AIAW E134 - Reflection 70B, Strawberry and other AI rumours

What if our AI models aren't as smart as we think? Ever wondered how AI intersects with entrepreneurship and the arts? Are bold claims in AI model development leading us astray? 

Join us as we uncover the surprising gaps between expectations and real-world performance of language models, featuring expert insights from Jesper Fredriksson and Robert Luciani. We dive deep into the challenges of model quantization and fine-tuning, and you'll hear first-hand about a fascinating case where a model trained for reasoning and reflection came up short

We tackle a recent controversy over a model that promised to surpass existing frontier models in reasoning, only to fall flat in reproducibility. The initial excitement quickly spiraled into skepticism and accusations of fraud. Through Jesper and Robert's expert lenses, we explore the steps taken to mitigate these issues and ponder whether the creator was merely misguided or intentionally deceptive. 

Our conversation also touches on the complex interplay between human cognitive biases and AI, the intriguing parallels between human memory and artificial intelligence, and the critical importance of navigating AI's transitional period with care. 

Tune in for a thought-provoking episode that challenges the boundaries of AI development and its future possibilities.

Follow us on youtube: https://www.youtube.com/@aiawpodcast