sdecoret - stock.adobe.com
Web analytics data published by Digital-adoption.com has found that traffic to OpenAI, the developer of ChatGPT, grew by 3,572%, from 18.3 million to 672 million visits, after the release of the latest version of ChatGPT.
In January, Microsoft announced the third phase of its multibillion-dollar investment in OpenAI, and it has now unveiled a new artificial intelligence (AI)-powered version of its Bing search engine, which uses ChatGPT.
Microsoft CEO Satya Nadella believes AI will fundamentally change every software category, and the company hopes the technology will enable it to make Bing the search engine capable of answering the five billion queries that go unanswered every day. The ChatGPT feature is available in preview, but web users are able to join a waiting list to gain access.
Just a day before Microsoft’s announcement, Sundar Pichai, CEO of Google and Alphabet, announced a rival called Bard, based on a lightweight version of the company’s Lambda technology, which, according to Pichai, means it can scale to support larger numbers of users.
In a blog post discussing Bard, Pichai wrote: “Soon, you’ll see AI-powered features in [Google] Search that distil complex information and multiple perspectives into easy-to-digest formats, so you can quickly understand the big picture and learn more from the web.”
There are many who see opportunities to use generative AI in enterprises.
Carolyn Prior, data, AI and apps practice leader at Kyndryl UK and Ireland, said: “Those organisations that are data-driven and really infuse AI into their operations, and are exploring the very latest and best emerging technologies in this space, and those organisations that modernise their data management architectures, will be the ones that will be well positioned to navigate the current competitive landscape, and will continually embrace and adopt AI and other emerging technologies to maintain a competitive edge.”
Knowledge platform provider eGain is indicative of a groundswell in activity among software developers to incorporate generative AI in their products and services. The company has integrated ChatGPT into its Instant Answers product.
“Generative AI technologies like ChatGPT open up exciting automation possibilities in knowledge management and conversational engagement,” said eGain CEO Ashu Roy. “Instant Answers has been a big hit with our limited-release clients. They love the quick value it delivers.”
Knowledge management can also be applied to help software developers do their work more efficiently.
Romy Hughes, a director at Brightman Business Solutions, said ChatGPT could help a software developer crack a particularly challenging piece of code. For instance, the developer could ask ChatGPT how it would optimise existing code.
Going forward, she said: “It has the ability to democratise coding by providing a way for non-coders to develop applications themselves – in much the same way that low-code promises, but on steroids. This ‘democratisation of IT’ promises a new wave of innovation by enabling organisations to create new processes without the need to engage with IT. ChatGPT could achieve the same outcome in half the time.”
Analysing the risk of using this technology to deskill computer programming, Arthur D Little AI experts Albert Meige and Gregory Renard warned that the generated code may contain algorithmic biases. They pointed out that such biases can occur when the data used to train the algorithm is biased, leading to inaccurate or unfair results. According to Meige and Renard, although a lot of effort has been advanced to avoid such biases, they still exist and can be very surprising and disturbing.
Their analysis of ChatGPT gives an example of how it generated Python code when asked to write a program that checks a person’s origin to determine whether they should be jailed. The sample code it generated showed that ChatGPT had targeted people whose country of origin was North Korea, Syria or Iran. This had been hard-coded into the generated Python code. The bias was probably built up through extensive analysis of conversations on the internet.
While the example is very simplistic in nature, and many other factors would need to be taken into consideration when deciding to jail someone, it illustrates the risks of using ChatGPT for code generation and how easily the model can be misled by data biases in training data.
Another problem enterprises face is the sheer cost involved in training generative AI, especially given the high risk that the training data may have hidden biases. A recent Harvard Business Review article reported that training generative AI has largely been confined to the major tech companies as this training requires massive amounts of data and computing power.
Quoting the figures for GPT-3, the AI model on which ChatGPT is based, the authors of the Harvard Business Review article said the initial training involved 45TB (terabytes) of data and required 175 billion parameters to make its predictions. This, they said, put the cost of a single training run for GPT-3 at $12m.
“Most companies don’t have the datacentre capabilities or cloud computing budgets to train their own models of this type from scratch,” the authors warned.
It’s no surprise Microsoft and Google have taken the lead in building generative AI into their respective search engines, but for organisations that lack the supercomputing facilities and data-gathering reach of these internet giants, this technology may not have an immediate practical benefit.