I have AI psychosis :(

Quick Overview

The speaker confirms he has "AI psychosis" in a satirical sense, explaining his behavior as a deliberate, fluctuating persona he adopts to keep his content engaging and unpredictable, while simultaneously providing a serious breakdown of OpenAI's financial losses and the implications of AI development and safety.

Key Points: OpenAI lost $5.09 billion in 2024 and $38.5 billion in 2025, according to internal financial data cited in the video. The speaker argues that OpenAI's massive losses are driven by extensive spending on research, development, and sales and marketing. Anthropic has implemented a 30-day data retention policy for enterprise users for safety purposes, which the speaker interprets as a strategic move to train models and mitigate jailbreak risks. The speaker clarifies that his "AI psychosis" is a creative choice to keep his content dynamic and that his underlying stance on AI safety and development remains consistent. The video highlights that many AI companies are currently under pressure to balance rapid innovation with safety and financial sustainability.

Context: The video features a content creator who discusses the current landscape of AI development, specifically focusing on the financial and safety challenges faced by leading companies like OpenAI and Anthropic. He contrasts his own satirical online persona with the serious, highly technical, and often opaque nature of the AI industry.

Detailed Analysis

The creator addresses the audience's curiosity about his "AI psychosis," framing it as a conscious, satirical approach to content creation that keeps his channel engaging in a rapidly evolving field. He pivots to a detailed examination of recently leaked internal financial data for OpenAI, revealing substantial losses in 2024 and 2025, which he attributes to heavy investment in R&D and market expansion. The analysis then shifts to Anthropic, discussing their shift toward a 30-day data retention policy for enterprise users, which he posits is a safety-driven measure intended to harden models against jailbreaking while simultaneously improving them through further training. The discussion concludes by reflecting on the broader, more serious tone currently permeating the AI industry, as companies move past the initial hype phase and face the realities of building and scaling complex, high-stakes technology.