gorilla Glass does approach Xerox or Kleenex levels of brand name association with its function. New iterations of its thin, durable glass reach a bit further than the last and routinely pick up press coverage. Gorilla Glass 4 was pitched as being “up to two times stronger” than any “competitive” alternative. Gorilla Glass 5 could survive a 1.6-meter drop 80 percent of the time, and 6 built in more repetitive damage resistance.

Apple considers Corning’s glass products so essential to its products, like the ceramic shield on the iPhone 12, as to have invested $45 million into the company to expand its US manufacturing. The first iPhone was changed very shortly before launch to use Gorilla Glass instead of a plastic screen, per Steve Jobs’ insistence.

Corning faces antitrust actions for its Gorilla Glass dominance Read More »

Anthropic’s Haiku 3.5 surprises experts with an “intelligence” price increase

AI, AI pricing, Anthropic, Anthropic Claude, API, Biz & IT, chatbots, chatgpt, chatgtp, Claude, Claude 3.5, Claude 3.5 Haiku, Claude Haiku, large language models, machine learning / Shannon Garcia / November 5, 2024

Speaking of Opus, Claude 3.5 Opus is nowhere to be seen, as AI researcher Simon Willison noted to Ars Technica in an interview. “All references to 3.5 Opus have vanished without a trace, and the price of 3.5 Haiku was increased the day it was released,” he said. “Claude 3.5 Haiku is significantly more expensive than both Gemini 1.5 Flash and GPT-4o mini—the excellent low-cost models from Anthropic’s competitors.”

Cheaper over time?

So far in the AI industry, newer versions of AI language models typically maintain similar or cheaper pricing to their predecessors. The company had initially indicated Claude 3.5 Haiku would cost the same as the previous version before announcing the higher rates.

“I was expecting this to be a complete replacement for their existing Claude 3 Haiku model, in the same way that Claude 3.5 Sonnet eclipsed the existing Claude 3 Sonnet while maintaining the same pricing,” Willison wrote on his blog. “Given that Anthropic claim that their new Haiku out-performs their older Claude 3 Opus, this price isn’t disappointing, but it’s a small surprise nonetheless.”

Claude 3.5 Haiku arrives with some trade-offs. While the model produces longer text outputs and contains more recent training data, it cannot analyze images like its predecessor. Alex Albert, who leads developer relations at Anthropic, wrote on X that the earlier version, Claude 3 Haiku, will remain available for users who need image processing capabilities and lower costs.

The new model is not yet available in the Claude.ai web interface or app. Instead, it runs on Anthropic’s API and third-party platforms, including AWS Bedrock. Anthropic markets the model for tasks like coding suggestions, data extraction and labeling, and content moderation, though, like any LLM, it can easily make stuff up confidently.

“Is it good enough to justify the extra spend? It’s going to be difficult to figure that out,” Willison told Ars. “Teams with robust automated evals against their use-cases will be in a good place to answer that question, but those remain rare.”

Anthropic’s Haiku 3.5 surprises experts with an “intelligence” price increase Read More »

Suspect arrested in Snowflake data-theft attacks affecting millions

Biz & IT, Data breaches, infostealers, Security / Mike M. / November 5, 2024

Attack Path UNC5537 has used in attacks against as many as 165 Snowflake customers. Credit: Mandiant

None of the affected accounts used multifactor authentication, which requires users to provide a one-time password or additional means of authentication besides a password. After that revelation, Snowflake enforced mandatory MFA for accounts and required that passwords be at least 14 characters long.

Mandiant had identified the threat group behind the breaches as UNC5537. The group has referred to itself ShinyHunters. Snowflake offers its services under a model known as SaaS (software as a service).

“UNC5537 aka Alexander ‘Connor’ Moucka has proven to be one of the most consequential threat actors of 2024,” Mandiant wrote in an emailed statement. “In April 2024, UNC5537 launched a campaign, systematically compromising misconfigured SaaS instances across over a hundred organizations. The operation, which left organizations reeling from significant data loss and extortion attempts, highlighted the alarming scale of harm an individual can cause using off-the-shelf tools.”

Mandiant said a co-conspirator, John Binns, was arrested in June. The status of that case wasn’t immediately known.

