While big competitors like Microsoft or Google are spending a lot of money and presenting the first products that use artificial intelligence based on large language models (Large Language Models, LLMs), little or nothing is heard from Apple in the context of the current hype about ChatGPT. This also seems to be due to the fact that the iPhone group’s AI department is still crunching massively.
Google-CEO vs. Apple-CEO
Since last autumn, there have been several departures of important employees and the overall strategy is apparently also lacking. This is reported by the IT trade magazine “The Information” from Silicon Valley. Accordingly, Apple AI boss John Giannandrea proposes with “organizational dysfunction” hereabouts. It remains unclear whether this can be resolved. But Apple urgently needs to do something: Compared to ChatGPT, Apple’s voice assistant Siri seems outdated – and a “Siri 2.0” is a long time coming.
Most recently, three well-known AI experts — Srinivasan Venkatachary, Steven Baker, and Anand Shukla — left Apple and went to Google. This happened back in the fall because they thought the search giant was “a better place to work on LLMs,” according to The Information. Alphabet CEO Sundar Pichai is said to have tried to attract the three Apple employees himself. At the same time, Apple boss Tim Cook tried to keep her from resigning. Now they’re working on Google’s efforts to reduce the cost of training LLMs and improve their accuracy — and products based on Google’s models.
Apple urgently needs to catch up
For years, Apple’s AI space has been slow to babbling along. When it comes to operation, Siri has long been considered to be no longer up to date – since other assistance systems such as Alexa or Google Assistant also had one or the other problem, this was less noticeable. So you have to know Siri commands well to use the voice assistant; real dialogue using previous announcements is almost impossible. ChatGPT, on the other hand, is extremely articulate and the companies are currently in the process of integrating such systems into their assistants and scoring points with their own chatbots. So far there is nothing to be seen from Apple here.
According to sources with whom The Information was able to speak, the Siri department has “turned into a mess characterized by petty turf wars between senior executives and heated arguments over direction.” Data generated by Siri and coming from the data science and engineering departments is simply not used for further training. “It’s a waste of time and money,” said a former Apple employee. There are still issues with decision making and the general approach to AI. Apple is rated as “too conservative” here.
For data protection reasons, the group likes to outsource AI applications to user devices and not to the cloud, so it can react less quickly than its competitors. Siri Responses would be written by a “team of 20 authors” rather than being generated by an LLM. Cook and his management team are said to have specifically ordered this so that Siri would not generate “embarrassing answers”. A project to completely rebuild Siri, which should start in 2019, was rejected internally. Among other things, it should lead to a lightweight variant of the system being able to be run locally. Although more requests are now executed locally, the weak basic technology remained.