Ai

On Thursday, October 22, Nvidia fell 4.4 percent. By Friday’s close of $210.96 the stock had posted its third consecutive losing week, sat 12 percent below the high it set on October 6, and was down 7.6 percent for the month, its worst month since last November (CNBC, October 24, 2026). Nothing in that week’s news explained the selling. At GTC Berlin the company announced a shipping container holding seven Vera Rubin racks, which cuts the time to bring a data center online from six months to fourteen days, and a supply-chain expansion that includes a $1.1 billion capacity advance to a Taiwanese testing house (CNBC, October 21, 2026; NVIDIA, October 21, 2026). Those are the announcements of a company that expects to sell an enormous amount of hardware. ...

On July 31 I argued that the danger in the artificial intelligence buildout lay less in whether the technology works than in how the capital behind it is arranged: hardware that depreciates faster than the debt secured against it, revenue that arrives slower than the projections promise, and a downside diffused onto private-credit funds, pension holders, and index investors who never chose the bet. On October 14, that argument acquired a price. ...

The United States and Iran described two different waterways on Wednesday. United States Central Command said in a post on X that traffic was crossing the Strait of Hormuz carrying commercial goods and energy supplies, “including 20 million barrels of crude oil,” and that “the United States and regional partners clearly control the strait.” Mohammad Reza Naghdi, a senior adviser to the Revolutionary Guards commander-in-chief, said the same day that the strait “is closed, and the armed forces have full control over it,” and that it would stay closed until Iran’s demands were satisfied. Both statements carry the date October 7, and they cannot both describe the same water. Crude was up on the day: the front month for West Texas Intermediate stood at $89.10, up 0.93 percent, as of 8:31 p.m. Eastern. ...

Yesterday, Apple released macOS 27. Golden Gate. It is the first version of macOS that will not run on a single Intel Mac, and it is the last with full Rosetta 2 support, the translation layer that let Apple silicon Macs run Intel software.1 Four machines that supported the previous release were dropped flat: the 16-inch MacBook Pro from 2019, the 13-inch MacBook Pro from 2020, the 2020 iMac, and the Intel Mac Pro from 2019.2 If you are reading this on one of them, yesterday was the day the current stopped. ...

On Tuesday, September 8, a 28-year-old researcher named Jacob Coxon resigned from Anthropic and said the people building the technology believe it could kill us all by the end of the decade. This week, the story did not fade, as stories usually do. It compounded. Evan Hubinger, Anthropic’s alignment lead, said he personally believed there was a greater than 10 percent chance AI could kill all humans within the next decade, and that the field did not yet have a plan to solve alignment.1 Paul Christiano, the former head of safety at the US Commerce Department’s Center for AI Standards and Innovation, said the recent pace of capability gains gave him “a meaningful risk that rapid acceleration in AI capabilities leads to catastrophic and irreversible loss of control in the very near term.”2 Two researchers named Wang, one at OpenAI and one at Anthropic, posted within hours of each other that recursive self-improvement was the danger “hard to overstate” and that “there is not yet a viable scientific plan” to make it safe.3 OpenAI’s chief scientist wrote that he expects the speed of progress to carry into systems that increasingly drive their own development, and called it “a time that calls for extreme caution,” adding, “I am concerned no one is prepared for the consequences of a continued rapid rise in machine intelligence.”4 ...

On Tuesday, September 8, 2026, a 28-year-old researcher named Jacob Coxon resigned from Anthropic and posted his reasons on X. He had previously worked at OpenAI. He wrote that neither company was acting responsibly, that they were “racing straight to self-improving superintelligence and gambling with our lives,” and that the people building the technology “earnestly believe that it could kill us all by the end of the decade.”1 Two of his colleagues, still employed, backed him up. Evan Hubinger, a lead in Anthropic’s alignment division, wrote that he personally believed there was a greater than 10 percent chance AI could kill all humans within the next decade, and that the industry did not yet have a plan to solve alignment.2 Samuel Marks, Anthropic’s scalable oversight lead, added the line that should have stopped the conversation: “In general, the more senior the employee, the more concerned they are.”3 ...

Three facts describe the current state of artificial intelligence infrastructure spending, and none of them are matters of opinion. The companies building that infrastructure have, over the past five years, repeatedly extended the accounting assumptions that determine how quickly their hardware loses value on paper: assumptions that flatter today’s earnings regardless of what eventually happens to the hardware itself. The revenue actually being generated by the AI industry falls far short, by any reasonable measure, of what would be required to justify the scale of capital currently being committed. And the financing structure built to bridge that gap has diffused the risk away from the technology companies themselves and onto pension funds, index-fund investors, and, in some cases, ordinary electricity ratepayers; parties who never chose to make a bet on artificial intelligence at all. ...

APA-Citable Sources for “The Optimal Use of Hermes” Below are formatted APA citations organized by which claim in the essay they support. Sources marked [CONFIRMED] were verified via direct API lookups. Sources marked [TRAINING] are drawn from the agent’s pretraining knowledge of well-known, highly-cited academic papers. Claim 1: Hermes Agent: an open-source agent framework from Nous Research (memory, skills, delegation, cron) Nous Research. (2026). Hermes Agent documentation. Retrieved May 21, 2026, from https://hermes-agent.nousresearch.com/docs/ ...

Correction notice — May 26, 2026. This essay was factually reviewed on 2026-05-26 and republished with specific corrections applied. Among them: removed fabricated citations, including an invented robot (“Shumana”), invented Stanford Cart developer names, and a fabricated RAND study author; corrected dates, attributions, and roles for figures including Minsky, McCarthy, Geoffrey Hinton, and the BDI framework’s origins; corrected specific historical details (SNARC’s 1951 build, the Strategic Computing Initiative’s 1983 launch, OpenAI Five’s release year, the DQN Atari count, Brynjolfsson et al.’s citation, and several others); and removed fabricated co-authors from the Silver et al. and Mnih et al. reference entries. A reader who compared the original May 21, 2026 version to the current text would see that the historical claims, attributions, and reference list have been tightened throughout. The work should be read as corrected, not as it first appeared. If you spot a remaining error, please email phil@huffmanwrites.org. ...

Update (June 2026): Hermes is no longer part of my workflow. It has been replaced by Claude, a more reliable and consistent AI assistant. While Hermes was groundbreaking in its approach, it leaped without looking too many times; its autonomous nature often led to unpredictable or undesirable outcomes. This essay remains online as a historical record of my experience with Hermes, but I no longer use or recommend it. About six weeks ago, I wrote a short essay explaining that I use artificial intelligence as part of my writing process. The core observation was simple: AI does not originate ideas, but it exposes structure. It mirrors patterns back at you, shows you where an argument repeats itself, amplifies confusion rather than resolving it, and generally behaves less like a writer and more like a diagnostic instrument. ...