Since its inception, ChatGPT, has garnered extensive attention. Within a mere half-year of its debut, an impressive 100 million users had adopted the platform, with many happily trading the cost of a monthly Netflix subscription for the advanced capabilities of the ChatGPT Plus version. Yet, a rising tide of unease seems to have gripped the AI community. Is ChatGPT, in fact, getting slower? This examination seeks to delve into this evolving narrative.
OpenAI’s ChatGPT has enjoyed rapid adoption, underpinned by its promise of mimicking human-like interactions. Its potential to even rival search engine titans like Google was fervently debated in the tech community. However, recent concerns have bubbled to the surface about the tool’s diminishing speed and the reliability of its responses.
Research from esteemed institutions, Stanford and UC Berkeley, paints a concerning portrait of ChatGPT’s evolving competency. In a span of merely three months between March and June 2023, the platform’s proficiency in tasks such as solving mathematics problems or generating code faced a marked downturn. The most alarming was the AI’s capability in identifying prime numbers, which plunged from a commendable 97.6% accuracy rate in March to a worrying 2.4% by June.
Many have taken to social platforms, voicing their discontent over the perceived decline in GPT-4’s performance compared to its predecessor, GPT 3.5. Santiago Valderrama, an AI luminary, didn’t mince words when suggesting that GPT-4’s performance seemed to be on a downward trajectory rather than showing signs of improvement.
Theories abound. Some suggest that in a bid to reduce operational costs, OpenAI could be deploying several smaller, specialised GPT-4 models. While these might mimic the behaviour of a more comprehensive model, they likely do so at the expense of performance.
As ChatGPT’s adoption skyrocketed, whispers grew louder about a potential reduction in processing power allocated per user. Perhaps it was the off-the-record remarks by OpenAI’s CEO, Sam Altman, that gave these whispers a more solid form. His comments regarding struggles in procuring sufficient GPUs and the resulting expansion delays were not meant for the public. However, in our age of digital connectivity, such secrets rarely remain hidden.
It’s speculated that the burgeoning popularity of ChatGPT could be its Achilles’ heel. As more users flock to the platform, it’s conceivable that the system could be facing a ‘bad neighbour effect,’ causing slowdowns due to overwhelming demand.
Following the rise of ChatGPT, tech giants like Microsoft and Google have been swift to weave AI-powered tools into their offerings. Yet, recent figures suggest a potential cooling in the AI fervour. Data reveals that downloads for ChatGPT and Microsoft’s Bing have witnessed a slump. In June, ChatGPT downloads on iPhones in the U.S. contracted by 38%, while web traffic to the platform receded by 9.7%.
ChatGPT has unarguably changed the AI landscape. But, as with many pioneering technologies, it’s not without its teething problems. While AI holds vast potential, challenges around scalability, cost, and consistent performance loom large. As the path ahead unfolds, it remains to be seen whether ChatGPT can regain its initial promise or if it becomes a salutary lesson in the delicate balance of innovation, demand, and consistent delivery.
If you want to chat to some humans about your IT needs, then please do get in touch, we’d be happy to help: Contact Us
Photo by Andrew Neel
< Back to all news