Rodney Brooks on Why We Should Calm Down About GPT

Large language models have taken the internet by storm. Their ability to generate human-like text and engage in conversations has fascinated people in all domains. But, it is crucial to approach Large Language Models(LLM) with caution and understand their limitations.

In a recent interview with IEEE Spectrum, Rodney Brooks shared his views on the furor over large language models.

?? Rodney Brooks is a renowned roboticist, author, and entrepreneur.

I read the rather lengthy interview so that you don’t have to! Read on to know Brooks's thoughts about GPT & LLMs.

On GPT-4 and Performance vs Competence

Rodney Brooks uses an interesting example to contrast performance and competence in AI. He talks about how a large language model, like GPT-4, can confidently answer a question about a specific topic. But when probed further, the LLM lacks the logical inference that a human would have. For instance, when you ask it a complex coding problem, it might give you an answer that sounds confident and correct. But, upon implementation, you might find that the solution does not work.

According to Brooks, this demonstrates the model's performance, but not its competence. Brooks emphasizes that these models may perform well, but they lack an underlying model of the world. So they are merely correlating language. Their lack of real-world connection and understanding is a significant limitation. Brooks believes that future iterations like GPT-5 or GPT-6 will not progress on these issues significantly.

On Large Language Models (LLMs) and Company Valuations

Next, Brooks discusses the enormous valuations of companies marketing large language models. For instance, OpenAI was reportedly valued at almost $30 billion. He is skeptical about the high valuations based on the applications of these models.

He thinks these models will make language interfaces smoother and improve language input and output. Yet, he doesn't believe them to completely change the field or bring about artificial general intelligence (AGI). He sees them as helpful tools, but not groundbreaking.

On AI and Employment

Brooks addresses the fear of AI eliminating entire classes of employment. He mentions the current shortage of radiologists and truck drivers is due to premature predictions of AI taking over these jobs. He argues that these predictions are often based on hype and a lack of understanding of AI's capabilities. While AI can automate certain tasks, it's far from replacing human competence in many fields. He warns against the potential real-world consequences of such premature predictions. Job shortages in certain professions are just one of those consequences.

On Self-Driving Cars

Brooks talks about the excitement over self-driving cars and how it has created unrealistic hopes. He says that even though people predicted it long ago, fully self-driving cars (level 5) are still a long way off.

He thinks that while level 2 and level 3 autonomous features in cars have gotten better, reaching level 5 is much harder. He also mentions how the hype around self-driving cars has slowed the development of infrastructure for them.

On Warehouse Robotics

Brooks talks about his current work in warehouse robotics. He mentions that 80% of warehouses in the U.S. have zero automation, and only 5% are heavily automated. He sees a significant opportunity for robotics in this space.

He emphasizes the need for robots that can work alongside humans in these warehouses. His company, Robust.AI, is working on creating such human-centered robots. His company has designed robots that are aware of people and treat them with respect. And in case of a Robot doing a dumb thing, humans can easily take control.

On Technological Initiatives

Brooks is hopeful about indoor farming's potential to address climate change. He sees it as a game-changing technology that can reduce the environmental impact of traditional farming.

By using technology, genetic engineering, and machine learning, indoor farming could provide a sustainable food supply. Brooks finds this development promising and it gives him hope for the future.

Conclusion

LLMs like ChatGPT are powerful, but they still have a long way to go. As OpenAI CEO Sam Altman tweeted, "ChatGPT is incredibly limited but good enough at some things to create a misleading impression of greatness. It's a mistake to be relying on it for anything important right now. It’s a preview of progress; we have lots of work to do on robustness and truthfulness."