In November 2022, OpenAI surprised the world with the launch of ChatGPT, its renowned chatbot powered by Artificial intelligence. More than two years after its iconic development, the company presented O3, a new family of reasoning models that promise further revolutionize the field of technology.
In detail, the company led by Sam Altman presented two versions of his recent algorithm: O3 standard – designed to process large amounts of data and perform general tasks – and O3 mini operawith a significantly smaller number of parameters and ideal for carrying out more specific tasks. Thus, the new OpenAI takes a step further and tries to get closer to the consecration of Artificial General Intelligence (AGI).
O3: the new OpenAI development that promises to get closer to Artificial General Intelligence
Released on December 20, 2024, the new OpenAI development made notable progress in: abstract reasoning, advanced mathematics and programmingamong others. O3 belongs to a new family of products based on ChatGPT that, unlike the renowned chatbot and its predecessor O1, invests more time – and money – to evaluate alternatives before generating a response for the user.
The evaluation ARC-AGIdesigned to measure progress of the different tools on their way to becoming an AGI, achieved a score of 87.5%, performance that is comparable to what humans can achieve. This type of evaluation includes, for example, the resolution of visual logic puzzlessimple for humans but difficult for computers.
artificial intelligence
The goal of AGI is to imitate the functioning of the human brain.
Depositphotos
In detail, O3 also obtained outstanding results in software engineering and in one Codeforce programming competitionmanaging to be ranked 175th worldwide. But the advances do not stop there: the new OpenAI family surprised with very high results in a complexity math assessment
Despite expectations, some detractors of the Artificial Intelligence industry call the announcement exaggerated. Such is the case of the same creator ARC-AGIwho pointed out that currently there are much cheaper tools that can achieve a 80% in said evaluation. The main obstacle that O3 encounters is manage to generalize their knowledge to other tasks.
For the moment, technology fans will have to wait to access the new models of OpenAI flagship reasoning. The company announced that security researchers can sign up for a waiting list to test the model and the North American firm hopes to launch its new advance to the public later, although it has not yet been confirmed whether it will be done in a paid or free format.
What is Artificial General Intelligence?
Broadly speaking, Artificial General Intelligence is the next step for AI developers. These types of automatic systems are, in theory, capable of successfully perform any intellectual task that humans perform.
That is, the AGI They will possess the ability to understand or learn any intellectual task that a human being can perform. Simply put, it is a type of tool whose objective is imite the cognitive abilities of the human brain.
In addition to the main characteristics mentioned above, AGI systems also possess certain other traits that differentiate them from the AI that we currently know. Among them are:
- Generalization ability: AGI can transfer your learnings and skills from one field of knowledge to another. This allows you to adapt effectively to new and unseen situations.
- Common sense knowledge: AGIs will also have extensive knowledge about the world, which may include different areas such as; facts; relationships and social norms. This generates deeper reasoning upon which decisions are made.
AGI research and development requires interdisciplinary collaboration between fields such as computer science, neuroscience and cognitive psychology.
The difference between AGI and “narrow” AI:
Compared to the current tools available to users, AGI presents a higher level of autonomy and generality of the functions they can perform. While the first is dedicated exclusively to solving tasks for which it was specifically programmed, AGI would allow a machine to operate in a wide range of activities thanks to its ability to generalize and transfer knowledge from one area to another.
Currently, the field of AGI is in an exploratory phase. However, the launch of O3 and the fierce competition between OpenAI, Google and X, among others, envision a not-too-distant future where these types of tools are part of people’s daily lives.
Source: Ambito