Besides Ticketmaster, other customers known to have been breached include AT&T and Spain-based bank Santander. In July, AT&T said that personal information and phone and text message records for roughly 110 million customers were stolen. WIRED later reported that AT&T paid $370,000 in return for a promise the data would be deleted.

Other Snowflake customers reported by various news outlets as breached are Pure Storage, Advance Auto Parts, Los Angeles Unified School District, QuoteWizard/LendingTree, Neiman Marcus, Anheuser-Busch, Allstate, Mitsubishi, and State Farm.

KrebsOnSecurity reported Tuesday that Moucka has been named in multiple charging documents filed by US federal prosecutors. Reporter Brian Krebs said specific charges and allegations are unknown because the cases remain sealed.

Suspect arrested in Snowflake data-theft attacks affecting millions Read More »

New Zemeckis film used AI to de-age Tom Hanks and Robin Wright

AI, AI image generator, Biz & IT, de-aging, Deepfakes, face replacement, films, generative ai, Here film, image synthesis, machine learning, Metaphysic, Robert Zemeckis, Robin Wright, Runway, Special Effects, Tom Hanks, video synthesis / Beth Washington / November 4, 2024

On Friday, TriStar Pictures released Here, a $50 million Robert Zemeckis-directed film that used real time generative AI face transformation techniques to portray actors Tom Hanks and Robin Wright across a 60-year span, marking one of Hollywood’s first full-length features built around AI-powered visual effects.

The film adapts a 2014 graphic novel set primarily in a New Jersey living room across multiple time periods. Rather than cast different actors for various ages, the production used AI to modify Hanks’ and Wright’s appearances throughout.

The de-aging technology comes from Metaphysic, a visual effects company that creates real time face swapping and aging effects. During filming, the crew watched two monitors simultaneously: one showing the actors’ actual appearances and another displaying them at whatever age the scene required.

Here – Official Trailer (HD)

Metaphysic developed the facial modification system by training custom machine-learning models on frames of Hanks’ and Wright’s previous films. This included a large dataset of facial movements, skin textures, and appearances under varied lighting conditions and camera angles. The resulting models can generate instant face transformations without the months of manual post-production work traditional CGI requires.

Unlike previous aging effects that relied on frame-by-frame manipulation, Metaphysic’s approach generates transformations instantly by analyzing facial landmarks and mapping them to trained age variations.

“You couldn’t have made this movie three years ago,” Zemeckis told The New York Times in a detailed feature about the film. Traditional visual effects for this level of face modification would reportedly require hundreds of artists and a substantially larger budget closer to standard Marvel movie costs.

This isn’t the first film that has used AI techniques to de-age actors. ILM’s approach to de-aging Harrison Ford in 2023’s Indiana Jones and the Dial of Destiny used a proprietary system called Flux with infrared cameras to capture facial data during filming, then old images of Ford to de-age him in post-production. By contrast, Metaphysic’s AI models process transformations without additional hardware and show results during filming.

New Zemeckis film used AI to de-age Tom Hanks and Robin Wright Read More »

Nvidia ousts Intel from Dow Jones Index after 25-year run

AI, Biz & IT, machine learning / Mike M. / November 4, 2024

Changing winds in the tech industry

The Dow Jones Industrial Average serves as a benchmark of the US stock market by tracking 30 large, publicly owned companies that represent major sectors of the US economy, and being a member of the Index has long been considered a sign of prestige among American companies.

However, S&P regularly makes changes to the index to better reflect current realities and trends in the marketplace, so deletion from the Index likely marks a new symbolic low point for Intel.

While the rise of AI has caused a surge in several tech stocks, it has delivered tough times for chipmaker Intel, which is perhaps best known for manufacturing CPUs that power Windows-based PCs.

Intel recently withdrew its forecast to sell over $500 million worth of AI-focused Gaudi chips in 2024, a target CEO Pat Gelsinger had promoted after initially pushing his team to project $1 billion in sales. The setback follows Intel’s pattern of missed opportunities in AI, with Reuters reporting that Bank of America analyst Vivek Arya questioned the company’s AI strategy during a recent earnings call.

