Submission + - Think You've Eliminated Chinese AI? Check the Model's Lineage, Cisco Says (securityweek.com)
The findings suggest that a US-labeled model could inherit behaviors, vulnerabilities, or biases from a Chinese base model. Cisco is calling for a model bill of materials documenting upstream dependencies.
As the researchers put it: “Models do not have passports. They have supply chains.”
Submission + - New York City 911 system missed 1700 calls due to software update
Submission + - Q-CTRL Achieves First GPS-Free Quantum Gravimetric Maritime Navigation Demo (q-ctrl.com)
Comment Re:Laying the groundwork (Score 5, Interesting) 124
For some definition of 'they', sure, there's always *someone* paranoid about the scenario.
However there's a fair amount more credible supposition over this one. Trump's sore loser reaction has paved the way for some dangerous moves.
In 2020, he only went to interfere *after* the fact, but he clearly interfered. When the results betrayed him, he sought to overturn them and when that failed he tried to ignore the electors and use their own anyway and as that failed he tried to get Pence to reject the election and as that failed he incited a mom to try to disrupt the proceedings. He learned that in a fair game he actually could lose and that was unacceptable and he was willing to do whatever it takes to replace a loss with a win.
Now he and opportunistic cohorts are already much more proactive about it. Every reason to anticipate any plausible excuse to rig or discard the elections as he has proven himself already willing to do so and is surrounded by a much more all-in administration than his first term. Already trying to mess with mail-in, already openly rigged the representative maps, already been working ICE and national guard troops to get the hang of domestic military operations. Stephen Miller once basically publicly fantasized about sending republican controlled national guard into democratic states against their will to make them bend to his will.
I do anticipate the election will be *held* and if and only if the results are a loss will they try anything like an 'emergency', but I absolutely expect ICE and national guards to be doing operations around strategic polling sites. Ostensibly about immigration or crime, but with the oh so coincidental side effect of voter suppression.
Comment Re:Better to fire all the "lieutenants" who agreed (Score 1) 55
I get the feeling that Zuck is in a similar boat to Richard Dawkins and Kent Overstreet, who have gone off the deep end and believe they have LLMs which are effectively equivalent to human companionship.
Remember Zuck celebrated that people could have AI "friends" and employees could talk to AI impersonations of their management and it would be just as good as actually getting to talk to management.
Nothing quite as publicly really off the deep end as Overstreet and Dawkins (who both seem to have decided that their favored LLMs must be 'girls', which strongly suggests they needed to gender it to make their "crush" make sense for them), with Overstreet getting extra creepy points for insisting his LLM is essentially a teenage girl. I would not be surprised if he at least more privately holds similar feelings and thus doesn't view this as "apocalyptic" bad.
Submission + - Microsoft Admits AI Training in New User Privacy Statment (microsoft.com) 1
"Not all personal data processed by Microsoft can be accessed or controlled via the tools described in this section."
Submission + - The First Standardized Test For AI Tampering Just Failed Every Model (studyfinds.com)
Submission + - Hewlett Packard Enterprise retroactively cuts Aruba lifetime warranty to 5 years (hpe.com)
Submission + - China's Moonshot in talks with Microsoft, Amazon, Google over K3 revenue sharing (reuters.com)
Any deal could mark the first big revenue-sharing pact between a Chinese AI firm and a major U.S. cloud company.
The discussions highlight how China's leading AI models, often far cheaper than Western offerings, are gaining traction in the U.S., despite national security concerns in Washington that have led to bans on exports of AI chips to China. They are also taking place despite critical comments about Moonshot from senior U.S. officials.
IPO-bound Moonshot is seeking up to a 30% share of revenue generated from K3-related services on Microsoft's Azure, Amazon Web Services and Google Cloud, according to the sources who declined to be identified because the discussions are private.
That would be in line with terms that sources have said the startup has outlined for major customers using the open-weight model.
Moonshot has come under fire from U.S. Treasury Secretary Scott Bessent who said last month that he might add it to a trade blacklist. U.S. officials have accused the Beijing-based company of stealing from Anthropic's most sophisticated model, Fable, to help create Kimi K3 and illegally acquiring Nvidia chips.
Founded in 2023 by Carnegie Mellon-trained AI researcher Yang Zhilin, Moonshot is backed by several Chinese tech giants including cloud computing giant Alibaba.
Submission + - Microsoft deepens Saudi Arabia AI ties despite longstanding privacy concerns (nerds.xyz)
The partnership could bring HUMAINs Arabic-language ALLAM models into Microsoft Foundry and Microsoft 365 Copilot. HUMAIN engineers will also work alongside Microsoft engineers to help organizations deploy AI systems.
The relationship deserves scrutiny. Human rights groups have previously raised concerns about Microsofts Saudi expansion, including privacy, surveillance, and government access to data.
Microsoft says its technology can meet enterprise and public-sector security and governance requirements. But as the company gets closer to a Saudi state-backed AI operation, questions remain about what happens when those requirements collide with Microsofts privacy and human rights commitments.
Submission + - It's now easy to know what Flock/ALPR plans your US locality has (tennesseedave.com)
Submission + - Lab Supply Companies Have Been Selling Antibodies Using Manipulated Images (arstechnica.com)
[...] The consequences of this discovery are pretty significant. If you buy an antibody believing it will generate clean, clear data and find it doesn’t, many researchers will assume the problem is them. That often leads them to spend time troubleshooting the procedure and trying slightly different conditions to get useful data. If the antibody can’t produce those results, all the troubleshooting will have been a waste of time and money. Nature’s coverage of the new findings includes quotes from several companies that sell these antibodies that, paraphrased, range from “we’ll look into it” to “we don’t think it’s a problem.” Researchers who have spent time frustrated while failing to get an antibody-based experiment to work may think otherwise.
Comment "leverages" (Score 1) 83
"leverages user audio systems"....what an interesting way to spell "spying".
Submission + - Citrix adds a Linux powered escape hatch for broken Windows PCs (nerds.xyz)
The recovery environment sits on its own partition and remains isolated from the Windows filesystem. It does not repair Windows, and it cannot help with failed hardware, missing network access, or unavailable Citrix infrastructure. However, it could give enterprises a faster way to keep employees working while IT investigates or repairs affected Windows installations.