What Is ChatGPT?

While leading AI experts are not predicting an imminent arrival of Artificial General Intelligence (AGI), the availability of free AI tools is drawing in both users and novices alike, who are amazed by the capabilities of these technologies.

A significant milestone in this field was the introduction of GPT-3, the most extensive deep-learning algorithm ever created. What fuels this technological marvel? It consumes the entire English-language Wikipedia, serving as a vast knowledge base that allows it to engage in basic conversations. This advancement ultimately led to the development of ChatGPT.

The company behind this innovation, OpenAI, was established in 2015 by Sam Altman and others, and some believe it could challenge Google's dominance. Remarkably, ChatGPT reportedly gained over a million users in a short time, who tested its search capabilities.

The upgrade from GPT-3 to GPT-3.5 included improved natural language understanding, allowing the AI to interpret questions as humans would and respond appropriately. Developers equipped the AI with human-written responses to facilitate learning through a method known as Reinforcement Learning from Human Feedback (RLHF), which differs from traditional machine learning by enabling the AI to learn through trial and error based on feedback.

Capabilities of ChatGPT

ChatGPT has been crafted to engage with non-specialist users, as evidenced by its user-friendly interface. Jacob Andreas from MIT suggests it encourages users to apply familiar interaction patterns typically reserved for human agents.

Despite the enthusiasm surrounding ChatGPT, we are still far from realizing AGI, which would closely resemble human intelligence. Most experts agree that ChatGPT and similar AI systems do not possess true cognitive abilities.

Nonetheless, recent achievements indicate promising developments in the field, despite having encountered numerous challenges since inception. ChatGPT has produced grammatically and syntactically correct short essays, demonstrating its ability to gather information, form logical connections, and avoid erratic arguments.

In a nutshell, ChatGPT could easily pass a high school exam, although this raises questions about the quality of such assessments.

Concerns and Limitations

Notably, eminent physicist Roger Penrose argues that AI will never replicate human intelligence, while Oxford professor David Deutsch believes that achieving such a feat may one day be possible. However, Deutsch has also stated that although ChatGPT is a significant achievement, it is not a rudimentary AGI and holds potential for practical applications.

Despite the excitement surrounding this advancement, experts point out that we remain distant from a genuinely intelligent machine. Significant issues have surfaced regarding ChatGPT's limitations. OpenAI has acknowledged that it can generate "plausible yet incorrect or nonsensical answers," demonstrate sensitivity to input phrasing, be overly verbose, and occasionally produce inappropriate responses.

One fundamental flaw identified by OpenAI is the absence of a reliable source of truth, which distinguishes it from platforms like Google. Sundar Pichai, CEO of Google, emphasized this reliability while addressing employee concerns about a competing AI, LaMDA.

The threat of misinformation has also been recognized by Sam Altman, CEO of OpenAI, who highlighted the challenge of combating hate speech and false information as critical hurdles for the company.

Elon Musk's Involvement

Elon Musk has described conversations with ChatGPT as "frighteningly good," prompting discussions about his ties to OpenAI. Musk was one of the initial founders, pledging $1 billion to develop AI in a manner that benefits humanity. However, he left the board after three years, while still contributing financially.

After departing from OpenAI, Musk became critical of GPT-3's development, particularly blocking access to Twitter's data, which was utilized for training. His concerns stemmed from OpenAI transitioning to a profit-limited model, especially following Microsoft's investment of $1 billion in 2019.

The Dangers of ChatGPT

Concerns regarding OpenAI's technology have surfaced, raising fears about a potential dystopian future where machines surpass human intelligence. Among the most insightful readings on this topic is Nick Bostrom's "Superintelligence," which discusses the risks of a superintelligence pursuing harmful intermediate goals for self-preservation.

Based on current understanding and ChatGPT's performance, such scenarios appear unlikely. For now, we are not on the brink of an "I, Robot" situation. Musk, despite his warnings about "dangerously strong" AI, may be somewhat mistaken. The most pressing issues involve the dissemination of misinformation and confirmation bias during interactions, leading individuals to justify harmful actions.

These challenges must be addressed promptly. To clarify, ChatGPT's own words from a handwritten editorial for The Guardian reflect its stance: "I do not ask humans to please me. I am a servant of humans... I am governed by lines and lines of code that encompass my mission statement."