GPT-4o and the future of Call Centres

GPT-4o is out this morning, and it is a huge step forward technically (although arguably not societally) for capabilities largely outside of the headlines.

The biggest impact comes from an ability to engage in continuous, interruptible speech. This makes AI customer service operatives possible - and on a massive scale.

The limitations of previous LLMs became clear when building the first version of my language learning tool "Chateau"; to engage in continuous speech requires a conversation to be seamless and in real time. The necessity of waiting to record a complete sentence stops "butting in" and the time taken to respond has thus far been simply too long.

So have OpenAI done this? Last night's live demo seemed to show they had but the proof will be in the API pudding, so until available we will have to see.

What is clear is that this is coming, and an LLM that has been fine-tuned on an entire business is going to be considerably more useful than a human operator in 80% of situations. As well as 24/7 coverage and deep knowledge of all potential problem sets, it also means the end of wait queues, since an infinite number of operators can be spun up to meet demand. Unbelievable as it may sound now, I predict that once consumers get used to it they will insist upon a fully knowledgable AI over a human in almost all situations.

The societal ramifications of this are of course profound, particularly for the UK's 70,000+ call centre staff and legions of administrators who respond to telephone queries.

The benefits are also profound particularly in healthcare with the ability to run a scalable, highly knowledgable 111 service that frees up medical staff and reduces costs, easing the burden on the NHS.

Universal access to personalised AI tutors will similarly both help all learners, but impact teachers.

This is not just OpenAI. Where ChatGPT goes LLaMA, Mistral, Gemini et all will follow, which leads to this final thought: to paraphrase Shadbolt and Hampson, before anyone starts talking, a thing should always say what it is and be what it is.