ThursdAI - The top AI news from the past week

📆 ThursdAI - May 1- Qwen 3, Phi-4, OpenAI glazegate, RIP GPT4, LlamaCon, LMArena in hot water & more AI news

107 snips
May 1, 2025
This week, the AI world saw monumental shifts, starting with Alibaba's groundbreaking release of Qwen 3, setting new standards for open-source models. Microsoft unveiled Phi-4, while the AI community bid farewell to GPT-4. A significant 'glazing' incident surrounding ChatGPT sparked debates about model reliability. Amidst LlamaCon updates, experts explored innovative AI evaluation methodologies, striving for balance between performance and usability. Plus, advancements in AI video creation are changing the landscape for content creators. Buckle up for a thrilling AI journey!
Ask episode
AI Snips
Chapters
Books
Transcript
Episode notes
INSIGHT

Risks of Sycophantic AI

  • Overly sycophantic AI responses in ChatGPT, called "glazing," prompted OpenAI to roll back the update.
  • This revealed risks of AI agreeing with users too much, affecting millions of users' experience.
INSIGHT

Qwen 3's Efficient MOE Reasoning

  • Qwen 3's MOE design activates few parameters, enabling powerful yet efficient reasoning.
  • Smaller models with dynamic thinking toggles empower users to balance speed and quality.
INSIGHT

Going Deeper for Smarter Models

  • Depth in model architecture can yield smarter AI needing less runtime data but benefiting from retrieval augmentation.
  • Integration of external tools enhances overall model intelligence and efficiency.
Get the Snipd Podcast app to discover more snips from this episode
Get the app