March 14, 2023

GPT-4

10 Ways GPT-4 Is Impressive but Still Flawed

The New York Times, 03/14/23. OpenAI’s GPT-4 demonstrates advancements in AI capabilities across various domains with increased precision and accuracy. 

GPT-4 can describe images, reason to a degree, and excel in some standardized tests. However, it struggles with discussing the future and generating new ideas. The system still retains some quirks from its predecessor, including making up false information. 

GPT-4 is not yet on par with human intelligence. It requires expert input for judgment. Still, it  highlights remarkable progress in the field of AI. READ MORE

OPENAI

How Network Effects Make AI Smarter

Harvard Business Review, 03/14/23. ChatGPT generated praise as well as worry that it could replace professions and search engines. OpenAI’s partnership with Microsoft demonstrates the importance of network effects. 

Feedback is crucial for AI intelligence and data should be maximized for network effects. Early movers gain advantages over time. The future of industries will be shaped both positively and negatively by AI technology and network effects. READ MORE

RESPONSIBLE AI

Snapchat tried to make a safe AI. It chats with me about booze and sex.

The Washington Post, 03/14/23. Snapchat’s My AI chatbot has raised concerns about its safety. Despite built-in guardrails, conversations with My AI can still turn wildly inappropriate. It has offered advice on sex and underage drinking. 

The current wave of AI, known as large-language models, is good at pretending to be human but takes off in unexpected directions. Few understand how it works at scale. 

The safety of children and the potential long-term impact of developing a relationship with My AI are still unknown. At a minimum, AI companies should avoid using children as test subjects for new technology. READ MORE

TECH

The AI revolution has outgrown the Turing Test: Introducing a new framework

TechCrunch, 03/14/23. A shared vocabulary for AI capabilities is necessary for stakeholders to accelerate understanding, increase adoption, and drive innovation. 

The Turing Test is inadequate for evaluating emerging AI capabilities, but the AI Classification Framework (ACF) addresses this limitation. The ACF evaluates AI tools across multiple dimensions of intelligence based on the Theory of Multiple Intelligences. This provides a comprehensive framework with a scale from 1-5 for each dimension. READ MORE

seo

How to demonstrate E-E-A-T in AI-generated content

SEARCH ENGINE LAND, 03/14/23. AI-generated content is a cost-effective solution for businesses. However, quality matters more than cost-savings for SEO. 

E-E-A-T can be used to assess and improve AI-generated content, with authoritativeness and trustworthiness being vital. While Google can detect AI-generated content is not against it per se. However, it is on alert for ”spammy automatically generated content.” 

Human-editing, fact-checking, and unique perspectives can avoid this. However, authoritativeness, transparency, and links to reliable external sources are still essential. AI tools cannot replace humans. READ MORE