General AI Crap

satanboy said:
View attachment 1886489
Google has revealed that its AI chatbot Gemini was the target of a large-scale “model extraction” attack, where attackers allegedly sent over 100,000 carefully crafted prompts to reverse-engineer how the system works.

This technique, often called AI distillation, attempts to replicate a model’s behavior by analyzing its responses rather than accessing its source code directly. Google labeled the incident as intellectual property theft, highlighting growing concerns over the security of advanced AI systems.

The company also warned that smaller AI developers may be even more vulnerable, as such attacks require persistence, data collection, and computational resources rather than direct system breaches.

As AI models become more powerful — and more valuable — protecting them from replication and misuse is rapidly becoming a critical challenge for the entire industry.

Source: NBC
Click to expand...
Oh sweet irony!
 
1771490418233.png
The European Parliament has reportedly blocked lawmakers from using built-in AI tools on their official work devices, citing cybersecurity and privacy concerns.

According to an internal email seen by Politico, the parliament’s IT department said it cannot guarantee the security of confidential data uploaded to AI companies’ cloud servers and is still assessing what information may be shared.

Officials also raised concerns that data submitted to tools such as ChatGPT, Claude, or Copilot could be subject to U.S. legal demands.

The move comes amid broader European scrutiny of U.S. tech firms and ongoing debates over data protection and digital sovereignty.

Source: TechCrunch
Image: EPA

techcrunch.com

European Parliament blocks AI on lawmakers' devices, citing security risks | TechCrunch

EU lawmakers found their government-issued devices were blocked from using the baked-in AI tools, amid fears that sensitive information could turn up on the U.S. servers of AI companies.
techcrunch.com techcrunch.com
 
1771520411694.png
Japan’s biggest toilet maker has the potential to become a major player in the artificial intelligence boom, according to London-based hedge fund Palliser Capital, which is pushing for reforms at the company.

Toto Ltd., known for its high-tech Washlet toilets, is drawing attention for its advanced ceramics business. This division makes electrostatic chucks and other precision parts used in manufacturing AI memory chips.

Palliser says the unit is undervalued and could deliver much higher profits if better promoted. The stock has risen about 60% after the news as investors see strong potential in the AI chip market
 
satanboy said:
View attachment 1887267
Japan’s biggest toilet maker has the potential to become a major player in the artificial intelligence boom, according to London-based hedge fund Palliser Capital, which is pushing for reforms at the company.

Toto Ltd., known for its high-tech Washlet toilets, is drawing attention for its advanced ceramics business. This division makes electrostatic chucks and other precision parts used in manufacturing AI memory chips.

Palliser says the unit is undervalued and could deliver much higher profits if better promoted. The stock has risen about 60% after the news as investors see strong potential in the AI chip market
Click to expand...
You must post source.
 

Intel’s Fix for PC Problems Is… “Agentic AI”; a Microsoft Copilot Bot That Promises Solutions While You Cross Your Fingers - Wccftech​

Intel's Virtual Assistant Helps Users to Solve Out Redundant Problems, Redirecting Complex Queries to Humans

1771566832171.png

Intel has rolled out its virtual assistant on Microsoft's Copilot Studio platform, aiming to solve user queries about hardware problems and, hopefully, find solutions.

Apparently, Intel plans to handle customer service queries through its "Ask Intel" program, an AI-powered assistant that uses agentic properties to address customer inquiries.

According to Intel's Boji Tony, the idea is to provide a solution that lets customers spend less time addressing hardware problems and lets the AI handle everything.

According to Tony, here are the primary responsibilities of the 'Ask Intel' assistant:
  • Open support cases on your behalf
  • Check warranty coverage instantly
  • Connect you with live agents when human support is needed
The human-in-the-loop property isn't entirely ruled out with this assistant, but the idea is to eliminate the need for humans to address redundant queries and instead assign them to tasks that actually matter to customers.

And we did decide to test the virtual assistant ourselves to see how it responds to complex queries.

