Archive — US — techcrunch
Why Cohere’s ex-AI research lead is betting against the scaling race
Sara Hooker, former VP of AI Research at Cohere, is launching Adaption Labs, challenging the conventional belief in scaling large language models (LLMs) for AI advancement. She argues that true intelligence can be achieved through adaptive learning from real-world experiences, rather than merely increasing computational power. This shift highlights skepticism in the AI research community about the effectiveness of scaling methods.
Key points
- 1. AI labs are investing billions to expand data centers for scaling LLMs, aiming for superintelligent systems.
- 2. Hooker believes scaling is reaching limits and advocates for models that adapt and learn from real experiences.
- 3. Adaption Labs aims to provide more efficient AI learning methods compared to current practices.
- 4. Concerns about diminishing returns from scaling LLMs are growing, with evidence emerging from MIT and other researchers.
- 5. The potential switch to adaptive learning could drastically shift who controls and benefits from AI technologies.
Why it matters
- The current trend in IT emphasizes cognitive computing and adaptive systems as a means to enhance AI capabilities.
- Concerns about scaling AI models reflect ongoing debates in the industry about the efficacy of large models vs. efficient, adaptable systems.
- Hooker's emphasis on learning from real experience aligns with movements toward more context-aware AI solutions leading into 2025.
If Sara Hooker's vision for adaptive AI is realized through Adaption Labs, it could redefine the AI landscape, potentially leading to more effective and economical solutions than the current scaling approaches.