AI Advances Beyond AGI with GPT-5.6 and New Models, GMI Issues Rare Buy Signal

Key Takeaways

1. GPT-5.6 achieves a breakthrough in solving the Benjamini-Hochberg conjecture, while multiple open-source models (GLM 5.2, Kimi, Grok 4.5) are released, signaling accelerated AI progress beyond AGI.
2. GMI Compounding Machine triggers a rare buy signal on RealVision, with a track record of 12x returns in weak markets versus 7x for buy-and-hold.
3. S&P 500 data shows extreme timing sensitivity: a $1 investment from 1990-2024 grows to $40, but missing the best 25 days yields only $8, while avoiding the worst 25 days yields $236.
4. Port Alpha announces a $3 billion next-generation shipyard in Texas, integrating autonomous systems and software-defined manufacturing, expected to create 10,000 jobs.
5. Economic data shows housing starts rose 19% beat expectations, building permits fell 3%, consumer confidence jumped to 54.4, and industrial output grew 0.1% in June.

1. AI Models and Infrastructure

1. Marc Andreessen reported that GPT-5.6 helped solve a long-standing conjecture regarding the Benjamini-Hochberg procedure's ability to control false discovery rates under correlated data, demonstrating significant capability improvements over GPT-5.5. The result is primarily conceptual but marks a milestone in AI-driven mathematical research. — via 1
2. @jason noted that the pace of AI innovation is accelerating, with multiple major model releases in the past two weeks including GLM 5.2, thinking machines, Kimi, and Grok 4.5. He argues that we are moving beyond AGI and that a moderate cooldown of AI hype could be beneficial. — via 1 2
3. @jason stated that AGI has already been achieved, open-source models are accelerating, token prices are dropping 90% annually, and autonomous robots are being weaponized, with China training robotic soldiers for war, urging US robotics companies to immediately engage in war training. — via 1

2. Markets and Investment

1. Raoul Pal highlighted that the GMI Compounding Machine on RealVision has issued a rare buy signal. The strategy, which compounds investments in weak periods, has achieved a 12x return, significantly outperforming a simple buy-and-hold's 7x return. — via 1
2. Ben Carlson provided a powerful illustration of market timing: $1 invested in the S&P 500 from 1990 to 2024 grew to $40 when held continuously; missing the best 25 days reduced it to $8, while avoiding the worst 25 days resulted in $236. Simultaneously missing best and worst yielded $47. This underscores the difficulty of market timing. — via 1
3. Ben Carlson also observed that despite some tech stocks declining sharply, the S&P 500 is only 1% off its all-time high, satirizing the notion of "buying when there's blood in the streets." — via 1
4. Liz Ann Sonders shared key economic data: June housing starts surged 19% above expectations, but building permits dropped 3%, missing estimates; June industrial output rose 0.1% in line with forecasts, maintaining an upward trend; July Michigan consumer sentiment rose to 54.4 from 49.5 previously. — via 1 2 3
5. Lyn Alden flagged ongoing U.S. government deficit concerns, arguing that neither party can curb spending, and the deficit remains severe. — via 1 2

3. Frontier Technology and Applications

1. Marc Andreessen announced the Port Alpha project: a $3 billion investment to build a next-generation shipyard in Brownsville, Texas, combining software-defined manufacturing with autonomous systems. The project is expected to create 10,000 jobs and have a significant economic impact. — via 1
2. Marc Andreessen reported that the Vesuvius Challenge has achieved a major breakthrough: a complete scroll has been read using new unfolding algorithms, with a $1 million grand prize and several $50,000 prizes now offered. — via 1
3. Marc Andreessen predicts that "Vibe research" will become the biggest trend of 2026, as AI is already good enough for people to conduct informal research using it. — via 1
4. @jason shared a concrete example: an employee built an internal application using Lovable in four hours, costing less than $2,000 per year, compared to $500,000 for a similar project two years ago. — via 1