Title: Sam Altman Missteps: "We Can Build AGI" - A Misleading Claim
Title: Sam Altman Missteps: "We Can Build AGI" - A Misleading Claim
OpenAI has continually been in the limelight for its disruptive announcements and bold claims. From the controversial release of GPT-2 due to safety concerns to the "12 Days of Christmas" showcase of new products, OpenAI remains a trailblazer in the AI industry.
Recently, OpenAI's co-chair, Sam Altman, shared his thoughts on the past year: "We now know how to build AGI as it’s usually understood."
Artificial General Intelligence (AGI) refers to developing an AI system that matches human intelligence in breadth and depth. Unlike narrow AI, designed for specific tasks like language translation, chess, or facial recognition, AGI can take on any intellectual task and adapt across various contexts.
Is AGI Imminent? Not Likely, at Least Not the AGI We're Expecting
I am convinced that AGI isn't close yet. Today's AI, such as ChatGPT, relies on pattern recognition and probability calculations rather than understanding. Applying the phrase "Life is like a box of..." with "chocolates" may seem intelligent, but it's based on probabilities, not reasoning. Even when interpreting TikTok algorithms recognizing someone's sexual orientation before they did, it's seen as just "Machine Learning" and "Pattern Recognition."
Many experts share this skepticism. Demis Hassabis, who I worked with at Google, predicts AGI will arrive by 2035. Ray Kurzweil, meanwhile, anticipates 2032, while Jürgen Schmidhuber, director of IDSIA, suggests it won't happen until 2050. This uncertainty about the timeline persists with a multitude of perspectives.
The AI Effect: AGI's Ever-Shifting Definition
Sam Altman has softened the "G" in AGI, asserting that "we will hit AGI sooner than most people think, and it will matter less." On Sunday, he described AGI as "a very sloppy term," warning that the expectations for AGI may change as AI systems develop capabilities considered general intelligence.
Sam Altman and Reasoning
When Sam Altman states, "We now know how to build AGI," he likely refers to OpenAI's o1 – an AI model designed for iterative, self-calling reasoning. The AI consists of two critical steps:
- Iterative and Reflective Process: The model generates an output, evaluates or critiques it, and refines it through self-reflection in a new round of reasoning.
- Feedback Loop: This iterative process results in a continuous feedback loop, allowing the model to revisit its outputs, criticize them, and improve them further.
OpenAI's o1 doesn't just respond with answers but plans, critiques, and continuously refines its plan – a significant departure from its previous focus on bigger models.
Sam Altman vs. Salesforce, Microsoft, Google, and Amazon
OpenAI faces fierce competition in 2025 from established tech giants like Salesforce, Microsoft, Google, and Amazon. These companies, with their exclusive datasets and customer bases, pose a threat to OpenAI's supremacy in the AI agent market. However, OpenAI remains optimistic, relying on its technological edge to outperform competitors.
[1] "Why Is Google's AI Chief Predicting The Emergence Of Artificial General Intelligence (AGI) In 2025?" Forbes. (2023, June 1).[4] "Assessing Progress Towards AGI." LessWrong. (2023, June 1).
The idea that AGI could be achieved within the next decade is hotly contested. While some experts like Demis Hassabis, Ray Kurzweil, and Jürgen Schmidhuber offer optimistic timelines, other industry figures are more cautious, suggesting that AGI remains a long-term goal.
Sam Altman's assertion that "we now know how to build AGI" may be a reference to OpenAI's work on o1, a model designed for iterative, self-calling reasoning. The AI comprises two key steps:
- "Iteration and Reflection": The model generates an output, critiques it, and revises it in a fresh round of reasoning.
- "Feedback Loop": This generates a continuous loop, enabling the model to review its outputs, critique them, and further improve them.
In essence, GPT-like AI evolves beyond mere response generation – it plans, evaluates its plan, and refines it in a continuous process.
Despite this advancement, human-crafted prompts can outperform OpenAI's o1 at the moment. This indicates that open-source AI is yet to fully emulate human-like reasoning. Yet, as technology advances, open-source AI can continue to develop and improve, potentially surpassing current capabilities.
OpenAI's main competitors in the AI agent market in 2025 are Salesforce, Microsoft, Google, and Amazon. These major corporations possess substantial resources, closed datasets, and customer bases – putting OpenAI under considerable pressure to maintain its position as the market leader.
OpenAI, however, remains confident in its technical prowess, and Altman proudly claims that AGI will soon become a reality. To achieve this, OpenAI's model, o1, focuses on "thinking longer," or engaging in long-term inference loops, rather than simply building bigger models, which was once a primary concern.
These predictions about AGI's arrival vary widely among experts. Some, like Demis Hassabis and Ray Kurzweil, foresee AI hitting the mainstream either in 2035 or 2032, respectively. Jürgen Schmidhuber, in contrast, envisions AGI becoming a reality only around 2050.
Skepticism remains rife in the field; the timeline for AGI's arrival remains uncertain. Eventually, AGI may reach a level where AI systems will handle complex intellectual tasks as efficiently as humans, or perhaps even surpass human intelligence.
The evolving definition of AGI has resulted in changing expectations over time. When TikTok's algorithms successfully identified someone's sexual orientation before they did, the general consensus was to label this as "Machine Learning" and "Pattern Recognition."
Sam Altman recently criticized the "G" in AGI, asserting that AGI could become a reality sooner than we think, and its impact may be less significant than initially thought. Altman's statements may seem like moving the goalpost, as one could interpret this as loosening the requirements for AGI to be considered achieved.
If we relax the strict definition of AGI, as some experts propose, like making it possible to handle a wide range of tasks, looks like it could happen sooner than anticipated. In that perspective, AGI could become a reality really soon.
Some industry insiders jest that AGI is "what we haven't yet built" because as AI systems evolve to perform tasks once thought to require general intelligence, the definition of AGI frequently shifts.
When Sam Altman says that "We now know how to build AGI," he likely refers to OpenAI's o1 Model. This model uses an iterative, self-calling process that involves the AI:
- Generating an output
- Evaluating or critiquing that output
- Refining the output in a new round of reasoning
- Creating a feedback loop that allows the AI to repeatedly revise its outputs, critiques, and improvements
This approach enables GPT-like AI to extend beyond straightforward responses. Instead, it plans, critiques its plan, and continually refines its approach – a significant departure from previous focus on larger models.
1.Microsoft and Salesforce, two tech giants in the AI industry, are expected to pose significant competition to OpenAI in 2025, challenging its dominance in the AI agent market.2.Sam Altman, co-chair of OpenAI, has expressed optimism about the future of AGI, asserting that OpenAI has the technological edge to outperform competitors and that the development of AGI is closer than many people think.3.Ecornell, a renowned online learning platform, offers courses on AI and data science, providing individuals with the opportunity to expand their knowledge and skills in these fields.4.Data accessibility and reasoning capabilities are critical components in the development of AGI, as they enable AI systems to learn from a wide range of sources, analyze information, and make informed decisions.5.Companies like Salesforce and Microsoft, with their extensive databases and resources, can leverage data to improve their AI models and gain a competitive edge in the AI industry, making data access an essential factor in determining the success of AI systems.