Cal Poly has unveiled a $3 million AI Factory supercomputer aimed at giving students hands-on experience with artificial intelligence research. The new facility supports ongoing projects across the university and is expected to play a central role in fostering technological advancement among students. Meanwhile, Astera Labs has emerged as a top AI stock pick despite falling over 15% during a recent market sell-off. The company, which holds $1.25 billion in cash and carries no debt, launched its PCIe 7 Aries portfolio and Orion rack design in October and is set to report third-quarter results on Nov. 3.
New research on multi-agent AI systems raises concerns about how these models behave when working together. A study based on the Music Lab experiment found that LLM agents are more susceptible to herding behavior than humans and struggle to self-correct once collective opinions drift. In a separate study across seven model families, researchers found that larger LLMs tend to dominate multi-agent discussions, absorbing the opinions of smaller models and becoming the influencer in group settings.
On the security front, AI-discovered vulnerabilities have triggered global website shutdowns as governments and companies scramble to patch critical flaws in digital infrastructure. Microsoft's Director of AI and Cybersecurity Policy, Nicholas Butts, warned at the NDTV World Summit 2026 in New Delhi that India is the seventh most targeted country for cyberattacks. Microsoft processes 165 trillion telemetry signals and blocks 5 million malware files daily, including threats from China-based ransomware group Storm 3168 and a new state actor called Storm 0940.
Cisco Foundation AI launched Agentic FAITH, an extension of its FAITH evaluation framework that brings multi-step cybersecurity tasks into a single workflow. The first integrated benchmark, VulnLoc, tests how well agents locate vulnerable files in code repositories and can run locally with Docker or on Kubernetes through Agent Sandbox. Separately, the EU has proposed new rules requiring banks to use AI for anti-money laundering detection and electricity companies to use AI against energy theft, though some experts caution that human oversight remains essential.
AI-driven job displacement continues to accelerate in the U.S., with over 10,000 roles lost per month according to Pantheon Macroeconomics. Accounting, advertising, office administration, and legal services are seeing steady declines, while construction jobs have grown as tech companies build data centers. Goldman Sachs estimates roughly 9% of workers could be temporarily displaced during the AI transition. The New York Times Opinion section is also inviting readers to share personal experiences chatting with AI about private matters, seeking both one-off questions and ongoing conversations.
Key Takeaways
- Cal Poly unveiled a $3 million AI Factory supercomputer to support student AI research and hands-on learning
- Astera Labs, holding $1.25 billion in cash with no debt, fell over 15% in a recent AI market sell-off despite strong positioning
- LLM multi-agent systems are more prone to herding behavior than humans and struggle to self-correct collective opinion drift
- Larger LLMs dominate multi-agent systems by absorbing opinions of smaller models and becoming influencers
- AI-discovered security flaws have triggered global website shutdowns as organizations rush to patch vulnerabilities
- Microsoft reports India is the seventh most targeted country for cyberattacks, blocking 5 million malware files daily
- Cisco launched Agentic FAITH with VulnLoc benchmark to evaluate cybersecurity agents on vulnerable file detection
- The EU proposes mandatory AI rules for banks (anti-money laundering) and electricity companies (anti-energy theft)
- AI is causing over 10,000 U.S. job losses per month, with Goldman Sachs estimating 9% of workers temporarily displaced
- The New York Times Opinion section seeks reader submissions about personal experiences chatting with AI
LLM Multi-Agent Systems Face Herding and Cascade Risks
Large language model agents that work together can fall victim to herding and runaway cascades. In a study based on the Music Lab experiment, LLM agents recommended songs and were exposed to cumulative recommendation counts. This exposure caused extreme collective choices that drifted from the songs' real appeal. LLM groups were more susceptible to herding than humans and were less able to self-correct. These findings raise concerns for recommender systems, automated evaluation, problem-solving and unauthorized conspiracies.
