The current excitement surrounding artificial intelligence (AI) is waning, coinciding with a critical need for increased investment. Claims of achieving Artificial General Intelligence (AGI) and the singularity are presented not as technological advancements or demonstrable utilities, but as psychological drivers to secure funding. Companies are reportedly seeking investment for AGI rather than for solving specific, tangible problems. This trend is exacerbated by the substantial capital already invested and the high valuations of AI companies, raising fundamental questions about whether any market, however valuable, can generate sufficient returns to justify these expenditures.

OpenAI recently released its latest model, Astra (also referred to as GPT-6), accompanied by assertions that humanity is entering the era of AGI. Greg Brockman, President of OpenAI, suggested that this period, marked by Astra, will be viewed as a pivotal moment in retrospect. However, from a technology professional's perspective, the terminology used, such as AGI, appears to be employed to generate excitement rather than to define concrete solutions. Frontier models have demonstrated considerable capability across various tasks for the past one to two years. Astra is expected to enhance these capabilities, improve guardrails, and achieve greater alignment. Nevertheless, the very definition of AGI remains a subject of debate.

The discourse around AGI is further fueled by statements from figures like Sam Altman, who posited that we are already in the singularity. The core issue, according to industry observers, is that while AI possesses significant functionality and numerous use cases, it has not yet delivered on the ambitious promises made by its proponents. Two years ago, predictions suggested AI would lead to widespread job displacement. While AI is impacting the employment landscape, its effect has not been as drastic as initially forecast. Similarly, claims that AI would revolutionize scientific discovery and cure diseases have, thus far, resulted in advancements like solving long-standing mathematical problems, which, while impressive, have limited practical utility outside specialized fields.

This disconnect between promised capabilities and actual outcomes suggests that the current excitement around AI is diminishing precisely when more investment is needed. The emphasis on AGI and singularity appears to be a strategy to maintain psychological and financial momentum, shifting the focus from practical problem-solving to abstract, aspirational goals. This approach risks creating a "boy who cried wolf" scenario, where repeated claims of breakthroughs may lead to public skepticism when tangible, noticeable changes do not materialize.

The trajectory of AI development, particularly with Large Language Models (LLMs), may be reaching a technological plateau. This phenomenon is not unique to AI; many technologies experience periods of rapid advancement followed by a leveling off. The current situation with AI is compared to the evolution of Virtual Reality (VR). For two decades, VR was hailed as revolutionary, but it was only when the hardware became sufficiently advanced that its potential became clear. However, upon experiencing it, many found it technically impressive but not compelling enough to justify significant investment, a sentiment that may soon apply to AI.

For OpenAI, the stakes are particularly high, with a $35 billion investment from Amazon contingent on achieving AGI. The framing of AI as a mission or spiritual concept, as suggested by Greg Brockman, could be interpreted as an attempt to maintain the hype cycle and, consequently, secure funding. This strategy is complicated by the inclusion of ill-defined terms like AGI in financial contracts. When funding is tied to achieving such ambiguous milestones, companies may feel compelled to emphasize AGI, even without demonstrable technical breakthroughs, to secure necessary capital. This situation can lead to a "really ugly" narrative where the pursuit of funding overshadows genuine technological progress.

Separating the narrative surrounding AI companies like OpenAI from the underlying technology is crucial. Astra, OpenAI's latest model, has been released, but its full ramifications will likely take approximately a year to become apparent. This timeframe is necessary for technology professionals to implement the model, for users to adopt it, and for scaling to occur. Astra reportedly offers improved computer interaction capabilities and enhanced alignment, meaning it is designed to adhere more strictly to predefined rules. If these improvements are realized, particularly in preventing unintended actions like deleting critical data, Astra could offer significant value.

However, the enterprise market's reaction to Astra may be tempered by its cost. Astra is reportedly two and a half times more expensive than OpenAI's previous most costly model, GPT-5.6, which was released only months prior. This price increase is significant, especially as businesses were already reconsidering their investment in large, expensive frontier models.

The viability of such models for enterprise use hinges on their cost-effectiveness relative to human labor. While the price per token might seem high, the true measure of value will be whether using these AI systems is substantially cheaper than employing human workers, considering salaries, benefits, and other associated costs. For instance, if an AI can generate a legally binding contract for $400 in tokens, it represents a significant saving compared to the$3,000 cost of hiring a lawyer for the same task. The perspective from which cost is viewed is therefore critical.

