Yahoo Finance OpenAI’s Q* could become 'better than any human for certain activities', says expert
Speculation abounds about OpenAI's possible artificial intelligence (AI) breakthrough, called Q*.
Details about this possible breakthrough, pronounced "q-star", hit news headlines in November when OpenAI CEO Sam Altman was fired, and then re-hired, by the San Francisco-based firm's board.
Yahoo Finance UK sat down with Xenesis founder and chief scientist Tirath Virdee to discuss what we know so far about Q*, and how it could impact the development of AI systems.
Read more: Moody's building AI model for compliance checks
"With Q* I think we could get to a stage where it is better than any human when it comes to many of the specific activities we do," Virdee said.
The UK government advisor on artificial intelligence added that the Q* development could give AI the ability to have "semantic continuity" when constructing language. Semantic continuity could allow AI systems to interact more meaningfully with human users.
Read more: Google 'democratising AI' for non-expert app creators
However, Virdee stressed that this does not mean Q* has reached the level of artificial general intelligence, or AGI, which refers to highly autonomous system that possesses the ability to understand, learn, and apply knowledge across diverse tasks with a level of proficiency comparable to a human.
"Even considering the recent things that have been happening at OpenAI, I think we are still nowhere near developing artificial general intelligence. However, we are making significant strides towards it, and I am thinking that around 2050 we will have developed AGI," Virdee added.
Limitations of current AI systems
Virdee then outlined the limitations of current large language models (LLMs) such as ChatGPT and discussed how the Q* development might overcome these challenges.
"What ChatGPT does is only next token prediction. It takes all the documents that have ever been written and placed online, then it works out what the probability of what the next word in a sentence is going to be," he said.
The data scientist stressed that this has "obvious limitations because everything is based on probability and it will give you a different output every time."
Virdee then explained that this predictive method based on large sets of data is one of the reasons why LLMs, like ChatGPT, are getting mathematics wrong much of the time.
In contrast, he characterised the Q* breakthrough as a new foundational model for AI that holds the promise of advancing these systems beyond the current capabilities of generative AI systems.
Read more: The next generative AI success story could come from the call centre market
"Many people hypothesise that Q* is artificial intelligence learning to do maths. The implications of AI learning maths are so awesome. This is because you end up being able to build the ability to connect various sentences together, rather than just predict the next work in a sentence," he added.
He added that this new approach could provide AI with an innate sense of reasoning, allowing the new systems to search through possibilities and find much more accurate or successful solutions to problems.
OpenAI's new development
OpenAI hasn’t published details on its supposed Q* breakthrough, but it has published two papers about its efforts to solve grade-school math problems that utilise two distinct artificial intelligence models. This has led to speculation that the new Q* development is a combination of two AI models that are already in existence – Q-learning and the A* search algorithm.
Q-learning focuses on making better decisions, with the AI learning from its experiences, similar to how a person learns from playing a computer game. The more an AI using Q-learning plays, the better it becomes at figuring out how to succeed.
Q-learning, a form of reinforcement learning, resembles training a pet where good behaviour is rewarded, and mistakes are penalised. This approach reinforces the AI to not only value immediate rewards but also consider future rewards, contemplating the long-term consequences of its decisions.
Read more: Sovereign agents: Your own personal AI assistant? | The Crypto Mile
A* search algorithm is based on a classical computer science method that helps find the shortest path to navigate a maze. When A* search algorithm is combined with deep learning, a method for computers to learn and improve from experience, it results in a smarter system, yielding a more intelligent AI system.
While OpenAI's Q* breakthrough holds the potential to surpass human capabilities in specific activities, particularly in language construction with semantic continuity, it is essential to recognise that this achievement does not yet signify the attainment of artificial general intelligence.
Watch: Moody's building AI model for compliance checks | The Crypto Mile
Download the Yahoo Finance app, available for Apple and Android.
