Leveraging AI 145 | Reasoning models competition intensifies, When will we achieve AGI? Grok access not through X, Microsoft Co-pilot Custom GPTs are back! and many more AI news for the week ending on November 30th 2024
Nov 30, 2024
The podcast dives into the transformative potential of new reasoning models in AI, highlighting competition among tech giants like OpenAI and Alibaba. Discussions reveal the mixed opinions on AGI arrival timelines, with experts predicting anywhere from one to ten years. The impact of recent AI incidents raises ethical concerns, particularly regarding safety and behaviour. Innovations from Microsoft, such as custom GPTs and LazyGraph technology, aim to enhance user experiences while addressing vulnerability in AI interactions, especially for teenagers.
AI Snips
Chapters
Transcript
Episode notes
Thinking Models
- The focus in AI models is shifting from scaling laws and GPU-heavy training to "thinking models".
- These models prioritize reasoning and problem-solving at the time of request.
Test-Time Compute Advantage
- Test-time compute models offer a significant advantage to companies with limited GPU access, particularly in China.
- This allows them to train smaller models faster and potentially close the gap on US companies.
AGI Timeline Debate
- Leading AI thinkers have differing opinions on AGI's arrival, with some predicting it within 2-3 years and others estimating 5-10.
- Demis Hassabis believes several breakthroughs are still needed, while Jan LeCun thinks current LLMs aren't the path to AGI.