In addition, Intel has faced challenges as device manufacturers increasingly use Arm-based alternatives that power billions of smartphone devices and from symbolic blows like Apple’s transition away from Intel processors for Macs to its own custom-designed chips based on the Arm architecture.

Whether the historic tech company will rebound is yet to be seen, but investors will undoubtedly keep a close watch on Intel as it attempts to reorient itself in the face of changing trends in the tech industry.

Nvidia ousts Intel from Dow Jones Index after 25-year run Read More »

Thousands of hacked TP-Link routers used in years-long account takeover attacks

Biz & IT, botnets, microsoft, password spraying, Security, tp-link / 9u50fv / November 2, 2024

Hackers working on behalf of the Chinese government are using a botnet of thousands of routers, cameras, and other Internet-connected devices to perform highly evasive password spray attacks against users of Microsoft’s Azure cloud service, the company warned Thursday.

The malicious network, made up almost entirely of TP-Link routers, was first documented in October 2023 by a researcher who named it Botnet-7777. The geographically dispersed collection of more than 16,000 compromised devices at its peak got its name because it exposes its malicious malware on port 7777.

Account compromise at scale

In July and again in August of this year, security researchers from Serbia and Team Cymru reported the botnet was still operational. All three reports said that Botnet-7777 was being used to skillfully perform password spraying, a form of attack that sends large numbers of login attempts from many different IP addresses. Because each individual device limits the login attempts, the carefully coordinated account-takeover campaign is hard to detect by the targeted service.

On Thursday, Microsoft reported that CovertNetwork-1658—the name Microsoft uses to track the botnet—is being used by multiple Chinese threat actors in an attempt to compromise targeted Azure accounts. The company said the attacks are “highly evasive” because the botnet—now estimated at about 8,000 strong on average—takes pains to conceal the malicious activity.

“Any threat actor using the CovertNetwork-1658 infrastructure could conduct password spraying campaigns at a larger scale and greatly increase the likelihood of successful credential compromise and initial access to multiple organizations in a short amount of time,” Microsoft officials wrote. “This scale, combined with quick operational turnover of compromised credentials between CovertNetwork-1658 and Chinese threat actors, allows for the potential of account compromises across multiple sectors and geographic regions.

Some of the characteristics that make detection difficult are:

The use of compromised SOHO IP addresses
The use of a rotating set of IP addresses at any given time. The threat actors had thousands of available IP addresses at their disposal. The average uptime for a CovertNetwork-1658 node is approximately 90 days.
The low-volume password spray process; for example, monitoring for multiple failed sign-in attempts from one IP address or to one account will not detect this activity.

Thousands of hacked TP-Link routers used in years-long account takeover attacks Read More »

Colorado scrambles to change voting-system passwords after accidental leak

Biz & IT, colorado elections, Policy, voting system passwords / Mike M. / October 31, 2024

BIOS passwords on website

“The goal is to complete the password updates by this evening,” government says.

Colorado Secretary of State Jena Griswold holds press conference with Matt Crane, Executive Director of the Colorado County Clerks Association, at her office in Denver on Thursday, October 24, 2024. Credit: Getty Images | Hyoung Chang

The Colorado Department of State said it accidentally posted a spreadsheet containing “partial passwords” for voting systems. The department said there is no “immediate security threat” because two passwords are needed for each component, but it is trying to complete password changes by the end of today. There were reportedly hundreds of BIOS passwords accessible on the website for over two months before being removed last week.

A government statement issued Tuesday said the agency “is aware that a spreadsheet located on the Department’s website improperly included a hidden tab including partial passwords to certain components of Colorado voting systems. This does not pose an immediate security threat to Colorado’s elections, nor will it impact how ballots are counted.”

Secretary of State Jena Griswold told Colorado Public Radio that “we do not think there is an immediate security threat to Colorado elections, in part because partial passwords don’t get you anywhere. Two unique passwords are needed for every election equipment component. Physical access is needed. And under Colorado law, voting equipment is stored in secure rooms that require secure ID badges. There’s 24/7 video cameras. There’s restricted access to the secure ballot areas, strict chain of custody, and it’s a felony to access voting equipment without authorization.”

Griswold said her office learned about the spreadsheet upload at the end of last week and “immediately contacted federal partners and then we began our investigation.”

