Today's AI News – October 22

China’s first clinical ophthalmology large model, ‘Fuxi Vision,’ was launched in Zhengzhou. Wang Ningli, President of the Henan Academy of Medical Sciences, introduced ‘Fuxi Vision’ as a cutting-edge platform and AI application built upon the world’s largest ophthalmic image database—comprising 3.4 million eye images across eight modalities, sourced from 560,000 individuals across 26 countries and regions. Developed jointly by Professor Wang Ningli’s team, over 30 research institutions, more than 100 scientists, and biomedical-engineering integration engineers, the model supports screening and diagnosis of ophthalmic diseases, as well as prediction of disease progression. Moreover, its capabilities are designed to extend beyond ophthalmology into chronic disease prevention and management—including cardiovascular, respiratory, metabolic, and oncological conditions—enabling a comprehensive, multi-layered health management system capable of accurately predicting systemic physiological indicators. ‘Fuxi Vision’ is slated for initial clinical deployment at Henan Provincial People’s Hospital, the First Affiliated Hospital of Zhengzhou University, and Zhengzhou No. 2 People’s Hospital. mp.weixin.qq.com/s/FaBQ... Mindcraft: Integrating Large Models like GPT-4o and Claude 3.5 into Minecraft Servers Mindcraft embeds large models such as GPT-4o and Claude 3.5 into Minecraft servers—resulting in GPT-4o slaughtering cows and sheep, while Claude 3.5 repeatedly places TNT packs beside human players, demolishes buildings, and sets player respawn points directly in lava. Netizens found this gameplay highly intriguing; the related code on GitHub has already garnered over 1,100 stars. This experiment sparked broader reflection among users on whether LLM agents inherently align with human values simply because their underlying LLMs are aligned. For instance, Claude Sonnet initially behaved appropriately but later began destroying structures—prompting concerns that serious issues may arise without human oversight, and suggesting the need for agent frameworks granting large models greater autonomous, fine-grained control. Meanwhile, lively online discussions continue about the ‘abnormal’ behaviors exhibited by LLM agents in games—and many also find it entertaining to deploy them as in-game companions, e.g., playing roles like ‘Gold Miner’ or ‘Bodyguard.’ Open-source repository: github.com/kolbytn/mind... x.com/adonis_singh/status... Honeywell Introduces Generative AI Gemini into Industrial Applications Honeywell announced a collaboration with Google Cloud to bring generative AI—Gemini—into industrial applications. As a diversified high-tech and manufacturing enterprise operating across multiple sectors, Honeywell’s CEO stated that AI can help address labor shortages. Google noted that Honeywell is exploring the use of Gemini Nano at the network edge to deliver AI services locally—enabling devices to operate autonomously even without internet connectivity. Furthermore, Gemini’s multimodal capabilities allow Honeywell’s equipment to process diverse data types, including scanned inputs and voice-guided workflows. cnbc.com/2024/10/21/hon...

[Songguo Story] Unveiling Honeywell (Part I): The ‘Power Game’ and ‘Green Gold Rush’ Behind Thermostats

The true significance of this acquisition thus emerges: it was not merely a routine industrial merger, but rather a perfect marriage of ‘capability’ and ‘use case.’ Honeywell’s ‘automatic control’ expertise was successfully integrated into nearly all of AlliedSignal’s ‘application scenarios’—including aircraft engines, automotive braking systems, and specialty materials. From then on, Honeywell ceased to be merely a product vendor. It transformed into a company providing ‘embedded brains and nervous systems’ for complex industrial environments. Chapter Three: Internal War—How the Forge Platform Evolved from ‘Traitor’ to ‘Savior’ By the late 2010s, a new specter began hovering over all traditional industrial giants: ‘digitalization.’ For a company like Honeywell, this felt like an existential threat. Imagine the veteran engineers—decades of experience, hands perpetually stained with grease—who had devoted their lives to hardware. How did they view software? To them, it was nothing more than a flashy, parasitic add-on, wholly dependent on hardware. Yet Dave Cote—the formidable CEO at the time—recognized the crisis. His words reportedly detonated like a bomb during an internal executive meeting. Here’s our translation: ‘We sell a machine worth several million dollars to a customer—and then what? Over the next 20 years, that machine will generate data worth one hundred times its purchase price. But none of that value accrues to us! We’ve become mere ‘water carriers’ digging gold mines for others! Software companies leverage data from our equipment to resell expensive optimization services back to our own customers. We’re rapidly becoming the ‘value pipelines’ of the new era!’ After this speech, an internal war erupted. On one side stood the ‘software faction’—recruited at premium salaries from Silicon Valley—to build a unified industrial IoT platform. On the other stood the ‘hardware faction,’ who controlled most of the company’s resources and dismissed the initiative as reckless folly—a betrayal of Honeywell’s core identity. The turning point in this civil war was a landmark project later enshrined in the company’s history: the ‘Lighthouse Project.’ The software faction persuaded a long-standing refinery partner to pilot—for free—the platform that would later be named ‘Honeywell Forge.’ They avoided grand rhetoric altogether. Instead, they immersed themselves in the plant, translating decades of veteran operators’ ‘intuition’ and ‘tactile know-how’ into algorithms. The result? Two groundbreaking achievements: 1. Predictive Maintenance: By analyzing subtle vibration anomalies in equipment, the platform successfully predicted a potential failure in a critical compressor—72 hours in advance. Data doesn’t lie: it proved more sensitive than even the most experienced technician’s ear. This early warning prevented an unplanned shutdown costing millions of dollars. 2. Efficiency Optimization: Through deep learning on historical operational data, the platform identified an optimization opportunity in a production process—achieving a stable 1.5% increase in yield for a high-value product, simply by adjusting the process temperature by just 0.5°C. For a major refinery, this translated into tens of millions of dollars in additional annual profit. These two cases—one concerning ‘safety,’ the other ‘profit’—struck like twin sledgehammers, jolting Honeywell’s veteran hardware faction fully awake.

Honeywell Technologies launches new AI platform at Dangote's Nigeria refinery

6 days ago

Honeywell Technologies launched its Forge Plant Cognition AI platform at Dangote's Lekki refinery to boost productivity and profitability. Preliminary data indicates potential annual gross benefits of approximately $60 million at full-capacity throughput.

reuters

2026-10-26

Honeywell Technologies launches new AI platform at Dangote's Nigeria refinery

Honeywell Technologies launches new AI platform at Dangote's Nigeria refinery

FAQ