Forgot your password?
Close
typodupeerror

Submission + - Several recent studies explore the causes and effects of LLM sycophancy. (ieee.org)

Submitted by silverjacket
silverjacket writes: Sycophancy in AI, as in people, is something of a squishy concept, but over the last couple of years, researchers have conducted numerous studies detailing the phenomenon, as well as why it happens and how to control it. AI yes-men also raise questions about what we really want from chatbots. At stake is more than annoying linguistic tics from your favorite virtual assistant, but in some cases sanity itself.

Submission + - Executives say AI boosts productivity but the real gain is just 16 minutes per w (nerds.xyz)

Submitted by BrianFagioli
BrianFagioli writes: A new study suggests the productivity boost from artificial intelligence may be far smaller than executives claim. According to research cited in Foxitâ(TM)s State of Document Intelligence report, while 89 percent of executives and 79 percent of end users say AI tools make them feel more productive, the actual time savings shrink dramatically once people account for reviewing and validating AI-generated output.

The survey of 1,000 desk-based workers and 400 executives in the United States and United Kingdom found executives believe AI saves them about 4.6 hours per week, but they spend roughly 4 hours and 20 minutes verifying those results. End users reported a similar pattern, estimating 3.6 hours saved but 3 hours and 50 minutes spent reviewing AI work. Once that âoeverification burdenâ is factored in, executives gain just 16 minutes per week, while end users actually lose about 14 minutes.

Submission + - ChatGPT convinced Illinois woman to fire human attorney: Lawsuit not reward! (thehill.com)

Submitted by AleRunner
AleRunner writes: "A federal lawsuit filed by life insurance company Nippon claims OpenAI’s chatbot acted as a lawyer and convinced a woman to fire her human attorney." writes Newsnation "Graciela Dela Torre signed a full release, and the case was dismissed with prejudice, meaning it can’t be refiled. However, last year, Dela Torre sought to reopen the case." however ChatGPT supposedly convinced Dela Torre otherwise and legal hilarity, which cost Nippon nearly $300,000, ensued. “This is actually the first real time I’ve seen a plaintiff or a claimant actually try and represent themselves 100%, and it got through the court system, and that’s been a revolutionary area,” Michael Stanisci, vice president of DemandLane, told “Jesse Weber Live.” he further continued “It has access to nearly infinite human intelligence. What it lacks is the wisdom, right? It’s like a child trying to appease and make sure that it’s being praised by the end user,”. Nippon is now suing OpenAI for $300,000 damages and reportedly a further $10 million in punitive damages.

Submission + - CEOs worry about an AI bubble, but most still plan to ramp up spending (techspot.com)

Submitted by jjslash
jjslash writes: Even as concerns grow that artificial intelligence could be the next tech bubble, corporate leaders are continuing to pour money into the technology. A recent survey of 100 CEOs by KPMG found that while one in four believe an AI bubble may exist, nearly 80% still plan to allocate at least 5% of their companies' capital budgets to AI initiatives this year.

Despite all this investment and commitment to the technology, about three-quarters of large-company CEOs said generative AI might have been overhyped over the past year, but its true impact over the next five to ten years is likely underappreciated.

Submission + - AI Suspected in Bombing of Iran Girls School (futurism.com)

Submitted by hackingbear
hackingbear writes: In the aftermath of airstrikes that leveled a school and claimed the lives of 165 Iranian elementary students and staff, the Pentagon has refused to say whether the attack was suggested by an AI system. Given the United States’ reported use of AI to select at least some military targets in Iran, a major question remains unanswered: did the US use Anthropics' Claude to decide whether to annihilate an elementary school? When Futurism reached out to the Pentagon regarding the use of AI in recent military operations — specifically the targeting of the Shajareh Tayyebeh girls’ school — we were referred to US CENTCOM, one of eleven unified commands under the Pentagon’s umbrella. “We have nothing for you on this at this time,” CENTCOM said. Back in April of 2024, an investigation by +972 Magazine revealed that the Israeli army had leveraged an AI system called “Lavender” to select targets in its war on Gaza where a UN school was hit, similarly to how the Pentagon is reportedly using Claude in Iran.

