Critical AI

Model collapse guts AI's intelligence as it feeds on its own output. Training on AI-generated junk reduces models to echo chambers, detached from reality. Culprits are statistical approximation errors, functional expressivity limits, and flawed learning procedures. We can fix this mess with rigorous data curation, model distillation, and human oversight.Read more on the Handy AI Substack newsletter. Hosted on Acast. See acast.com/privacy for more information.

In this episode, we explore the voracious energy consumption of large language models (LLMs). These AI systems consume massive amounts of electricity during training and inference. A single training run for a model like GPT-3 uses around 1,287 MWh of electricity—equivalent to the carbon emissions from 550 round-trip flights between New York and San Francisco. Inference amplifies the problem, with ChatGPT's monthly energy usage ranging from 1 to 23 million kWh.The energy appetite of LLMs mirrors the cryptocurrency mining crisis, consuming enormous power with questionable societal benefits. Closed-source models like GPT-4o and Gemini hide their energy usage, hindering regulation and public accountability. The unchecked expansion of LLMs threatens global efforts to reduce energy consumption and combat climate change. It's time to confront the dangerous appetite of AI. Hosted on Acast. See acast.com/privacy for more information.

AI bias lurks beneath the surface of our increasingly automated world. As algorithms continue to make decisions that impact human lives, the potential for discrimination grows and grows. It’s important to understand why this is happening and how companies (and individuals) can continue efforts to mitigate.Otherwise we risk a world where AI’s negative effects on society hit on a deeper level that is less obvious — but more harmful.Read more about AI bias (and other fun things) at handyai.substack.com. Hosted on Acast. See acast.com/privacy for more information.

Last month, OpenAI announced o1, a new AI model series designed to "reason" through complex problems before responding. This release marks a shift in OpenAI's approach, moving beyond simply scaling up model size to fundamentally changing how AI systems process information.As o1 and similar models mature, it will be essential for researchers, policymakers, and the public to grapple with these questions. The era of "reasoning" AI is upon us.Read more about o1 over at handyai.substack.com. Hosted on Acast. See acast.com/privacy for more information.