The Memo - 30/Jul/2025
OpenAI/Google IMO gold, HLE 70% ceiling, Neuralink patients 8 + 9, and much more!
To: US Govt, major govts, Microsoft, Apple, NVIDIA, Alphabet, Amazon, Meta, Tesla, Citi, Tencent, IBM, & 10,000+ more recipients…
From: Dr Alan D. Thompson <LifeArchitect.ai>
Sent: 30/Jul/2025
Subject: The Memo - AI that matters, as it happens, in plain English
AGI: 94%
ASI: 0/50 (no expected movement until post-AGI)OpenAI CEO to the US Federal Reserve (22/Jul/2025):
What if AI gets so smart that the President of the United States can’t do better than following ChatGPT-7’s recommendation, but can’t really understand it either? What if I can’t make any better decision about how to run OpenAI, and I just say, ‘You know what? I fully hand it over. ChatGPT-7, you’re in charge. Good luck.’ That might be the right decision in any individual case. But it means that society has collectively transitioned a significant part of decision-making to this very powerful system that’s learning from us, improving with us, evolving with us, but in ways we don’t totally understand.
There were 19 big models released in July 2025, mostly from China, see them all on the Models table. I’ve also added a new GPT-7 analysis placeholder page at LifeArchitect.ai/gpt-7. Some of my quantitative research including datasets, compute, models (and the ‘ALScore’!) was recently featured in large works by Cornell, and USTC with UT Austin and Georgia Tech.
Here’s a comment by an audience member during the live Q&A portion of one of my keynotes in Australia yesterday:
Contents
The BIG Stuff (OpenAI/Google IMO gold, HLE 70% ceiling, Neuralink patients 8 & 9…)
The Interesting Stuff (RobotEra L7, Meta Superintelligence, Anthropic psychiatry…)
Policy (AI.gov, Chinese smuggling chips, UK and OpenAI partnership…)
Toys to Play With (Assistant via ElevenLabs, Netflix El Eternauta…)
Things I’ve Been Thinking About (Work and jobs…)
Next (SingularityU invite, roundtable…)
The BIG Stuff
OpenAI and Google LLMs get IMO gold medal (15/Jul/2025)
In a remarkable display of AI capabilities, internal large language models from OpenAI and Google have achieved gold medal-level performance in the International Math Olympiad (IMO), adhering to the same stringent conditions as human participants (two 4.5-hour exam sessions and no external tools or internet access). This achievement challenged even the expectations of experts like Prof Terry Tao, who just a few weeks before the win said: ‘It won’t happen this [upcoming 2025] IMO. The [frontier model] performance is not good enough in the time period.‘ (15/Jun/2025)
OpenAI announce, OpenAI LLM proofs,
Google announce, Google LLM proofs (PDF)
See it on my AGI countdown: https://lifearchitect.ai/agi/
About 30% of Humanity’s Last Exam chemistry/biology answers are likely wrong (23/Jul/2025)
‘Humanity’s Last Exam’ (HLE) is a challenging AI benchmark comprised of 2,500 questions from various scientific disciplines, designed to test the limits of frontier language models. Recent findings by research team FutureHouse reveal that more than 29% of the chemistry and biology questions have answers that contradict evidence found in peer-reviewed literature.
…when extrapolating to the full dataset, we expect 29.3% of the questions to be directly conflicted by research, [only] 51.3% to be supported by research, and 19.3% to be nuanced, depending on assumptions or opinions.
Read more via FutureHouse.
I have updated my Mapping analysis to include this new ceiling: https://lifearchitect.ai/mapping/
Note that my ASI threshold is unchanged (and indeed corroborated) by this finding. It reads: ‘It is likely that any model with a primary score at >50% on HLE, and a secondary score at >90% on GPQA is an ASI system.‘
Read more: https://lifearchitect.ai/asi/
OpenAI prepares to launch GPT-5 in August (24/Jul/2025)
OpenAI is set to launch its highly anticipated GPT-5 model in early August, showcasing its new capabilities including integrated reasoning. This model is a significant step toward achieving artificial general intelligence (AGI), streamlining the use of large language models by unifying various series into a more cohesive system. OpenAI's CEO has expressed excitement, noting the model's impressive ability to answer complex questions instantly, highlighting a future where AI plays an even more integral role in problem-solving and innovation.
Read more via The Verge.
The early release model (13/Jul/2025) was made available to testers this month, and is called gpt-5-reasoning-alpha-2025-07-13.
My GPT-5 page: https://lifearchitect.ai/gpt-5/
OpenAI agreed to pay Oracle $30B a year for 4.5GW Stargate data center services (22/Jul/2025)
Note the security cameras and noise warning on the Stargate racks above. In The Memo edition 16/Oct/2024 I estimated that each rack is about as loud as a helicopter (101dB).
OpenAI has entered into a groundbreaking agreement with Oracle, committing to pay US$30B annually for data center services. This deal is part of a larger initiative, dubbed Stargate, which aims to build a massive data center infrastructure capable of supporting OpenAI's expanding AI capabilities. The contract, which involves 4.5 gigawatts of capacity, reflects the growing demand for computational power and positions both OpenAI and Oracle as significant players in the AI industry. This partnership underscores the accelerating scale of AI operations and the pivotal role of cloud providers in supporting such growth.
Together with our Stargate I site in Abilene, Texas, this additional partnership with Oracle will bring us to over 5 gigawatts of Stargate AI data center capacity under development, which will run over 2 million chips... Oracle began delivering the first Nvidia GB200 racks last month [Jun/2025] and we recently began running early training and inference workloads, using this capacity to push the limits of OpenAI’s next-generation frontier research. (OpenAI, 22/Jul/2025)
Read more via TechCrunch, OpenAI, Axios, and an unusual perspective by WIRED (‘…Struggles to Get Off Ground‘).
The Memo features in recent AI papers by Microsoft and Apple, has been discussed on Joe Rogan’s podcast, and a trusted source says it is used by top brass at the White House. Across over 100 editions, The Memo continues to be the #1 AI advisory, informing 10,000+ full subscribers including RAND, Google, and Meta AI. Full subscribers have complete access to all 30+ AI analysis items in this edition!
The Interesting Stuff
Reflections on OpenAI (15/Jul/2025)
Nearly everything is a rounding error compared to GPU cost.