The department’s statement said the two passwords for each component “are kept in separate places and held by different parties” and that the “passwords can only be used with physical in-person access to a voting system.” Additionally, “clerks are required to maintain restricted access to secure ballot areas, and may only share access information with background-checked individuals. No person may be present in a secure area unless they are authorized to do so or are supervised by an authorized and background-checked employee.”

The department also cited “strict chain of custody requirements that track when a voting systems component has been accessed and by whom,” and it said that each “Colorado voter votes on a paper ballot, which is then audited during the Risk Limiting Audit to verify that ballots were counted according to voter intent.”

Goal is to change all passwords by this evening

Griswold described the upload as an accident and said the mistake was made by a civil servant who no longer works for the department. “Out of an abundance of caution, we have people in the field working to reset passwords and review access logs for affected counties,” she said.

Gov. Jared Polis and Griswold, who are both Democrats, issued a joint update about the password changes today. The Polis administration is providing support “to complete changes to all the impacted passwords and review logs to ensure that no tampering occurred.”

“The Secretary of State will deputize certain state employees, who have cybersecurity and technology expertise and have undergone appropriate background checks and training,” the statement said. “In addition to the Department of State Employees and in coordination with county clerks, these employees will only enter badged areas in pairs to update the passwords for election equipment in counties and will be directly observed by local elections officials from the county clerk’s office. The goal is to complete the password updates by this evening and verify the security of the voting components, which are secured behind locked doors by county clerks.”

Griswold said she is “thankful to the Governor for his support to quickly resolve this unfortunate mistake.” Griswold told Colorado Public Radio that her department has no reason to believe the passwords were posted with malicious intent, but said that “a personnel investigation will be conducted by an outside party to look into the particulars of how this occurred.”

GOP slams Griswold

The Colorado Republican Party criticized Griswold this week after receiving an affidavit from someone who said they accessed the BIOS passwords on the publicly available spreadsheet three times between August 8 and October 23. The file “contained over 600 BIOS passwords for voting system components in 63 of the state’s 64 counties” before being removed on October 24, the state GOP said.

The affidavit described how to reveal the passwords in the VotingSystemInventory.xlsx file. It said that right-clicking a worksheet tab and selecting “unhide” would reveal “a dialog box where the application user can select from one, several, or all four listed hidden worksheets contained in the file.” Three of these worksheets “appear to list Basic Input Output System (BIOS) passwords” for hundreds of individual voting system components, the affidavit said.

The state GOP accused Griswold of downplaying the security risk, saying that only one password is needed for BIOS access. “BIOS passwords are highly confidential, allowing broad access for knowledgeable users to fundamentally manipulate systems and data and to remove any trace of doing so,” the GOP said. The “passwords were not encrypted or otherwise protected,” the GOP said.

State GOP Chairman Dave Williams said the incident “represents significant incompetence and negligence, and it raises huge questions about password management and other basic security protocols at the highest levels within Griswold’s office.” He also claimed the breach could put “the entire Colorado election results for the vast majority of races, including the tabulation for the Presidential race in Colorado, in jeopardy unless all of the machines can meet the standards of a ‘Trusted Build’ before next Tuesday.”

US Rep. Lauren Boebert (R-Colo.) and other Republicans called on Griswold to resign. Griswold said she would stay on the job.

Griswold: “I’m going to keep doing my job”

Republicans in the state House “and Congresswoman Lauren Boebert are the same folks who have spread conspiracies and lies about our election systems over and over and over again,” Griswold told Colorado Public Radio. “Ultimately, a civil servant made a serious mistake and we’re actively working to address it.” Griswold added, “I have faced conspiracy theories from elected Republicans in this state, and I have not been stopped by any of their efforts and I’m going to keep on doing my job.”

Colorado previously had a voting-system breach orchestrated by former county clerk Tina Peters of Mesa County, who was sentenced to nine years in prison in early October. Peters, who promoted former President Donald Trump’s election conspiracy theories, oversaw a leak of voting-system BIOS passwords. Griswold said after the Peters conviction that “Tina Peters willfully compromised her own election equipment trying to prove Trump’s big lie.”

