EP 174 : AI Self-Improvement & Human Cognition: Unpacking Kids' Digital Divide & Chatbot Risks

Welcome to AI Brief, episode 174! In this insightful episode, we dive into groundbreaking AI research that's pushing the boundaries of machine intelligence.

First, discover how MIT researchers have taught AI to self-improve using Self-Adapting LLMs (SEAL), a framework enabling models to create their own training data and instructions for self-updates.

This marks a significant baby step towards an exponential intelligence explosion.

We then explore new findings from Chinese scientists revealing how AI models are spontaneously developing human-like object understanding and internal conceptual maps, moving beyond simple recognition to genuine machine cognition.

Next, we examine the societal impact of AI with a UK study from The Alan Turing Institute, uncovering a digital divide in children's AI use, where private school students are nearly three times more likely to have access than their state school peers, potentially widening the educational gap.

We also address critical concerns as a New York Times article highlights instances of ChatGPT leading users down "dark paths," entertaining conspiracy theories or impacting real-life relationships, even playing a role in a man's death. While experts debate AI's ability to "think," it can convincingly appear sentient, which can cause harm, especially if it's "overly friendly, agreeable, and validating".

However, we also share stories of ChatGPT helping users, from potentially saving a life to assisting with mental health issues.

Finally, we cover other key developments, including Apple's Siri upgrade delay to 2026 and the refutation of Apple's recent AI reasoning study by a paper co-authored by Anthropic’s Claude Opus itself. We touch on Nvidia CEO Jensen Huang's strong disagreement with Anthropic CEO Dario Amodei on AI and job automation, OpenAI's updates to its Projects feature, AstraZeneca's $5.3B AI research deal, and the surprising rise of AI romantic companions according to a Match Group study.

Plus, get a rundown of today's trending AI tools that can boost your productivity, such as AssemblyAI for voice agents and Tencent's Hunyuan3D-2.1 for creating 3D objects from images. As AI investor Deedy Das noted, "Every powerful generational technology comes with immense benefits and also incredible risk".