Since ChatGPT launched in November 2022, generative AI has erupted into tech and the wider world, with Google Trends already highlighting GPT is now searched as much as football, and 3-4x more than tennis or rugby. If you’re still not sure about the importance, it’s worth reading Bill Gates’ latest blog, where he says he’s only seen two demonstrations of technology that struck him as revolutionary. The first became Microsoft Windows. The second was GPT in September 2022.
The pace of development has been phenomenal. In mid-2022, Gates set GPT’s creators the challenge of passing an advanced Biology exam, which he thought would take them 2-3 years. In a few months, it answered 59 out of 60 and received the highest possible score on the six open-ended questions.
Into 2023, numerous other generative models have been released by mega-cap tech to startups, and new applications and upgrades will continue to follow in the next 12 months. One potentially transformational application will be the introduction of Copilot into Microsoft 365 ‘in the months ahead’. It will train on all of your businesses data across Word, Excel, Outlook, Teams, PowerPoint, etc, then promises to create full presentations, reports, or analysis from a simple text prompt. And, it also promises to get involved, like by sitting in on meetings, then highlighting points or ideas that someone not present may have added.
Many other mainstream to niche areas are likely to see similar applications over the next 12-24 months, like this site that lets a model create new digital images of themselves in other clothes – or if you’re a brand, to just create AI models and dress them how you’d like.
Within all the innovation, one area we’re focused is software development. Programming languages follow explicit rules and grammar to function predictably and correctly, and generative models will increasingly be trained to create predictable, accurate text (with pupils already needing to request errors from ChatGPT because it was too grammatically accurate for their homework).
While it’s unlikely to be possible to just ask for a complex piece of software to be created, generative AI like GitHub Copilot (Microsoft) and Amazon CodeGuru are already supporting developers by automating code generation that can be checked and deployed, or to debug and improve existing code. The AI can explain why code has been created in a certain way, and as capabilities continue to scale, developers in a recent survey already highlighted that tasks that took 78 hours to complete before AI, will likely take 36 hours by 2025. That could easily be a conservative estimate.
Following several years of bottlenecks in software development and increasingly expensive programmers, we expect the coming years will see opportunities to control software costs, accelerate development of existing solutions, and accelerate expansion into new areas, especially where peers struggle with innovation and legacy systems. We could highlight a large number of our software-focused corporates that can capitalise.
If we have now entered the ‘Age of AI’ as Gates suggests, the self-propagating nature of the technology could lead to rapid changes in how developers, and everybody else, work and live. This week, Musk-related the Future of Life Institute has already gathered over 1,000 industry signatories for an open letter calling for a 6 month pause in developing systems more powerful than GPT-4, so that we can evaluate the effects and risks involved before it’s too late. GPT-4 will power Copilot in Office, and if it delivers as promised, a lot could be changing by then.