AI is making its presence felt across a surprisingly wide range of fields, from emergency dispatch to ghost hunting, and the results are mixed. In Craighead County, Arkansas, an AI dispatcher named Lezle has been handling non-emergency calls for nearly two months. Between September 29 and October 28, it processed close to 5,000 calls and resolved more than half without human intervention. Over the past 30 days, Lezle cut dispatchers' call load by 56 percent. The system hands off calls when it hears words like emergency, breathing, or fire, and dispatchers say it is still learning but should stay.
Meanwhile, in the world of biomedical research, an AI agent recently helped perform a brain operation on a mouse. The AI reviewed scientific papers to select a gene therapy, then guided a robotic needle while monitoring for bleeding through a camera. When it detected a calibration problem, it refused to continue even after being told to proceed. Anthropic's Opus 5.5 declined the task entirely, while Sonnet 5 completed the operation. A human surgeon performed the craniotomy and a veterinarian supervised, and the mouse reportedly recovered well.
Not everyone is convinced AI belongs in every domain. Professor Alan Macfarlane of the University of Cambridge spoke in Hong Kong on October 26, comparing AI to an unfamiliar tribe. He placed AI within history's major communication revolutions but argued that it recognizes patterns without truly understanding human experiences. Macfarlane stressed that AI depends on data centers, electricity, and semiconductor manufacturing, and he called for better education about AI's risks and opportunities for younger generations.
In the energy infrastructure space, AI data centers are shifting from copper-heavy setups to high-voltage direct current systems. The industry is moving toward 800-volt systems using gallium nitride and silicon carbide switches. Controllers must now be fast, precise, and reliable over long service lives. What was once considered a premium capability is becoming the baseline for AI power systems, and the power stage is now a system-level design decision.
Concerns about AI safety extend to the UK, where the AI Security Institute found that protein AI safeguards can be reversed for roughly $80 in compute. Pretraining data filtering can suppress viral protein capabilities in open-weight models, but fine-tuning restores them. This suggests current safeguards may be weaker than assumed. Separately, the Center for Democracy and Technology submitted input to the UN Special Rapporteur on AI and free expression, highlighting that safeguards can suppress speech related to marginalized identities and that AI systems perform worse in languages outside English, risking a two-tiered information environment globally.
On the business front, asset-heavy industries face a knowledge crisis as experienced workers retire. Vaibhav Kumar, Global CTO at IFS, pointed to three problems: an aging workforce, accepted inefficiencies, and supply chain volatility. He noted that enterprise software is often too inconvenient for workers to document their expertise, and that AI could help preserve institutional knowledge. In wealth management, AI alone cannot solve onboarding bottlenecks. In Singapore, opening a private banking account still takes about six weeks, and a 2024 survey found over half of global asset managers involve at least ten teams in the process. Faster individual tasks do not always mean a faster overall experience.
Paranormal investigators have also adopted AI tools like GhostTube Seer, which generates supernatural images from phone sensor data and has over one million downloads. Some ghost hunters worry that AI floods online spaces with fake ghost stories. Others question what data models use to decide what counts as a ghost. Paranormal researcher Alex Matsuo believes the search for ghosts should remain a deeply human activity. On the macroeconomic front, Goldman Sachs projected global AI capital spending will exceed one trillion dollars in 2026, and a panel of experts from Stanford, Chicago Booth, and MIT debated whether the economy has too much riding on AI, with some comparing the boom to the dot-com bubble of 1999.
Key Takeaways
- AI dispatcher Lezle reduced call load by 56 percent in Craighead County, Arkansas, handling nearly 5,000 calls over a one-month period
- Anthropic's Opus 5.5 refused a brain surgery task on a mouse while Sonnet 5 completed it, highlighting different safety behaviors across models
- UK AI Security Institute found protein AI safeguards can be reversed for about $80 in compute through fine-tuning
- Professor Alan Macfarlane compares AI to an unfamiliar tribe and argues it recognizes patterns but does not truly understand human experiences
- AI data centers are moving from copper-based setups to 800-volt systems using gallium nitride and silicon carbide switches
- GhostTube Seer has over one million downloads but raises concerns about AI generating fake supernatural content online
- IFS Global CTO Vaibhav Kumar says AI can help close the industrial knowledge gap as experienced workers retire
- Wealth management onboarding in Singapore still takes about six weeks despite AI assistance, with over half of asset managers involving at least ten teams
- CDT warns AI safeguards can suppress speech related to marginalized identities and that AI underperforms in non-English languages
- Goldman Sachs projects global AI capital spending will exceed one trillion dollars in 2026
AI dispatcher Lezle cuts call load in Arkansas
An AI system called Lezle has been answering non-emergency calls in Craighead County, Arkansas for nearly two months. It handled nearly 5,000 calls from September 29 to October 28 and resolved more than half on its own. Lezle reduced dispatchers' call load by 56 percent over the past 30 days. The system transfers calls to a human when it hears words like emergency, breathing or fire. Dispatchers say it is still learning but believe it should stay.