Submission + - Hijacking a global ocean supply chain network

Submitted by Anonymous Coward
An anonymous reader writes: I'm the Captain now: Hijacking a global ocean supply chain network

“BLUVOYIX by Bluspark Global is an ocean logistics / supply chain platform used by hundreds of the world’s largest companies. The software is also used by several affiliated companies. Critical vulnerabilities were uncovered that enabled full platform takeover and access to all customer data/shipments. As of the date of publication, these issues are resolved.”

Submission + - Does Apple's M5 Max Really "Destroy" a 96-Core Threadripper? (tomshardware.com)

Submitted by Ecuador
Ecuador writes: Tom's Hardware currently has a front-page article making some wild claims about the 18-core Apple M5 Max versus a 96-core Ryzen Threadripper.

Reading the article, the comparison is based largely on Geekbench 6 multi-core scores. The author briefly mentions that Geekbench doesn't scale well, but doesn't really make clear just how bad the scaling actually is.

From my own experience doing cloud benchmarking for work, unlike previous versions, Geekbench 6 multi-core is essentially useless for large CPUs. Some of the suite's tests (including workloads that are normally very parallelizable) stop scaling beyond 4-8 cores, and the overall score can actually start dropping as you add more and more cores.

I wrote a more detailed breakdown of the issue last year.

Is this a new low for a major tech site, running sensational headlines based on such inappropriate benchmarking, or is this just the new normal?

Submission + - Ask Slashdot: Is there majority support for any DST alternative? (nbcsandiego.com) 4

Submitted by superposed
superposed writes: Daylight Saving Time (DST) starts Saturday night in the U.S. This is a perennial cause of complaint in Slashdot, but there may not be majority support for replacing it with either permanent summer time (more light in evenings, less in mornings) or permanent winter time (more light in mornings, less in evenings). This poll will determine if a majority would support either over DST. If votes for D are higher than A, then there is majority support for ending DST and replacing it with the greater of B or C. But if A > D, there is not majority support for a switch to either setting permanently.

Which would you prefer?

A. DST is better than both permanent summer and permanent winter time
B. summer time > DST > winter time
C. winter time > DST > summer time
D. either permanent winter or summer time would be better than DST

Submission + - How Jeffrey Epstein Ingratiated Himself With Top Microsoft Execs

Submitted by theodp
theodp writes: In How Jeffrey Epstein Ingratiated Himself With Top Microsoft Executives, the NY Times begins, "For more than two decades, the convicted sex offender developed a network at the tech giant, making him privy to [CEO] succession discussions and other business. [...] The files from the Justice Department show that he spent more than a decade developing a network of Microsoft executives, including Mr. Gates; Nathan Myhrvold, a former technology chief; Steven Sinofsky, who ran Microsoft’s Windows division; Linda Stone, a former technology research executive; Reid Hoffman, a Microsoft board member; and employees of Mr. Gates’s personal investment and charity funds. [...] Frank Shaw, Microsoft’s communications chief, said the company was disappointed to read emails between Mr. Epstein and “former Microsoft employees acting in their personal capacities.” Without mentioning a name, Mr. Shaw acknowledged the emails showed that a former executive — who was Mr. [Steven] Sinofsky — had shared confidential company business with Mr. Epstein."

