Bloomberg’s Mark Gurman said that Apple is testing a range of new large language models in an internal setting.
Apple is believed to be developing several AI models of different capacities. Gurman claimed that the Cupertino-based company has deployed versions with 150 billion, 33 billion, and 3 billion parameters. The 150 billion parameter model draws its power in the cloud, which means it’s more powerful compared to models that work solely on-device. For comparison, the Apple Intelligence model has 3 billion parameters.
The internal tool, ‘Playground’, allows Apple to run benchmarks and said that its capability is ‘approaching the quality of recent ChatGPT rollouts’. Gurman added that the company executives were having ‘philosophical differences’ but did not mention any specifics. Also, the LLMs had a tendency to ‘hallucinate’, or provide incorrect answers. It’s not clear whether Apple will introduce or implement these LLMs in the near future.
