Is ChatGPT reaching its pinnacle? What is the likely impact for lawtech and insurtech?

Contacts: Richard West and Karim Derrick

Recent observations indicate a potential inflection point in the use of ChatGPT, the renowned generative AI technology. Despite its integration to empower Microsoft’s Bing search engine and the relentless buzz surrounding its prowess on social media, ChatGPT has experienced a decline of up to 10% in user engagement compared to last month’s usage.    

Amidst this perplexing trend lies a multitude of contributing factors that warrant deeper exploration. One noteworthy element is users’ discernment of a trade-off between the accelerated performance and diminished quality of ChatGPT’s responses. While it has become swifter in generating outputs, some users have voiced concerns over the declining standards of its responses.  

OpenAI, the driving force behind ChatGPT, has remained silent on the matter, which has sparked conjectures within the AI community. Among these speculations is the notion that there might be a concerted effort to channel AI models towards specific knowledge domains to achieve comparable results at reduced costs.  

Moreover, another issue looms on the horizon – one that has been anticipated for some time now: the emergence of generative AI as an AI echo chamber. Large language models like ChatGPT extensively scrape data from the public internet, yielding powerful outputs seemingly indistinguishable from human-generated content. However, as TechRadar reports, a substantial portion of online content is now produced by generative AI. Consequently, these language models are being trained on each other’s outputs, leading to a feedback loop, that could erode the overall quality of AI generated content. Though the impact may be marginal at present, the problem is expected to exacerbate over time. This phenomenon draws parallels with the social media realm, where algorithms amplifying popular content have inadvertently propagated biases and undesirable behavior, that has become the background noise of all social media use. 

This begs the question of whether the current approach to transformer-based large language models has reached its zenith. Recent development has primarily focused on scaling up models, with the number of parameters becoming the de facto measure of progress. However, there are indications that moving away from sheer size might be a more promising direction. OpenAI’s reluctance to disclose the training dataset for their models has raised concerns within the AI community, with suspicions of potential data protection breached prompting investigations across the EU.  

Nevertheless, large language models remain a permanent fixture in the AI landscape. As the hype cycle appears to plateau, attention shifts toward pragmatic applications. Channelling development efforts into refining data quality for AI training and incorporating methods to ensure control over inferences from AI with symbolic and rules-based systems holds tremendous potential. Legal and insurance services stand to benefit significantly from this approach, provided that safety and transparency are assured.  

At Kennedys IQ we have been at the forefront of deploying multiple hybrid large language model use cases, already demonstrating tangible value for our clients. The prospects of introducing these applications to the market in the coming weeks and months are promising. So, as we unveil the trajectory of ChatGPT and its implications, it is still evident that the future of AI in lawtech and insurtech holds boundless possibilities. Stay tuned for further developments and innovative breakthroughs. Exciting times lie ahead, stay tuned!