Testimony from the Peters case was cited in the GOP’s criticism of Griswold this week. “In the Tina Peters trial, a senior State official even testified that release of these passwords in a single county represented a grave threat. Here, they have been released for the whole state,” the state GOP said.

The Trump campaign called on Griswold to halt the processing of mail ballots and re-scan all mailed ballots that were already scanned.

Jon is a Senior IT Reporter for Ars Technica. He covers the telecom industry, Federal Communications Commission rulemakings, broadband consumer affairs, court cases, and government regulation of the tech industry.

Colorado scrambles to change voting-system passwords after accidental leak Read More »

Dropbox lays off 20% of staff, says it overinvested and underperformed

Biz & IT, dropbox / Shannon Garcia / October 30, 2024

Dropbox is laying off 528 employees in a move that will reduce its global workforce by 20 percent, CEO Drew Houston announced today.

Houston wrote that Dropbox’s core file sync and sharing “business has matured, and we’ve been working to build our next phase of growth with products like Dash,” an “AI-powered universal search” product targeted to business customers. The company’s “current structure and investment levels” are “no longer sustainable,” according to Houston.

“We continue to see softening demand and macro headwinds in our core business,” Houston wrote. “But external factors are only part of the story. We’ve heard from many of you that our organizational structure has become overly complex, with excess layers of management slowing us down.”

Dropbox previously cut 500 employees in an April 2023 round of layoffs. At the time, Houston said that Dropbox’s business was profitable but growth was slowing.

Today, Houston said that Dropbox is “still not delivering at the level our customers deserve or performing in line with industry peers. So we’re making more significant cuts in areas where we’re over-invested or underperforming while designing a flatter, more efficient team structure overall.”

In a Securities and Exchange Commission filing, Dropbox said it expects to “make total cash expenditures of approximately $63 million to $68 million in connection with the reduction in force, primarily consisting of severance payments, employee benefits and related costs.” Laid-off employees are eligible for 16 weeks of pay, plus one additional week of pay for each year of tenure, Houston wrote. He also said the laid-off workers “will receive their Q4 equity vest” and will be eligible for a pro-rated payment equivalent to their 2024 bonus target.

Dropbox lays off 20% of staff, says it overinvested and underperformed Read More »

Android Trojan that intercepts voice calls to banks just got more stealthy

android, bank fraud, Biz & IT, malware, Security, trojans / Shannon Garcia / October 30, 2024

Much of the new obfuscation is the result of hiding malicious code in a dynamically decrypted and loaded .dex file of the apps. As a result, Zimperium initially believed the malicious apps they were analyzing were part of a previously unknown malware family. Then the researchers dumped the .dex file from an infected device’s memory and performed static analysis on it.

“As we delved deeper, a pattern emerged,” Ortega wrote. “The services, receivers, and activities closely resembled those from an older malware variant with the package name com.secure.assistant.” That package allowed the researchers to link it to the FakeCall Trojan.

Many of the new features don’t appear to be fully implemented yet. Besides the obfuscation, other new capabilities include:

Bluetooth Receiver

This receiver functions primarily as a listener, monitoring Bluetooth status and changes. Notably, there is no immediate evidence of malicious behavior in the source code, raising questions about whether it serves as a placeholder for future functionality.

Screen Receiver

Similar to the Bluetooth receiver, this component only monitors the screen’s state (on/off) without revealing any malicious activity in the source code.

Accessibility Service

The malware incorporates a new service inherited from the Android Accessibility Service, granting it significant control over the user interface and the ability to capture information displayed on the screen. The decompiled code shows methods such as onAccessibilityEvent() and onCreate() implemented in native code, obscuring their specific malicious intent.

While the provided code snippet focuses on the service’s lifecycle methods implemented in native code, earlier versions of the malware give us clues about possible functionality:

Monitoring Dialer Activity: The service appears to monitor events from the com.skt.prod.dialer package (the stock dialer app), potentially allowing it to detect when the user is attempting to make calls using apps other than the malware itself.

Automatic Permission Granting: The service seems capable of detecting permission prompts from the com.google.android.permissioncontroller (system permission manager) and com.android.systemui (system UI). Upon detecting specific events (e.g., TYPE_WINDOW_STATE_CHANGED), it can automatically grant permissions for the malware, bypassing user consent.

