Michael Zeller, head of AI at Temasek, revealed at a developer conference in Singapore that new layers must be integrated into the generative AI stack to achieve accuracy.
Zeller said that a data prompt and orchestration layer are necessary, but the real game-changer lies in a data and oversight layer. This critical layer would assess, verify, and cross-check the model against reliable sources, something currently lacking in the field.
Validating and inspecting generative AI models has great potential, but there is a concern about the overhyped expectations surpassing the actual progress in the field. Despite this, Zeller believes that the development will progress quickly, possibly faster than Gartner’s Hype Cycle prediction of two to five years.
Zeller also pointed out that traditional AI is more practical and cost-effective for specific uses compared to generative AI, which can be expensive and unpredictable in its outputs. The current software stack for generative AI is not well-equipped to handle the large datasets and complex models needed for accurate results.
Zeller added that generative AI is still in its early stages of development and that it is important to be realistic about its capabilities. He said that generative AI is not a magic bullet and that it will not replace traditional AI.
The sources for this piece include an article in TheRegister.