ChatGPT: an “iPhone moment” – but what next?

How ChatGPT has “lit the fire” for generative AI

The release of ChatGPT was an “iPhone moment”, signalling a breakthrough in revolutionary AI models, a former Intercom executive told Third Bridge Forum. ChatGPT, the latest step in OpenAI’s deployment of “increasingly safe and useful AI systems”, is seen by many as the most tangible example yet of how AI will transform businesses and society on an unimaginable scale.

“It was absolutely an iPhone moment and things will never be the same again. The world has changed and will change even more in the future in ways we probably can’t even imagine today.” Former executive at Intercom

Despite the fanfare, only a handful of jobs can currently be fully replaced by generative AI, primarily because of its limitations around accuracy and logical reasoning. But advancements in machine learning, computing power and data storage are fuelling rapid growth in this field. The COVID-19 pandemic also fundamentally changed society’s relationship with technology. Experts we have spoken to say the enterprise chatbot industry has accelerated by a decade in just three years, with many more players expected to disrupt the space over the short to medium term. Using AI programmes such as ChatGPT as personal assistants could soon be the norm – leaving humans to retain the lead in complex skills such as public speaking, team management and critical thinking.

It is also important to recognise that virtual assistants and chatbots – notably Siri, Alexa and OK Google – are already an accepted part of our lives. However, they have not garnered anywhere near the level of enthusiasm and excitement that ChatGPT has, a former manager at OpenAI told Forum. What differentiates ChatGPT, according to the specialist, is that no activation is required. “ChatGPT is one of the first releases to the general public of what OpenAI has been up to with its AI models and what it’s capable of.” As the former Intercom executive said, it “lit” people’s imagination. “Immediately, you could imagine how you could use this, and that is part of what also lit the fire for generative AI.”

Major ChatGPT enhancements are already underway. In January this year, the former OpenAI manager we spoke to anticipated ChatGPT moving to GPT-4 with 100 trillion parameters – a significant leap from GPT-3’s 175 billion parameters – within months of its first release. Less than four months later, GPT- 4 is already available to certain OpenAI users. And there is no sign of a slowdown. We learned that regular users of ChatGPT should notice significant improvements every 3-6 months and that positive feedback on GPT-4 has been building for some time. Answers are expected to become increasingly accurate and concise as the technology evolves, eventually with voice matching capabilities.

Which industries might generative AI break into first? Customer support tops the list of areas that are ripe for disruption and it is “coming fast”. Currently, AI can answer around 30% of customer service questions, but a specialist we interviewed believes that number could grow to 60-80% in the short term. This means the effectiveness of each customer service representative could “double, and then double again, and soon double again.” Across Forum Interviews, we have heard that marketing, sales, financial services, gaming and healthcare are among the segments with the greatest potential for transformation.

How ChatGPT is reshaping the competitive landscape

What we have seen so far with ChatGPT is merely the tip of the iceberg for generative AI. Our experts agree it is reasonable to expect an explosion of new businesses and products as developments blossom. Hot on the heels of ChatGPT-4’s release, Google launched its Bard chatbot in March as it seemingly “seeks to make up lost ground”.¹https://www.ft.com/content/c9e97913-bc88-4845-a5b4-fe8f6b813274

Rather than competing directly with ChatGPT – particularly with OpenAI already reducing its pricing – many other companies are expected to focus on building layers on top of ChatGPT to create customised models. But to be successful, we were told players will need to truly add value to the core ChatGPT infrastructure – offering something that is unique and perhaps an entirely new use case, thereby building a durable and structural advantage.

As we know, one of ChatGPT’s biggest limitations currently is that it is only based on public data captured through 2021. Nonetheless, many companies will also expect to train on their own datasets, and incorporate their own tone and voice. We have heard this will be a “rich” area for software vendors as they carve out their niche. Interviews suggest bespoke content will become one of the key differentiators that enterprises will seek out vs general-purpose generative AI. HubSpot, Intercom, Salesforce and Genesys, as well as newer entrants like Jasper and Typeface, are already making moves here.

ChatGPT has also seemingly sparked a race among Big Tech companies as they ramp up their own AI efforts. Microsoft’s investment in OpenAI/ChatGPT has been at the forefront of discussions on this, notably regarding potential threats to Google. Google reportedly declared “code red” over fears competitors may eat into its search business.²https://www.nytimes.com/2023/01/20/technology/google-chatgpt-artificial-intelligence.html%20 But some experts have cast doubt over the likelihood of this – at least for now. The former OpenAI manager we spoke to believes that although ChatGPT will significantly enhance Microsoft’s AI capabilities across multiple offerings, Google will not necessarily lose out if it maintains momentum in this burgeoning field. Similarly, a former senior director at Google said the partnership arguably presents the biggest threat to Google Search in recent memory, but overall sees it as only “hype” so far. They also suggested it could take years for Bing and ChatGPT to achieve the perception of “trustworthiness” that Google enjoys.

