Sam Altman has been writing on his blog for a longer time than he has been part of the AI sector. Currently the CEO of OpenAI, Altman initiated his blog, named “Sam Altman,” in 2013 while employed at YCombinator, where he would shortly ascend to the role of president. The early entries do not reference AI, instead concentrating on B2B startup resources, topics for dinner gatherings, and skepticism towards UFOs.

Altman experienced an epiphany: “The most accomplished founders do not aim to establish companies,” he stated. “They embark on a quest to forge something resembling a faith.” Fast-forward to his most recent 2025 blog entry, entitled “The Gentle Singularity,” and it appears that he has achieved his objective.

“We have crossed the event horizon; the ascent has commenced,” Altman starts, adopting a prophetic tone. “Humanity is nearing the creation of digital superintelligence.” Nevertheless, scientific evidence does not substantiate the idea that humanity is on the verge of achieving Artificial General Intelligence. Present models are effective for handling vast data sets but can be extremely erroneous in different contexts. AI hallucinations are becoming more prevalent with chatbots, leading to tangible complications.

There are no progressions in reasoning, as underlined in a recent paper indicating that AI models sometimes fail to identify answers even when presented. However, Altman remains hopeful about the future, forecasting that by 2026, systems will produce original insights, and by 2027, robots will undertake real-world activities. By 2035, he anticipates advancements from high-energy physics to the colonization of space.

Altman’s assertions are speculative, utilizing terms like “likely,” “may,” and “maybe” to carry significant weight. Despite his almost religious faith in AGI, Altman cannot entirely disregard reason. His assertions, like “ChatGPT is already more formidable than any human who has ever existed,” lack substantiation.

Altman’s recent blog entry also addresses ChatGPT’s consumption of energy and water, representing his initial commentary on the matter. Since the emergence of ChatGPT in 2023, OpenAI and competitors such as Google and Microsoft have been reticent about their data center consumption. Sasha Luccioni, climate lead at HuggingFace, remarked on the absence of transparency.

Altman eventually disclosed some figures, stating that an average ChatGPT request consumes approximately 0.34 watt-hours and 0.000085 gallons of water. However, he did not furnish any additional particulars or specifications regarding the model, and OpenAI has not addressed subsequent inquiries. Altman appears to minimize the energy and water necessary for model training, which raises concerns among researchers.

Now that Altman has responded to the “often curious,” he has diminished justification to retain information. Making all data public would enable others to validate his figures, similar to established scientific methodology. In the meantime, discussions about energy and water consumption in data centers are escalating in the US. Luccioni has created an AI Energy Leaderboard that illustrates the variability among open-source AI models.

This matter is crucial as companies seek to reduce energy expenses, and there is active industry involvement. Meta, Microsoft, and Google are already represented on the board. Can OpenAI afford not to join? Ultimately, the response hinges on whether Altman is constructing a business or a belief system.