OpenAI Q* Reasoning Rumor is Similar to Meta AI Planning Approach

by

Meta is working on adding latent space planning/search to large language model AI.

1️⃣H-GAP (https://arxiv.org/abs/2312.02682)
2️⃣Diffusion World Model (https://arxiv.org/abs/2402.03570)
3️⃣TAP (https://arxiv.org/abs/2208.10291)
4️⃣LaMCTS (https://arxiv.org/abs/2007.00708)
5️⃣LaP3 (https://arxiv.org/abs/2106.10544)
6️⃣LaSynth (https://arxiv.org/abs/2107.00101)
7️⃣LaMOO (https://arxiv.org/abs/2110.03173)

Rumored OpenAI Q* Is Optimization in Abstract Representation Space

The innovation in Q* lies in its optimization process, conducted not within the space of possible text strings but in an abstract representation space. Here, thoughts or ideas are represented in a form that allows for the computational minimization of the EBM’s scalar output, akin to finding the path of least resistance in a landscape. This process involves gradient descent, a method for finding the minimum of a function, applied to iteratively refine these abstract representations towards those that yield the lowest energy in relation to the prompt.