AI Job Training Challenges Persist as NYT Cooking Launches AI Tool

Key Takeaways

• Training AI to replace human jobs requires huge effort, money and ongoing human supervision
• NYT Cooking launched an AI tool on iOS for ingredient swaps, serving size changes and weight measurements across 23,000 recipes
• India is identifying AI-resistant jobs in handicrafts, handloom, culinary arts and tourism
• AWS Continuum security agent scores 89% on CyberGym-E2E benchmark for finding and fixing code bugs
• 32 NVIDIA B300 GPUs were found operating inside China Telecom despite U.S. export controls
• Power consumption monitoring can detect AI training runs in data centers at least 96% of the time
• Sam Altman warns one company could gain control of key AI resources and block competitors
• African experts call for stronger links between AI training and employment across the continent
• Tesla and Waymo may follow the same industry-leadership pattern now emerging in AI infrastructure
• AWS Continuum fixed the original historical bug in only 37.8% of tasks despite strong bug detection rates

Training AI to replace human jobs is its own hard work

Training AI and robots to do familiar human tasks takes huge effort, money and supervision. Workers must repeat tasks hundreds of times so machines can learn. Even after training, humans must keep managing and monitoring AI. The article questions whether replacing workers with AI is worth the ongoing work it creates.

NYT Cooking adds AI tools for recipe changes

The New York Times Cooking team launched an AI tool that helps cooks modify recipes. Users can request ingredient swaps, adjust serving sizes and measure by weight. The tool is rolling out on the iOS app after the team found it could not handle 23,000 recipes manually. AI made these features possible.

India searches for AI-proof jobs and global strengths

Commerce Minister Piyush Goyal said the Commerce Ministry is identifying jobs less vulnerable to AI. He spoke at an event in Berlin on October 23, 2026. India wants to lead in areas like handicrafts, handloom, culinary arts and tourism. He said traditional skills and human creativity still matter and India will support its 1.4 billion people.

AWS AI security agent fixes 89% of bugs in test

AWS made its Continuum security agent generally available on Thursday. It scored 89% on the CyberGym-E2E benchmark by finding and fixing code bugs. The agent found bugs in 92.5% of tasks and kept tests passing in 89% of cases. It only fixed the original historical bug in 37.8% of tasks, and AWS ran the tests itself.

32 NVIDIA B300 GPUs found inside China Telecom

A security firm found 32 NVIDIA B300 GPUs operating inside China Telecom. The GPUs were found through a tool exposed on the open internet. U.S. export controls limit sales of the B300 to approved buyers. NVIDIA called this a small part of its business. The discovery shows top AI hardware is still reaching China.

AI infrastructure shift mirrors old tech industry changes

The author compares today's AI infrastructure race to the autonomous vehicle boom of 2022. He previously worked at Luminar, a lidar company for self-driving cars. He noticed that tech firms like Tesla and Waymo often led major industry shifts, not the old established companies. Now he sees the same pattern emerging in AI infrastructure and optical networking.

African experts urge AI training tied to jobs

Experts at a BUILD 2026 panel in Abuja called for stronger links between AI training and employment across Africa. They urged investment in local innovation, industry-driven education and African-language AI datasets. Speakers highlighted the need for practical experience, mentorship and better computing resources. They said governments, universities and businesses must work together to build inclusive tech capacity.

Power use may reveal secret AI training in data centers

Researchers found that power consumption can help detect if a data center is training an AI model. During training, GPU activity creates a power rhythm that special tools can track from outside the building. In tests, the method detected at least 96 percent of training runs and had a low false positive rate. This approach could help regulators audit AI activity without relying on company reports.

Sam Altman warns AI race may lead to one company holding all power

Sam Altman outlined three possible endings to the AI race. He said the most dangerous outcome is one company becoming so powerful it controls key resources and blocks competitors. He suggested such a company might disable others' hardware to maintain dominance. Altman favors global oversight or a shared research lab instead, similar to nuclear energy regulation.

Sources

• The hard work of replacing hard work
• How, and Why, We’re Adding A.I. Tools to NYT Cooking
• Commerce Ministry identifying AI-proof professions, sectors where India can take global lead: Goyal
• AWS says its AI security agent fixes 89% of bugs
• Security Firm Discovers 32 NVIDIA B300 GPUs Operating Inside China Telecom, Showing U.S. Export Controls Aren’t Curbing Top-Tier AI Hardware From Getting Out
• The AI infrastructure race: I've seen this movie before
• Experts Urge Africa to Link AI Training with Jobs
• Can power consumption reveal whether a data center is secretly training AI?
• Sam Altman predicts AI race could end with ‘terrible’ scenario where one company holds all power, cuts off resources