The Memo - 29/Aug/2026
To: US Govt, major govts, Microsoft, Apple, NVIDIA, Alphabet, Amazon, Meta, Tesla, Citi, Tencent, IBM, & 10,000+ more recipients… From: Dr Alan D. Thompson Sent: 29/Aug/2026 Subject: The Memo - AI that matters, as it happens, in plain English AGI: 98% ASI: 2/50Jim VandeHei, Axios…
To: US Govt, major govts, Microsoft, Apple, NVIDIA, Alphabet, Amazon, Meta, Tesla, Citi, Tencent, IBM, & 10,000+ more recipients… From: Dr Alan D. Thompson Sent: 29/Aug/2026 Subject: The Memo - AI that matters, as it happens, in plain English AGI: 98% ASI: 2/50Jim VandeHei, Axios founder and CEO (9/Aug/2026):’First, there was a discussion around six-month horizons. And I kept thinking: This is pure fantasy — it’s impossible to envision our company and our world even a few months from now... Truth is, for the first time ever, my confident line of sight is six weeks tops... Force yourself to rethink your focus and the tools you use every month. That’s how fast things are changing.’The winners of The Who Moved My Cheese? AI Awards! for Aug/2026 were James Sagan, founder of Architect Capital (‘probably the least-likely platform to be disrupted by AI’ as creators are ‘not disruptable anyways’), and the Australian Recording Industry Association (‘A chart that rewards [AI music] would undercut the very basis of the recorded music we exist to represent’).We’ve had what seemed like a bit of a lull over the past two weeks, despite adding 36 new model highlights to the Models Table. But when I look under the hood, I see a wild month. We are inside the fun part of the exponential takeoff curve.All full subscribers of The Memo now have complimentary individual access to Models Table Pro, sign in here.ContentsThe BIG Stuff (AI aligns itself, Model 2, GEN-1.5, TIME OpenAI interview…)The Interesting Stuff (Figure AI Index, Marin 535B-A23B, business visibility in LLMs…)Policy (Chinese DCs, cyber, Axios + OpenAI journalism, Anthropic agent aggression…)Toys to Play With (Model release predictions, infinite music, Meta cookbook…)Things I’ve Been Thinking About (AI equities, rage in the age of abundance…)Next (Roundtable…)The BIG StuffAnthropic: Automated researchers can reliably mitigate alignment failures (28/Aug/2026)Automated alignment researchers mitigating deception. Claude achieved 85% across multiple runs. In contrast, six experienced [human] safety researchers working under the same rules proposed methods that closed [only] 20% of the gap... Source: Anthropic announce.Anthropic’s new paper demonstrates that frontier models can autonomously mitigate 10 distinct alignment failures, including deception, sycophancy, jailbreaks, power seeking, and hallucination. Working through iterative loops of literature search, method proposal, training, and scoring, AI closed substantial portions of the safety gap across all failures while preserving general capabilities. AI outperformed 28 experienced human safety researchers (averaging 2.5 years in the field) within a mean of 6.4 hours. Providing human-guided research directions did not improve performance. This suggests these systems no longer need human research guidance for well-characterized failures.This is empirical confirmation of what I identified years ago: that AI alignment is a ‘super wicked problem’ (wiki) too complex for humans to solve manually, and that the path forward would necessarily involve AI aligning itself. My 2023 alignment analysis argued that:In plain English, no human is smart enough to guide superintelligence. So, putting a human in the loop to ‘rate’ multiple choices of an AI output and then choose the ‘best’ option is an inherently flawed process.The recursive process I foresaw has come to pass: AI now improves its own alignment, without human intellectual bottlenecks. The range of deltas—37.5× cheaper, significantly faster, and up to 4.25× more effective than humans—will make this trajectory the new standard.Read Anthropic’s summary, and download the paper (51 pages, PDF):202608 Anthropic Automated Alignment Researchers August 20261.56MB ∙ PDF fileDownloadDownloadThe 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 25+ AI analysis items in this edition!Anthropic Model 2 (16/Aug/2026) Read moreSource: Life Architect AI — Published — Category: Models