The report highlights the vast costs of training a model.
“Based on variables released by Google et al., you’re paying circa $1 per 1,000 parameters. This means OpenAI’s 175B parameter GPT-3 could have cost tens of millions to train. Experts suggest the likely budget was $10M.”
The report features PolyAI’s ConveRT model as a prime example of how data-efficient models can outperform large models in specific dialogue use cases, offering an affordable, pre-trained solution to enterprise clients.
Check out what our CTO, Shawn Wen, has to say about PolyAI’s contributions in this video:
You can read more in this ZNet write up of the report.
The report also showcases PolyAI’s voice assistant for restaurant bookings, which launched last year with a large UK restaurant group. The report highlights how our voice assistant can:
- understand noisy speech from telephone lines and has a success rate of >90% for an average 8-turn conversation.
- learn from interactions much faster than their predecessors like Siri or Alexa.