Artificial intelligence can sometimes be impressively smart, but it also has its moments of being not so sharp when it misses obvious details. Apple’s researchers have taken a step forward in addressing this challenge.
The researchers have discussed a feature of large language models that helps them understand what is displayed on a screen, known as reference resolution. This capability is outlined in a paper that emphasizes Apple’s commitment to staying competitive and enhancing Siri, its virtual voice assistant.
The study underscores the importance of using this technology to make assistants more adept at engaging in natural conversations. By grasping the nuances and context of the content on the screen, assistants like Siri can provide responses that feel more human-like.
The ability to decipher ambiguous references in human speech, such as ‘they’ or ‘that,’ based on context is crucial for a conversational assistant like Siri. This understanding is vital for users to communicate naturally with the assistant or engage it in a dialogue.
Apple’s focus on benchmarking against rivals such as GPT-3.5 and GPT-4 is apparent in the research. The study revealed that Apple’s models not only matched GPT-4 with its smallest version but also surpassed it significantly with its larger models.
This aligns with Apple’s recent comments on AI during an earnings call, where CEO Tim Cook hinted at exciting AI features to be unveiled later in the year. This cautious approach of doing work before talking about it has been consistent with Apple’s strategy.
Apple has already begun to fulfill this commitment by highlighting AI features in the new M3-chip-powered MacBook Air release. This move signifies the company’s focus on integrating AI technology into its products.
The research conducted by Apple’s team likely represents just the beginning of the company’s AI initiatives. It showcases Apple’s dedication to advancing AI capabilities to enhance user experiences and stay competitive in the market.