GPT-5: It Just Does Stuff by Ethan Mollick

In his now famous piece still relevant piece from 2019 Sutton teaches us that the critical lesson from decades of AI research:leveraging computation through general methods is far more effective than incorporating human knowledge into AI systems.Original article here.

Today we're covering our first scientific paper. The authors present "Agentic Web," a comprehensive conceptual framework for the next phase of the internet, where autonomous AI agents act as primary intermediaries.Let us know how you like this format by emailing us, details at the end of the podcast!Article is here: Agentic Web: Weaving the Next Web with AI AgentsResearch paper is here: Arxiv research link

Ines Lee—a former economics professor who led content creation for productivity expert Ali Abdaal—reframes the challenge as opportunity: Instead of chasing AI engineering skills, become T-shaped with deep expertise in one area plus broad capabilities across related domains. Her three-phase framework shows exactly where humans add irreplaceable value, proving that markets reward relevance over credentials. Plus: Five steps to start building your T shape.Original article hereReferences from this article T-shaped professional concept by Tim Brown (IDEO cofounder) McKinsey report on rising demand for socio-emotional skills

In this article, Leon explains why he supports the right of educators and academics to refuse the use of artificial intelligence and why he believe some of the criticisms levelled against resistors are flawed.Original article here: The Right to ResistReferences from this article The Myth of Inevitable AI Publishers' Open Letter Against AI Artists and Writers Open Letter on AI Copyright Netherlands Academics Open Letter on AI Australian/US Educators Open Letter Refusing GenAI AI Ethics - Technology Flaws Network Effects and Social Media Companies Instructure and OpenAI Partnership Environmental Costs of AI AI Copyright Issues AI Bias Issues AI Privacy Concerns AI Data Collection Issues Microsoft Research on GenAI Impact on Critical Thinking Rethinking Assessment for GenAI Ebook The Myth of Inevitable AI Podcast Episode It's Uncomfortable on the Fence But At Least the View is Nice

This article underscores the need for educational reform in the face of technological evolution. As we think about the role of education, we must consider how to cultivate a mindset of curiosity and resilience in our students—essential skills for navigating an uncertain future. Original article here: Learning to LearnReferences from this article Encouraging Metacognition in the Classroom | Poorvu Center for Teaching and Learning Metacognitive Strategies | Center for Teaching Innovation Teaching Metacognitive Strategies in the Classroom Fostering Metacognition to Support Student Learning and Performance | CBE—Life Sciences Education Metacognition | Teaching + Learning Lab

Ethan Mollick shares examples about what GPT-5 gets right and what it just decides it's going to do without being asked. Despite its advancements, GPT-5 still requires human oversight to correct errors and validate results, emphasizing the need for collaborative interaction with AI.Original article: GPT-5: It Just Does StuffReferences from this article Google Gemini 2.5 with Deep Think achieving gold medal at the International Math Olympiad 3D city builder simulator created with GPT-5

Lee's perspective on AI reframes the conversation from fear to practical management strategies. He argues that concerns about losing control over AI agents may be overstated. He suggests that with proper supervision, similar to how humans manage tasks and responsibilities, organizations can effectively integrate AI systems without ceding too much control. Rather than viewing AI as a threat, Lee advocates for a balanced approach that emphasizes human oversight and robust review processes to harness AI's potential safely.Original article here: https://www.understandingai.org/p/keeping-ai-agents-under-control-doesnt References from this article:2024 paper by Yoshua Bengio and Geoffrey Hinton on AI riskAgent Week series - Reinforcement Learning ExplainedAgent Week series - How AI Agents Got Good at Using ToolsWhat I Learned Trying Seven Coding AgentsPrinciple of Least PrivilegeApollo Research study on AI scheming capabilitiesWhy I'm Not Worried About AI Taking Over the Physical WorldPredictions of AI Doom Are Too Much About the FutureWhy I'm Not Afraid of Superintelligent AI

Recent studies show we can get better results by using generative AI. But to get them at scale, our organizations would have to change, and that's slow and difficult. genAI itself may change all that.Original article: https://www.wildworldofwork.org/p/ai-is-fast-automation-is-slow-can

Organizations are messy “garbage cans,” and our instinct is to tame that mess with handcrafted process and bespoke agents like Manus. The Bitter Lesson says the opposite often wins—systems trained for outcomes, scaled with compute, tend to outpace careful craft.References from this article Moving off the Map: How Knowledge of Organizational Operations Empowers and Alienates by Ruthanne Huising The Garbage Can Model 43% of American workers have used AI at work (research paper) The Bitter Lesson by Richard Sutton (2019 essay) Our first AI-powered teaching games AI ability to work autonomously is increasing very rapidly Manus Tips for building agents from the Manus team ChatGPT agent OpenAI used reinforcement learning to train their AI Atlantic article on chess ratings Original article link

This study challenges the common assumption that better AI models alone will yield superior results, emphasizing the critical role of the human that's in the loop. References from this article: In a large-scale experiment MIT Initiative on the Digital Economy Leading the AI-Driven Organization courseOriginal article