The piece focuses on the belief in Artificial General Intelligence (AGI) among some figures at OpenAI and how this belief impacts engineering practices in AI. Elon Musk's founding of OpenAI stemmed from his view of Demis Hassabis as a potential supervillain capable of creating AGI, which Musk believed could either save or doom humanity. OpenAI’s co-founder, Ilya Sutskever, has made striking statements about the promise and peril of AGI, even conducting a dramatic ritual involving a burning effigy symbolizing a deceitful AGI. This kind of fervent belief has shifted from science fiction to a mainstream mindset within Silicon Valley. Karen Hao's insights reveal that OpenAI's development of GPT-2 was based on the "pure language" hypothesis, which assumed that AGI could emerge solely from language training. This belief led to an aggressive scaling of models, necessitating vast resources, including data centers that consume enormous amounts of water and power, often relying on polluting energy sources. The article criticizes the environmental toll and ethical issues stemming from the data exploitation necessary to train these models, especially given that LLMs (large language models) are pushed forward under the guise of potentially delivering massive value despite the uncertain probabilities and fictitious values assigned to AGI's success. The author argues that this fixation on AGI is a barrier to effective engineering. LLMs and similar technologies fail to address problems efficiently and ethically. Instead of pursuing AGI as a universal solution, the piece advocates for a more pragmatic approach. It suggests focusing on smaller, purpose-built models for specific tasks, conducting thorough cost-benefit analyses, and making clear trade-offs. This shift away from the AGI fantasy would allow for genuine engineering improvements without the associated harms to people and the environment.