Thore Graepel, a former DeepMind researcher, has launched a startup named Metis Reasoning. According to a report by The Decoder, the company intends to apply the search principles used in AlphaGo to general AI reasoning in an effort to address the problem of hallucinations.
The Proposed Approach
Rather than relying solely on knowledge stored directly within model weights, the startup envisions using an explicit knowledge ledger. The intended function of this ledger is to ensure that every conclusion generated by a model can be directly checked against underlying evidence.
What Remains Unclear
Despite the conceptual proposal, significant questions remain regarding the technology’s viability. The announcement did not specify a working system, nor were any benchmarks provided to demonstrate how the approach performs in practice.
Observers will need to wait for concrete demonstrations or technical papers to determine whether AlphaGo’s search methodology can successfully transition to general reasoning tasks.
Lead image: Christina Morillo / Pexels
