Since OpenAI’s release, industry giants have followed suit with announcements that generative AI is being integrated into Google Workspace, Microsoft Office 365 and Amazon Web Services. Apps are springing up everywhere.

It feels like we’re on the brink of something remarkable, not unlike Apple’s iPhone moment, especially given these models can be fine-tuned so they become industry, and company, specific. Experiments with Deloitte’s own data to generate reports and PowerPoint presentations, for example, have shown impressive results.

But just how good is this technology?

In the US, different iterations of the GPT family – OpenAI’s large language model – were applied to the three components of the Uniform Bar Exam, which is used in 36 states. This included the latest version, GPT4, which was released in March.

On multiple choice, GPT4 got 76%, which is better than the average student. The exam also consists of six essay questions and a practical task. Overall, the latest model scored 297, which is above the pass mark set by each state. Interestingly, this was zero-shot analysis without pre-training.

Today, most general counsels are under pressure to spend less on legal providers and headcount despite having more work than ever. For lawyers specifically, generative AI can help tackle that ‘more for less’ challenge. In the near future, huge swathes of tasks could be carried out differently and that will mean important productivity gains. The ability to generate good quality text, at a lower cost, could also see many more patents, for example. There are big societal applications too. According to the Organisation for Economic Co-operation and Development, fewer than half of the world’s population have realistic access to courts and lawyers. Publicly available algorithms could mean a fairer justice system if generative AI can offer more accurate answers to legal questions than a standard web search can now.

It's also important to note the difference between automation and innovation. Automation is about task substitution; innovation is how we use technology to do what currently isn’t possible. And that’s what’s really exciting.

Most computer scientists were not expecting this six months ago. If the future has arrived, it did so very quickly.

Conversations have already started around risk, trust and bias, and the confidentiality of company data, but generative AI is significant. Not for what it is now, but for what it will be. What we see today are prototypes; the technology will continue to advance at a rapid pace, offering better and more realistic interactions.

Hyper personalisation, or the ability to train models for individuals, will see them become true personal assistants, at home and in the workplace. Can they go from expressing knowledge to offering opinions and judgements? Undoubtedly.

Greater productivity, and the ability of large language models to produce code, will likely lead to fewer expensive failures for start-ups. That could mean a shift in how venture capitalists invest money, how we innovate and how we grow next-generation organisations.

But perhaps most importantly, this technology can be used to analyse vast amounts of data, and derive insights, on some of the world’s most pressing problems. And task automation will free up the time to dedicate to it. Solving climate change, broader sustainability, better education and skills – this could be where generative AI’s greatest value really lies.