Suggestions
Kevin Owyang
Creative Executive with a passion for merging storytelling, strategy, and data for impactful ROI.
Kevin Owyang is the Creative Director at Desert Tech, known for his entrepreneurial spirit and innovative approach to design. He has a background in producing complex web projects, demonstrating his ability to handle intricate design requests and adapt to client changes effectively. His experience includes a notable educational background from the MIT Sloan School of Management, which has contributed to his success in driving growth and creativity within his roles.
In addition to his work at Desert Tech, Owyang is also active in discussions about the intersection of technology and design, often engaging with topics such as artificial intelligence and its implications in creative fields. His professional journey reflects a commitment to leveraging design for impactful solutions in technology-driven environments.12
Highlights
- False information is a major threat. In India, the risk is highest.
- Unfiltered, it feeds AI training - the next versions of Claude, ChatGPT, etc.
In India the risk of misinformation and disinformation is greater than the risks of infectious diseases, illicit economic activity, inequality (wealth, income) and labor shortages.
In the US it is the 4th-6th most dangerous risk -- comparable to El Salvador, Saudi Arabia, Pakistan, Romania, Ireland, Czechia, Sierra Leone, France and Finland.
Agentic AI experts often point to the reliability and availability of clean training data as a huge problem. For example, the Sharepoint of a large enterprise cam contain various versions of critical data ranging from product specification, testing and analysis, consumer feedback, etc.
Sometimes it's updated drafts based on feedback. Sometimes it's curt notes never intended for public consumption. Sometimes it's completely inaccurate observations made by someone who, later learned they were wrong, and updated the document in a completely new file directory structure.
The challenge is finding ways to tell the AI what is true, what is new, and what is most reliable.
NOW MAGNIFY that problem. Outside the boundaries of a corporate Sharepoint, there agentic AI is being used to present falsehoods as truth. And all this becomes part of the training model of other LLMs.
I'm curious what solutions are on the horizon? Let me know in the comments.
- One ChaptGPT query vs One Google search: 10x electricity
- Generative AI needs 5 billion cubic m of water for cooling in 2027 What does MS Copilot say about it? I asked and it told me
Your LinkedIn post touches on critical environmental concerns related to generative AI. Let's break it down:
-
Energy Intensity: Comparing one ChatGPT query to a typical Google search, the former consumes nearly 10 times as much energy¹. This highlights the substantial power draw of generative AI systems.
-
Water Demand: Generative AI models require significant water for cooling their processors and generating electricity. For instance, OpenAI's GPT-4 data center in West Des Moines, Iowa, used about 6% of the district's water in a single month². Globally, AI's water demand could rival that of the United Kingdom by 2027³.
-
Carbon Footprint: Training large language models emits substantial CO2. In fact, training one such model produces as much carbon as the entire lifetime emissions of five gas-powered cars⁴.
These environmental challenges underscore the need for sustainable AI practices. As the industry grows, we must prioritize energy efficiency, rethink data center design, and explore alternative energy sources. 🌍🔌💧
Want the footnotes and sources? Comment below and I can send them to you!