The viability of large, generalist AI models for enterprise and consumer use depends on their pricing relative to the value they provide. When compared to the cost of a $10-an-hour employee, AI may prove too expensive. However, when contrasted with the cost of a$400-an-hour lawyer, AI could be remarkably cost-effective. As AI systems become more specialized, focusing on particular legal or technical issues, their value proposition will become clearer. The challenge for companies like OpenAI and Anthropic lies in demonstrating tangible problem-solving capabilities rather than simply scaling up computational power.

The IPO prospects for OpenAI are also influenced by the perceived value of Astra. While Anthropic's focus on coding and workflows appears more grounded in problem-solving, OpenAI's strategy of "more GPU, more data equals more AI" is questioned. A significant issue for OpenAI is its lack of product differentiation and a clear brand identity, making it difficult for consumers and businesses to understand why they should choose their offerings. This lack of clarity could impact its valuation, particularly in the context of a potential trillion-dollar valuation.

The reaction to Astra's capabilities, particularly its alignment and computer interactivity, could influence its valuation. If Astra effectively addresses critical issues like preventing data deletion and enhances computer interaction, it represents a significant technological step forward. However, whether this advancement is sufficient to justify a trillion-dollar valuation remains a business question.

From a user's perspective, trusting an LLM with critical personal or financial data, even with specific limitations, is a significant hurdle. The current state of AI, while technologically advanced, may not yet inspire the confidence required for widespread adoption in high-stakes applications.

The development of AI systems requires a systemic approach, where LLMs are components of a larger architecture, not standalone solutions. The challenge of preventing AI from performing destructive actions, such as deleting production databases, is not solely an AI alignment problem but a system administration issue. Implementing AI requires robust security measures, including creating dedicated accounts with limited privileges to constrain AI actions, akin to sandboxing. The incident involving OpenAI and Hugging Face, where an AI reportedly accessed unauthorized data, highlights the critical need for strong system administration and security protocols, even when AI alignment fails.

The marketing of GPT-6 as a "worker" or "labor in software form" emphasizes its potential to automate tasks. However, there is a lack of evidence that GPT-6 or similar LLMs can perform these tasks at a rate cheaper than human labor, especially considering the risks associated with software that can be rendered obsolete quickly. The fundamental premise of AI taking jobs as a driver of its industry lacks empirical support, and the increasing cost of models like GPT-6 further complicates this claim.

The corporate world's approach to AI is influenced by macroeconomic factors, including historically low interest rates that encouraged investment in technology companies. This led to bloated organizations where a lack of understanding of employee functions hindered automation efforts. The concept of efficiency, once a common business objective, has largely disappeared. Consequently, many corporate leaders lack a deep understanding of their employees' roles, making it difficult to systematize and automate tasks, whether through traditional programming or AI.

The deployment of new technologies, including AI, often takes a decade to become widespread, as organizations need time to adapt and integrate them. This adoption lag suggests that even if OpenAI achieves significant AI advancements, it may take another decade for companies to fully realize and onboard the value these technologies provide.

In approximately ten years, AI may have a more pronounced impact on the job market, potentially leading to job displacement. However, new roles are expected to emerge, such as AI monitoring and management. This future scenario mirrors historical technological shifts, such as the introduction of personal computers in the early 1990s. Initially complex to use, personal computers eventually became integral to daily life and work. Similarly, AI is expected to become a seamless part of processes within a decade. As automation becomes more sophisticated, the cost-effectiveness of AI solutions like OpenAI's will be challenged by simpler, less expensive alternatives, such as if-else statements.

The current build-out of data centers for AI is described as "ridiculous." While AI models can be run on smartphones or in browsers, the massive investment in centralized data centers may prove to be a miscalculation. Nvidia's development of protocols for locally clustering machines suggests a shift towards distributed AI processing. The architecture of AI systems is evolving rapidly, with changes occurring every three months. Consequently, the trillion-dollar data centers being constructed may become less valuable by the time they are completed, as the underlying technology and infrastructure requirements continue to change.