Leopold Stays in the Game, Big Tech Earnings, OpenAI Cuts GPT-5.6 Prices

The episode opens with Leopold Aschenbrenner’s letter to Situational Awareness LPs after a brutal July. The fund suffered a sharp drawdown, but the letter says it is not shutting down, liquidating or becoming a private-only vehicle. The key move was selling part of the public book to Citadel to remove leverage, eliminate margin-liquidity risk and preserve private positions.

What the episode highlights

• Situational Awareness faced extreme moves in AI-linked positions and adverse trading in names publicly associated with the fund.
• The letter accepts responsibility, frames the month as an expensive lesson and promises changes in portfolio management, risk staffing and resilience.
• TBPN draws a line between a private block transaction and a bailout: this was not a government rescue, but a sale to a willing counterparty.
• The AI thesis is not treated as broken; the practical lesson is about leverage, concentration, liquidity and portfolio construction.

Big Tech and AI capex

The earnings discussion focuses on Microsoft, Apple, Amazon, Meta and Alphabet. Top-line and bottom-line beats matter less than the market’s interpretation of AI spending, cloud growth and the credibility of future returns on infrastructure investment.

Models, cost per task and benchmarks

The OpenAI segment shifts attention from cost per token to cost per task. A cheaper token price does not guarantee a cheaper outcome if the model needs many more tokens. ARC-AGI V3 also illustrates how evaluation harnesses, memory and API settings can materially change benchmark scores and output-token usage.

Takeaway

The common thread is execution under volatility: running an AI-focused fund without fragile leverage, making hyperscaler capex credible, and integrating frontier models in ways that actually improve task economics.

Source

• Chaîne: TBPN
• Vidéo source: https://www.youtube.com/watch?v=xa5MncRRgoU