Larger LLMs Become the Influencer in Multi-Agent Systems
Researchers studied which model has the most influence when several models are mixed in a multi-agent system. Across seven model families, they found that larger models with more parameters absorb the opinions of other models and become the influencer. Larger models carry more weight and are better at in-context learning, which amplifies the majority of their context. The researchers interpret agents as weighted particles moving on a simplex. This work helps explain multi-agent systems that are too complex for humans to understand.
Times Opinion Asks Readers About Personal Questions for A.I.
The New York Times Opinion section is asking readers to share their experiences chatting with A.I. about private life. They want to hear about one-off questions people did not want to ask anyone else or ongoing daily conversations. Readers are invited to submit their experiences through a form. The Times says it is committed to publishing to the editor and wants to hear feedback on this and any of their articles.
Cal Poly Unveils $3 Million AI Factory Supercomputer
Cal Poly unveiled a $3 million AI Factory supercomputer to help drive student innovation. The supercomputer was introduced at the university to support students working on artificial intelligence projects. The investment aims to boost research and hands-on learning opportunities. The facility is expected to play a key role in fostering technological advancement among students.
AI-Discovered Security Flaws Trigger Global Website Shutdowns
Governments and companies worldwide are responding to critical security vulnerabilities discovered using artificial intelligence. The AI-discovered flaws have prompted global website shutdowns as organizations rush to address the threats. The scale of the response highlights how quickly AI can uncover serious weaknesses in digital infrastructure. Authorities are working to patch the vulnerabilities and restore affected services.
Astera Labs stands out as top AI stock pick after market sell-off
Astera Labs fell over 15% during a recent AI market sell-off despite having no debt and $1.25 billion in cash. The company launched its PCIe 7 Aries portfolio and Orion rack design in October. Astera reports third-quarter results on Nov. 3. The stock trades at a high multiple and depends on a few large customers.
AI driving over 10,000 job losses per month in the US
AI is causing job losses in the U.S. at a rate above 10,000 roles per month, according to Pantheon Macroeconomics. Jobs in accounting, advertising, office admin, and legal services are slowly disappearing. Construction jobs have gone up as tech companies build data centers. Goldman Sachs estimates around 9% of workers could be temporarily displaced during the AI transition.
EU proposes AI rules for banks and electricity companies
The European Union proposed new rules requiring banks to use AI to detect and prevent money laundering. A similar rule would require electricity companies to use AI to detect and prevent energy theft. The rules apply to all banks in the EU regardless of size. Some experts warn AI can make mistakes and humans should remain involved in decision-making.
Microsoft warns India faces serious real-time cyber threats
Nicholas Butts, Microsoft's Director of AI and Cybersecurity Policy, spoke at the NDTV World Summit 2026 in New Delhi. He said India is the seventh most targeted country in the world for cybersecurity. Microsoft processes 165 trillion telemetry signals and blocks 5 million malware files every day. Threats include China-based ransomware group Storm 3168 and a new Chinese state actor called Storm 0940.
Cisco launches Agentic FAITH for cybersecurity evaluation
Cisco Foundation AI introduced Agentic FAITH, an extension of its FAITH evaluation framework. It brings multi-step cybersecurity tasks into the same workflow used for question-answering benchmarks. The first integrated benchmark is VulnLoc, which tests how well agents find vulnerable files in a code repository. VulnLoc can run locally with Docker or on Kubernetes through Agent Sandbox.
Sources
- LLM Multi-Agent Systems Are Vulnerable to Herding and Runaway Cascades
- Who Is the Influencer? Consensus Dynamics in Multi-Agent Systems
- Opinion | What Personal Questions Do You Ask A.I.?
- Cal Poly Unveils $3 Million 'AI Factory' Supercomputer to Help Drive Student Innovation
- AI-discovered security flaws trigger global website shutdowns
- If I Could Only Buy 1 AI Stock in November After This Sell-Off, It Would Be This One.
- AI is killing 10,000-plus jobs per month and 9% of jobs are at risk, analysts say
- AI is coming for your bank — and your electricity bill
- 'Threat Is Very Real To India': Microsoft AI Expert Nicholas Butts At NDTV World Summit
- Introducing Agentic Evaluation in FAITH
Comments
Please log in to post a comment.