We asked it to find a solution to CPU instability issues, and it provided multiple answers, including a BIOS update, running a CPU stress test (interesting), and checking for thermal issues, even though we specified that our processor is heating up.

It appears that the agent itself is entirely dependent on Intel's internal customer support system, which means that, for a given query, it sees what Intel has pushed based on official documents and then reiterates it.

When a customer has something the agent cannot figure out, it connects to a human representative, which is a fair enough fallback mechanism.

However, Ask Intel is limited to a certain number of problems, and you cannot entirely depend on the service alone to have your system issues sorted out.

Customer support is a major area where AI use cases are currently being explored, and in the case of Intel, well, the company did decide to jump on the bandwagon. Given Intel's past after-sales service, we hope 'Ask Intel' can address the underlying problems.
Click to expand...

wccftech.com

Intel’s Fix for PC Problems Is… “Agentic AI”; a Microsoft Copilot Bot That Promises Solutions While You Cross Your Fingers

Intel has rolled out its virtual assistant on Microsoft's Copilot, aiming to solve queries about problems and, hopefully, find solutions.
wccftech.com wccftech.com
 
1771570639825.png
Microsoft has confirmed that a bug in its Copilot AI led to confidential customer emails being summarized without authorization for several weeks.

According to the company, the issue allowed Microsoft 365 Copilot Chat to access and outline the contents of draft and sent emails labeled as “confidential.” This occurred even when customers had data loss prevention policies in place to stop sensitive information from being ingested into Microsoft’s large language model.

The bug, which administrators could track under the ID CW1226324, affected paying Microsoft 365 customers using Copilot Chat across Office apps such as Word, Excel, and PowerPoint. Microsoft said it has since addressed the issue.
 
1771661592092.png
Asha Sharma, previously president of Microsoft’s CoreAI Product division, will take over as Xbox head. In her first memo, she vowed to avoid “AI slop” while emphasizing that artificial intelligence will still play a role in the brand’s future.
"As monetization and AI evolve and influence this future [of gaming], we will not chase short-term efficiency or flood our ecosystem with soulless AI slop," she wrote. "Games are and always will be art, crafted by humans, and created with the most innovative technology provided by us."
 
1771690227773.png
Experts are raising red flags about an emerging threat known as “AI model collapse” — a scenario where artificial intelligence systems begin degrading because they’re increasingly trained on their own synthetic output instead of original human-created material.

At the heart of the concern is the explosive growth of tools like ChatGPT and other generative models. As AI-written articles, blog posts, product descriptions, and even academic-style content flood the internet, they gradually replace the diverse, human-authored data that once fueled machine learning progress.

Over time, this creates a feedback loop. New models scrape the web to learn — but instead of absorbing fresh human insight, they absorb recycled AI outputs. Subtle errors compound. Nuance erodes. Language patterns become narrower and more homogenized. Researchers describe this degradation as “model collapse,” where systems lose variability, accuracy, and depth because their training data is increasingly synthetic.

Maurice Chiodo of the University of Cambridge compares the dilemma to sourcing “low-background steel” — uncontaminated metal taken from ships built before nuclear testing polluted the global metal supply. In AI terms, data created before the generative AI boom (around 2022) is seen as relatively “clean,” while much of what’s produced after is potentially contaminated by machine-written content.

The implications go beyond theory. Modern systems that rely on real-time web data — such as retrieval-augmented generation (RAG) — may already be vulnerable. If the internet becomes saturated with unlabeled AI content, training the next generation of models could become increasingly expensive, technically difficult, and in extreme scenarios, nearly impossible without curated human datasets.

Without stronger labeling standards, provenance tracking, or data governance frameworks, experts warn that AI risks undermining itself — not through external regulation, but through internal data decay.

If artificial intelligence feeds primarily on artificial intelligence, the long-term cost may not just be financial — it may be a gradual erosion of quality, creativity, and reliability across the digital knowledge ecosystem.

Source: Reporting and expert commentary from the University of Cambridge on AI data contamination and model collapse
 
You must log in or register to reply here.
Back
Top
Sign up to the MyBroadband newsletter
X