AI: The next frontier
Inflexion’s latest Portfolio Exchange took place 12 months after the launch of ChatGPT. With 1m sign-ups in the first five days, 100m active users by January 2023 – and now c.180m users per month, it has been one of the most successful product launches of all time. The hype is justified, according to an expert panel hosted by Inflexion Partner Alex Mathers.
Dramatic developments in recent months have brought AI further to the fore. “Whilst there will no doubt be lots of developments in the near future, it will be game-changing in the long-term,” Alex enthuses. Already using it within the firm and across the Inflexion portfolio, he feels the future is already here.
Observations were shared and discussed by a panel of professionals which included Chris Stibbs, Chair of Times Higher Education, a rankings and data business backed by Inflexion; Leonid Zhukov, VP of Data Science at BCG-X and Director of BCG Global AI Institute; and Jan Beitner, Assistant Director at Inflexion focused on Data and AI.
AI is changing fast and can now address multiple issues and solve specific problems.
“We have all have seen ChatGPT in action, with generative AI able to create instances of almost any modality responding to the user prompt (query),” says Leonid. The model generates the results based on the patterns it learned from the training data. Lots of data is needed to train the algorithms, and ChatGPT is currently drawing on two decades of internet usage. “What is less in the media is what’s happening on the tech side and what data scientists do with the models,” points out Leonid. It used to be that data scientists would use one model for one problem but recently the concept has changed. “We have a new type of models, called foundation models, which has capabilities to solve multiple problems. These models come already pre-trained and with much smaller effort and you can adapt or fine-tune them to solve a very particular problem you have,” he enthuses, adding it really changes how data scientists are working. In particular, far less data is needed to adapt these models to company-specific tasks than previously, allowing mid-market companies to fully benefit from this new technology.
AI is not just about efficiencies but also about product enhancement.
Chris has seen very practical applications for it, including a data business which has replaced its website’s traditional keyword search function with something they built on ChatGPT to allow for intuitive site navigation. “You can have an interactive discussion with ChatGPT on how to find what you want on that website. It’s truly a game changer yet was built in-house,” he explains.
Another application that stands out for him is for an English language school. Traditionally they catered to the needs of those with high proficiency in language, and customers/students get a certificate at the end of it, making them different to apps like Duolingo that cater for more causal learners. The launch of Duolingo Max, however, has blurred the boundaries by incorporating ChatGPT to allow for interactive role play. Says Chris: “It’s taken a product to a new level, bringing it close to the experience of interacting with a real human being, opening up a new client base to Duolingo as it can now deliver a high proficiency service.”
If you don’t have a modern tech stack, it’ll be hard to make the most of AI.
But if you do, the integration with AI should be fairly painless with virtually no new skills needed as advanced data science skills are not required, according to Jan. “Our portfolio is going down multiple routes, with some simply buying new software like Github Copilot to boost coding productivity by c10-20%.” This, he is quick to point out, is a pretty massive boost for $20pcm. Others are additionally using chatbots and some are doing more bespoke things, such as automation. Chambers, a B2B information services provider in the legal sector, illustrates this, with Inflexion supporting significant investment into the firm’s research function including AI to achieve real productivity increases.
Humans still have the edge over AI for logic.
Leonid shared an experiment from BCG on this, which allowed people to do an exercise with and without ChatGPT: where creativity was needed, ChatGPT outperformed the quality of solutions by 40%. But when analytical thinking and logic were required, the AI underperformed. “It’s because the tech isn’t yet capable of that. It’s currently helpful with creative tasks, like marketing, but lacks understanding for logical things so we need to be careful where we use it and what skills it can enhance.” He caveats his sentiment with a reminder that this is today’s frontier and it moves extremely fast; in a year or even a month it may have evolved. “For now, it’s good to have the AI models as a brainstorming partner and an efficiency booster rather than blindly trusting them. It’s co-pilot and a helper rather than a substitute for humans,” Leonid suggests.
AI is more than a product disruption; it’s a full business-model disruptor – and that’s a much bigger challenge.
“It’s exciting but also scary times,” Chris cautions, speaking from his 25+ years of experience in the publishing industry. He reminds the Exchange that publishing has been scarred from previous disruptions. He reminds guests how the internet slashed the price of advertising as everything could be published everywhere, and it destroyed the margins of the entire print industry. “Those who survived were those whose USP was their brand and uniqueness. It’s The Economist, NY Times, The Times, who flipped their models from advertising-led to content-led. Content creators need to be able to define their USP and build and protect it.” Using AI purely for efficiency gains is short-sighted he says, suggesting that a laser focus on USP can make people and products future-proof. He also suggests ensuring OpenAI knows it cannot use your openly, on the web available information to train models like ChatGPT “Any upcoming legislation will be relevant for the future. What’s going into ChatGPT now and has gone in before is already out of the bag.” However, using your proprietary data with the enterprise version of ChatGPT is not an issue, as they do not retain your data. “If you trust Microsoft with your data on Sharepoint, you can trust it with your data being fed to GPT-4", Jan adds.
While the future remains to be seen, what is clear for now is that AI is here to stay. Concludes Leonid: “The disruption is here so you can’t hide. Experiment with it, see it as a tool, but be mindful of its future strength as it will disrupt all business models.”