Newsletter
Weekend Briefing: 14 January 2024
This Week from Artificiality: Running with Scissors, 10 Research Obsessions for 2024, Barbara Tversky & Spatial Cognition, Interpreting Intelligence Part 3
This Week from Artificiality:
OpenAI isn’t just running fast & breaking things—it’s running with scissors. And that could be a major issue in 2024.
Key Points:
- OpenAI released a "GPT Store" for user-created GPTs and "ChatGPT Team" for shared team access to ChatGPT.
- GPTs have flaws—they lack IP protection as their instructions are easily extracted.
- The GPT Store lets developers "sell" GPTs that can be easily replicated by downloading the instructions but the GPT Store provides no revenue share details, global payments, or effective search/discovery.
- Some claim the GPT Store is OpenAI's "app store moment" but it lacks basic, expected app store features.
- As the market moves to valuing workflow integration over raw AI potential, OpenAI's weaknesses could damage its current market leadership.
In our Artificiality Pro update for January, we covered our 10 research obsessions for 2024.
Key Points:
- Agentic AI. Personal generative AI that is able to act as a personal agent will redefine how we use the web and/or apps.
- AI Inside. Shift from novel AI apps (ChatGPT) to apps with AI inside (CoPilot) to provide AI benefits within existing workflows.
- Edge AI. AI in the cloud benefits from scale but is challenged by cost and privacy. Mobile solves these challenges but is lagging in capability.
- AI-Enhanced Learning. The integration of generative AI in learning and skill development is revolutionizing the educational landscape.
- Human Centered Generative AI. Humans will only embrace AI if it enhances their desire for purpose and agency.
- Interpretability. Interpretability enables critical value-proposition generative AI.
- Memory vs Margins. Increasing memory increases context and usefulness but memory is expensive.
- Source Dilemma. Closed-source vs open-source will be an important, path dependent choice that affects auditability, cost, quality, security, and vendor lock-in.
- Trust. In order for AI to be useful, we need to know if/when to trust it.
- World of Workflows. Generative AI’s impact on work will be multifaceted but at its core, the route to higher productivity involves making decisions about whether we want AI to compete with or complement our cognition.
We're revisiting one of our most thought-provoking episodes, originally recorded in April 2022, featuring Barbara Tversky, the author of Mind in Motion: How Action Shapes Thought.
Key Points:
- Spatial thinking involving acting in spaces we inhabit is fundamental to human cognition, but has been neglected in favor of focusing on language.
- Spaces like geographic distance are distorted in our minds by perspective and other heuristics. These distortions also appear in conceptual spaces like color, politics, etc.
- External symbolic representations like ancient cave drawings are early evidence of abstract thought and have enabled scientific and societal progress.
- Sketching facilitates discovery and creativity through ambiguity that allows reinterpretation and making inferences. Digital tools can lack this flexibility.
- Spatial thinking is multimodal, involving senses like wind, smell and texture, not just visual. Blind people can have strong spatial cognition.
Three spooky things we’ve learned in 2023 about how neural networks learn. Part 3.
Key Points:
- Memorization lays a foundation for learning general patterns, similar to human learning. We define intelligence by the ability to generalize.
- Understanding how models learn shows they can exhibit spookily intelligent learning.
- As we understand more how models learn, mechanistic interpretability has moved from nascent to an engineering problem. This will lead us to new understanding of intelligence.