More broadly though, OpenAI may have an inherent edge compared to the tech behemoths. For example, according to one expert, Google currently lacks a clear strategic direction whereas other players are focusing on model application and revenue generation in vertical fields. “This is fundamental and the biggest difference between OpenAI and the three giants, namely, Google, Meta and Microsoft.”

At a high level, generative AI is “still in the era of 1,000 flowers blooming”, but experts say consolidation could begin in as little as 1-3 years. One specialist noted how Google and Microsoft’s internet and computing breakthroughs changed the world in ways that no one anticipated. “This is the level of companies that will be created, so I expect to see some new behemoth, massive, massive companies created over the course of the next decade.”

“I expect to see some new behemoth, massive, massive companies created over the course of the next decade.” Former executive at Intercom

Companies in adjacent areas such as chips and data storage are also critical to the continued development of generative AI. We heard Nvidia is “particularly well positioned” and Intel and IBM were highlighted as potential beneficiaries too.

China’s position following ChatGPT’s release

As a country that is ranked among the top three countries for AI “vibrancy”, many have wondered how China will respond to the launch of ChatGPT.³https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-next-frontier-for-ai-in-china-could-add-600-billion-to-its-economy A former NLP executive estimated it could take domestic Chinese players 2-3 years to develop models that are 80% as powerful as ChatGPT. “Compared with GPT-2, GPT-3 hasn’t made a big technological breakthrough in algorithms and mainly focuses on the improvement of data volume and quality.”

Discussing the technical entry barriers to ChatGPT, three key challenges were highlighted. First and foremost, unlike the English speaking internet world, China has less open source data and lower data quality due to preliminary data cleansing. If domestic companies intend to develop an equivalent of ChatGPT, the expert believes the biggest barrier lies in data. To emphasise this point, the specialist suggested that Chinese data may represent only 2-5% of ChatGPT’s training data, and is the biggest reason why ChatGPT performs better in English than in Chinese.

Secondly, training large models such as GPT-3 typically requires the efficient use of “hundreds of servers or over 1,000 GPUs” – though less of an obstacle for companies that have already trained models with hundreds of billions of parameters, including Baidu, Alibaba, Tencent, Huawei, Beijing Academy of Artificial Intelligence and Inspur. US chip export restrictions could also impact engineering and computing in 12-18 months after China’s Nvidia inventory starts to dry up. We heard the gap between China and the US in computing and engineering is currently insignificant.

Lastly, although information about InstructGPT – ChatGPT’s underlying technology – can be learned from, there may be some hidden “tricks” known only by OpenAI related to fine- tuning, optimisers and algorithms. Funding is also critical, with the cost of training ChatGPT potentially amounting to USD 3m per use case. In terms of technical ability, however, the US is not worlds ahead of China. So far, China’s leading NLP companies include Inspur, Zhiyuan Community and Huawei, according to the specialist. Baidu is particularly “promising” owing to its engineering and data capabilities, and like OpenAI is expected to follow the decoder-only route with manual fine-tuning. “Baidu has launched some open- source models whose codes are not open- source. It also has announced in its paper an ERNIE (enhanced representation through knowledge integration) model.” However, the expert does not have high expectations at this stage for Baidu’s Ernie Bot given the myriad challenges still facing most players.

Get comfortable becoming a “prompt engineer”

Generative AI has been driving conversations in the tech community for years, but as we have heard, with relatively limited commercial progress until 2023. Specialists are impressed by the quality of ChatGPT so far, but the full scope of its impact remains unknown – as do the potential risks. For example, a “whole new layer” of governance will be required as individuals adopt personal assistants and the data landscape becomes even more complex. Likewise, data privacy and security will undoubtedly require a new compliance framework.

What we have heard, however, is that based on the arrival of ChatGPT the “best and brightest” software companies are already re-prioritising their strategic goals and figuring out how to become “AI-first” companies. Our experts say that any company that fails to recognise and act on this shift risks being replaced by next-generation companies who are repositioning for a once-in-a-generation wave of technological transformation.

“The best and brightest software companies are dropping what they’re doing and re-prioritising based on this and figuring out how they can become AI-first companies.” – Former executive at Intercom

Although there understandably remains a significant degree of apprehension – even scepticism – about what really is next after ChatGPT, we were told individuals and businesses should “embrace this change”. The former NLP executive noted that ChatGPT’s performance depends on high-quality prompts or instructions, meaning its output will only be as good as its input. Being able to get the best out of ChatGPT is seemingly becoming a highly sought-after skill in itself, with the former Intercom executive advising everyone to “get comfortable being a prompt engineer”.

With the technology world now in flux, our experts exhorted all companies to consider how AI could change or support their business model. One expert encouraged us all to leverage the broad power and creativity of the organisation to “think about what is possible”.