Quantum Blockchain Technologies, the AIM-listed blockchain technology research and development company, has confirmed the first successful generation of AI Oracle models trained directly on data from an unnamed ASIC manufacturer's mining rig.
The milestone follows a June announcement in which QBT reported completion of the structured operational dataset collection from the manufacturer's platform, which served as the training input for the AI models developed by the company's research team at the University of Milan.
The team encountered a significant technical hurdle: the ASIC manufacturer's data differs materially from data previously generated using the Bitaxe Gamma platform, which runs on Bitmain's BM1370 chip, both in structure and statistical characteristics.
Adapting the AI learning framework to account for the new architecture required substantial retargeting of the models' learning objectives, work QBT says is now complete.
Two implementations of the AI Oracle, a standard version and a more compact variant, are being evaluated in parallel, with the team assessing trade-offs between predictive capability, computational efficiency and deployment characteristics.
Live testing on the manufacturer's Mining Development Kit, connected to a commercial mining pool, will begin once the R&D team determines performance has reached a satisfactory threshold.
A technical review meeting between QBT and the ASIC manufacturer's engineering team is being scheduled, at which an updated project timeline covering the remaining steps to live testing will be presented.