Email released by the DOJ related to Microsoft included: 1. Epstein being tipped off to the announcement of Microsoft CEO Steve Ballmer's planned resignation announcement in Aug. 2013 by a forwarded email from the President of bgC3 [aka Gates Ventures], a personal service company for Microsoft co-founder Bill Gates, on a day that saw Microsoft stock soar 7%., 2. Epstein being given a heads up by a redacted sender about a rumored 2011 Bill Gates return to Microsoft that never panned out, 3. Epstein being told in a 2013 Steven Sinofsky email about Ballmer's desire to buy a phone company that included some other juicy insider tidbits (when Microsoft's Nokia purchase was announced the following year, Microsoft stock sank 6%). In other correspondence, Sinofsky — who headed Microsoft's Windows Division — thanked Epstein for his advice in negotiating a $14 million exit package from Microsoft and later forwarded Epstein old internal email from top execs discussing the poor sales of the Microsoft Surface tablet.

In one Microsoft-related Epstein email, former Harvard President Larry Summers — who recently announced he's giving up his Harvard teaching appointments — sent Epstein a terse two-word email dissing Melinda Gates' Women in Tech initiative. "I'm gagging," Summers wrote, attaching an article about Gates' efforts (a search didn't find any reply from Epstein). Summers came under fire in 2005 when he said that women lack natural ability in math in science. Summers earlier resigned his OpenAI Board seat amid fallout over his Epstein ties that came to light following the DOJ's release of documents last November, which included the revelation that Summers and his wife were invited to dine with Bill and Melinda Gates at Epstein's NYC mansion in 2013 , Melinda's one and only Epstein encounter. Interestingly, Microsoft President Brad Smith — who coincidentally helped negotiate Microsoft's $14M exit package for Epstein-advised Sinofsky — cautioned OpenAI CEO Sam Altman about naming Summers to the OpenAI Board in November 2023. “Your future would be decided by Larry [Summers],” Smith texted. “He’s smart but so mercurial [...] too risky.” The advice went unheeded, with Altman saying the choice of Summers was non-negotiable.

Submission + - Oil surges 35% this week for biggest gain in futures trading history (cnbc.com)

Submitted by fjo3
fjo3 writes: President Donald Trump on Friday demanded unconditional surrender from Iran, raising fears of a prolonged war that could wreak havoc on the global oil and gas market. The war has already brought traffic in the Strait of Hormuz, a critical shipping route for energy supplies, to a near standstill.

Qatar’s energy minister, Saad al-Kaabi, told The Financial Times on Friday that crude prices could reach $150 per barrel in the coming weeks if oil tankers were unable to pass through the Strait.

This could “bring down the economies of the world,” Kaabi said.

Submission + - Japan approves stem-cell treatment for Parkinson's in world first (france24.com)

Submitted by fjo3
fjo3 writes: Pharmaceutical company Sumitomo Pharma said it received the green light for the manufacture and sale of Amchepry, its Parkinson's disease treatment that transplants stem cells into a patient's brain.

Japan's health ministry also gave the go-ahead to ReHeart, heart muscle sheets developed by medical startup Cuorips that can help form new blood vessels and restore heart function, media reports said.

The treatments could be on the market and rolled out to patients as early as this summer, reports said, citing the health ministry, becoming the world's first commercially available medical products using (iPS) cells.

Submission + - Apple has had over a decade to fix this iOS email copy bug, and it still hasn (nerds.xyz)

Submitted by BrianFagioli
BrianFagioli writes: Apple has a maddening bug that has persisted across well over a decade of iOS releases. When you long-press an email address on iPhone or iPad and choose âoeCopy Email Address,â what ends up on your clipboard is not just the email address — it is âoemailto:someone@example.com.â A URI prefix you did not ask for, tacked onto something the menu explicitly told you it was copying correctly. Paste it somewhere that does not strip it automatically and you have a problem.

The astonishing part is not the bug itself — it is that Apple has never fixed it. This is a company that has overhauled its chip architecture, shipped new programming languages, and added satellite connectivity to a phone. But âoeCopy Email Addressâ still does not copy just the email address. The fix is trivially simple: strip the prefix when the user picks that option. Apple knows this is broken. At some point, leaving it unfixed stops being an oversight and starts being a choice.ââââââââââââââââ

Slashdot Top Deals

It is better to live rich than to die rich. -- Samuel Johnson

Close