Skeptics with a K

Episode #436

October 16, 2025

Key Takeaways Copied to clipboard!

  • The Turing Test, as commonly applied today, is a poor and misleading metric for gauging artificial intelligence because it primarily tests linguistic mimicry, can be passed by non-intelligent programs like ELIZA, and is frequently failed by genuinely intelligent entities like animals. 
  • The current trend in AI development focuses on creating ever-larger language models, driven by the unproven hypothesis that scale alone will lead to Artificial General Intelligence, while ignoring critical issues like energy consumption and the need for efficiency improvements. 
  • The final QED conference will feature panels on weight stigma and the crucial topic of integrating evidence into policymaking, alongside a live recording of the European Skeptics Podcast. 

Segments

AI Impact on Software Engineering
Copied to clipboard!
(00:02:22)
  • Key Takeaway: AI-generated code often exhibits poor practice, security vulnerabilities, and over-complicated solutions, requiring significant quality assurance time from human engineers.
  • Summary: There is an argument in software engineering to replace junior engineers with AI, but this risks eliminating the training pipeline for future senior engineers. AI code is prone to producing insecure and hard-to-maintain solutions. Engineers using AI code spend substantial time verifying its quality and adherence to style guides.
AI Energy Consumption Comparison
Copied to clipboard!
(00:06:59)
  • Key Takeaway: The power consumption of large language models is rapidly approaching country-scale levels, though it remains less than the energy usage of cryptocurrency mining.
  • Summary: Newer AI models can be made smaller and more efficient by removing rarely executed pathways post-training without performance loss. Popular LLMs like Gemini and ChatGPT are collectively approaching the power consumption scale of entire nations. Bitcoin mining alone consumes electricity comparable to the Czech Republic, exceeding the consumption of Thailand.
Critique of the Turing Test
Copied to clipboard!
(00:12:25)
  • Key Takeaway: The Turing Test is an inadequate measure of general intelligence because it only tests linguistic mimicry and is frequently failed by intelligent non-humans and passed by non-intelligent programs.
  • Summary: The original Turing Test (Imitation Game) proposed in 1950 was a proxy to determine if a machine could ’think’ by impersonating a human well enough to fool a judge. Modern implementations are poorly defined regarding judge type, duration, and pass threshold, leading to inconsistent results, such as the program Eugene Goostman passing in 2014 by claiming to be a 13-year-old Ukrainian boy.
QED 2025 Final Announcements
Copied to clipboard!
(00:39:05)
  • Key Takeaway: The final QED conference schedule includes panels on weight stigma (‘The Fat of the Matter’) and evidence-based policymaking, alongside a live show featuring the European Skeptics Podcast.
  • Summary: The European Skeptics Podcast will host a live show at QED, joining other live podcast recordings. A panel titled ‘The Fat of the Matter’ will address weight stigma with experts including Dr. Asher Lamy (’the fat doctor’). The final announced panel, ‘Evidence in Action,’ will focus on integrating scientific evidence into parliamentary policymaking.