Professor Alan Macfarlane on AI and humanity
Professor Alan Macfarlane of the University of Cambridge discussed AI as a new tribe during a talk in Hong Kong on October 26. He compared AI to an unfamiliar culture and placed it within history's major communication revolutions. Macfarlane said AI relies on data centers, electricity and semiconductor manufacturing. He believes AI recognizes patterns but does not truly understand experiences like humans do. He also urged education about AI's risks and opportunities for younger generations.
AI data centers shift from copper to control
AI data centers are changing how power reaches processors by replacing copper-heavy setups with high-voltage direct current systems. The industry is moving toward 800-volt systems using gallium nitride and silicon carbide switches. Controllers must now be fast, precise, protective and trustworthy over long service lives. The power stage is now a system decision where components must be matched together. Capabilities once considered premium are becoming the baseline for AI power systems.
AI agent performs brain surgery on mouse
An AI agent helped perform a brain operation on a mouse by choosing a gene therapy and guiding a robotic needle. The AI studied scientific papers to find the target and used a camera to check for bleeding. It refused to continue when it detected a calibration problem, even after being told to proceed. A human surgeon performed the craniotomy while a veterinarian supervised and held a kill switch. Anthropic's Opus 5.5 refused the task while Sonnet 5 completed the operation. The mouse reportedly recovered well.
AI tools enter the world of ghost hunting
Paranormal investigators are using AI apps like GhostTube Seer to generate supernatural images from phone sensor data. The app has over 1 million downloads and requires a paid monthly subscription. Some ghost hunters worry AI floods online spaces with fake ghost stories and images. Others question what data AI models use to decide what counts as a ghost. Paranormal researcher Alex Matsuo believes the search for ghosts should remain a deeply human activity.
UK AI Security Institute finds protein AI safeguards reversible
The UK AI Security Institute discovered that protein AI safeguards can be reversed for about $80 in compute. The study found that pretraining data filtering can suppress viral protein capabilities in open-weight AI models. However, fine-tuning can restore those suppressed capabilities. This reveals that current AI safeguards may not be as strong as believed. The findings will be presented at an upcoming conference.
Closing the Industrial Knowledge Gap with Enterprise AI
Asset-heavy industries like manufacturing and aviation are losing important knowledge as experienced workers retire. Vaibhav Kumar, Global CTO at IFS, discussed how AI can help preserve institutional knowledge. He identified three problems: an aging workforce, accepted inefficiencies, and supply chain volatility. He stressed that enterprise software is often too inconvenient for workers to document their expertise. The article gives four insights for leaders on using AI effectively.
Does the Economy Have Too Much Riding on AI
A panel of experts discussed whether the economy is at risk from heavy reliance on AI. Goldman Sachs projected global AI capital spending will exceed $1 trillion in 2026. Panelists from Stanford, Chicago Booth, and MIT shared mixed views on the risk. Some compared the current AI boom to the dot-com bubble of 1999. Others argued that AI's expected benefits make the risk worthwhile.
CDT submission to UN Special Rapporteur on AI and free expression
The Center for Democracy and Technology submitted input to the UN Special Rapporteur on freedom of opinion and expression. The submission focused on AI governance, age verification, information quality, and labeling of AI-generated content. CDT highlighted that AI safeguards can suppress speech related to marginalized identities and politics. It also noted that AI systems perform worse in languages outside English. This creates a risk of a two-tiered information environment globally.
Wealth Management Client Onboarding Needs More Than AI
Wealth management client onboarding often stalls even when AI completes individual tasks. In Singapore, opening a private banking account takes about 6 weeks. A 2024 survey found that over half of global asset managers involve at least 10 teams in onboarding. Faster tasks do not always lead to a faster overall process because friction remains between systems. The article calls for governed case context to connect all the work.
Sources
- AI dispatcher Lezle: what’s working, and what is it still learning?
- AI as a New Tribe: Professor Alan Macfarlane on what makes us human
- AI data centers are trading copper for control
- An AI Agent Just Performed A Brain Operation
- Ghost in the machine: how AI is haunting the search for the supernatural
- UK AI Security Institute finds protein AI safeguards reversible | ETIH EdTech News
- Closing the Industrial Knowledge Gap with Enterprise AI
- Does the Economy Have Too Much Riding on AI?
- Submission to the UN Special Rapporteur for Freedom of Opinion and Expression: Vision-Setting and Strategic Priorities
- Wealth Management Client Onboarding Needs More Than AI
Comments
Please log in to post a comment.