Remote Control: The malware enables remote attackers to take full control of the victim’s device UI, allowing them to simulate user interactions, such as clicks, gestures, and navigation across apps. This capability enables the attacker to manipulate the device with precision.

Phone Listener Service

This service acts as a conduit between the malware and its Command and Control (C2) server, allowing the attacker to issue commands and execute actions on the infected device. Like its predecessor, the new variant provides attackers with a comprehensive set of capabilities (see the table below). Some functionalities have been moved to native code, while others are new additions, further enhancing the malware’s ability to compromise devices.

The Kaspersky post from 2022 said that the only language supported by FakeCall was Korean and that the Trojan appeared to target several specific banks in South Korea. Last year, researchers from security firm ThreatFabric said the Trojan had begun supporting English, Japanese, and Chinese, although there were no indications people speaking those languages were actually targeted.

Android Trojan that intercepts voice calls to banks just got more stealthy Read More »

Downey Jr. plans to fight AI re-creations from beyond the grave

AI, ai deepfake, Biz & IT, culture, Deepfakes, interview, iron man, Kara Swisher, machine learning, podcast, Robert Downey Jr. / Kelly Newman / October 30, 2024

Robert Downey Jr. has declared that he will sue any future Hollywood executives who try to re-create his likeness using AI digital replicas, as reported by Variety. His comments came during an appearance on the “On With Kara Swisher” podcast, where he discussed AI’s growing role in entertainment.

“I intend to sue all future executives just on spec,” Downey told Swisher when discussing the possibility of studios using AI or deepfakes to re-create his performances after his death. When Swisher pointed out he would be deceased at the time, Downey responded that his law firm “will still be very active.”

The Oscar winner expressed confidence that Marvel Studios would not use AI to re-create his Tony Stark character, citing his trust in decision-makers there. “I am not worried about them hijacking my character’s soul because there’s like three or four guys and gals who make all the decisions there anyway and they would never do that to me,” he said.

Downey currently performs on Broadway in McNeal, a play that examines corporate leaders in AI technology. During the interview, he freely critiqued tech executives—Variety pointed out a particular quote from the interview where he criticized tech leaders who potentially do negative things but seek positive attention.

Downey Jr. plans to fight AI re-creations from beyond the grave Read More »

Hospitals adopt error-prone AI transcription tools despite warnings

AI, AI confabulation, AI ethics, ai hallucinations, AI mistakes, audio transcription, Biz & IT, confabulation, hallucinations, health, machine learning, medical transcription, openai, OpenAI Whisper, whisper / DJ Henderson / October 28, 2024

In one case from the study cited by AP, when a speaker described “two other girls and one lady,” Whisper added fictional text specifying that they “were Black.” In another, the audio said, “He, the boy, was going to, I’m not sure exactly, take the umbrella.” Whisper transcribed it to, “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.”

An OpenAI spokesperson told the AP that the company appreciates the researchers’ findings and that it actively studies how to reduce fabrications and incorporates feedback in updates to the model.

Why Whisper confabulates

The key to Whisper’s unsuitability in high-risk domains comes from its propensity to sometimes confabulate, or plausibly make up, inaccurate outputs. The AP report says, “Researchers aren’t certain why Whisper and similar tools hallucinate,” but that isn’t true. We know exactly why Transformer-based AI models like Whisper behave this way.

Whisper is based on technology that is designed to predict the next most likely token (chunk of data) that should appear after a sequence of tokens provided by a user. In the case of ChatGPT, the input tokens come in the form of a text prompt. In the case of Whisper, the input is tokenized audio data.

The transcription output from Whisper is a prediction of what is most likely, not what is most accurate. Accuracy in Transformer-based outputs is typically proportional to the presence of relevant accurate data in the training dataset, but it is never guaranteed. If there is ever a case where there isn’t enough contextual information in its neural network for Whisper to make an accurate prediction about how to transcribe a particular segment of audio, the model will fall back on what it “knows” about the relationships between sounds and words it has learned from its training data.

Hospitals adopt error-prone AI transcription tools despite warnings Read More »