According to Nvidia CEO Jensen Huang, human-level artificial intelligence (AI) may be achieved within the next five years, with hallucination, a significant obstacle in the field, potentially easy to address.
Huang made these remarks during his speech at the Nvidia GTC developers conference in San Jose, California, on March 20.
In his keynote, Huang discussed artificial general intelligence (AGI) and emphasised the importance of benchmarking in its development. He suggested that defining AGI through specific tests, where software outperforms humans by a significant margin, could lead to its realisation within five years.
Regarding “hallucinations,” which arise as unintended consequences of training large language models, Huang proposed a straightforward solution: requiring AI to verify answers by conducting research before providing responses. This approach aims to mitigate the issue of AI generating inaccurate information not present in its dataset.
While various AI models already offer features to provide sources for their outputs, such as Microsoft’s CoPilot AI, Google’s Gemini, OpenAI’s ChatGPT, and Anthropic’s Claude 3, fully resolving the hallucination problem could have profound implications for industries like finance and cryptocurrency.
At the present, caution is essential when using generative AI systems in contexts where accuracy is critical. For instance, ChatGPT’s user interface warns users about potential mistakes and advises cross-checking important information.
In finance and cryptocurrency, accuracy is paramount, as it directly impacts profits and losses. Hence, the current reliance on generative AI systems is limited in these fields.
Although experiments with trading bots powered by generative AI exist, they are often restricted by predefined rules to prevent autonomous trading. If hallucination issues were fully resolved, these AI models could potentially operate independently, executing trades and making financial decisions without human intervention. Thus, solving the problem of AI hallucinations could pave the way for fully automated trading systems.