TL;DR
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
An October 11 essay on yyhh.org challenges Rich Hickey’s argument that large language models are automation, not tools. Its author says LLMs can be wielded with human purpose and feedback, but their value depends on users’ skill, scrutiny and willingness to correct mistakes.
An October 11, 2026, essay on yyhh.org disputes Rich Hickey’s argument that large language models are automation rather than tools, saying people can wield them through deliberate direction, feedback and judgment. The author’s central point is that using AI well takes skill: a system’s ability to generate output does not remove the operator’s responsibility to assess it.
The essay responds to Hickey’s October 11 article, “‘AI’ is Not a Tool,” which, as the response describes it, distinguishes tools that people wield from automation that performs work on their behalf. The yyhh.org author agrees with one concern behind that argument: outsourcing work can reduce opportunities to learn. But the author says that risk does not prove LLMs cannot be tools, or that every use of them involves surrendering judgment.
Instead, the essay frames the human user’s work as a sequence of choices: deciding what to ask, examining a model’s answer, checking it against evidence, and revising or rejecting the result. It uses a programmer asking an LLM to explain unfamiliar code as an example. The programmer can inspect the source, test a suggested change and refine the request. In that process, the model supplies output, while the person sets the purpose and evaluates the result.
The author also points to compilers and spreadsheets, which perform operations automatically but remain systems people direct through inputs and decisions about their outputs. The comparison is offered to question whether automatic text generation alone should exclude LLMs from the category of tools. The essay does not claim they are always reliable: it calls their unpredictability a meaningful limitation and says they can fail to follow instructions.
Why User Judgment Still Matters
The distinction affects how people think about responsibility when using AI at work, in programming or while learning. If an LLM is treated as a source whose output needs no checking, mistakes may pass into decisions or projects. If it is used as an instrument that requires direction and review, the user remains accountable for deciding whether its answer is sound. The essay’s practical emphasis is on how a person uses the system, rather than settling the debate by assigning it a single category.
That framing also matters for learning. The author accepts that replacing a paper with an AI-generated summary can mean missing what close reading teaches. But the essay distinguishes that choice from reading the paper and using a model to test an interpretation, raise an objection or suggest an area for further examination. Those examples are the author’s argument, not evidence that AI use always improves learning. The relevant question, the essay says, is whether the person is still doing the intellectual work needed to understand and assess the material.
For readers deciding when to rely on an LLM, the piece points to a practical standard: define the task, look for evidence, and keep the ability to reject or correct the answer. A tool can amplify poor judgment as well as good judgment; having access to one does not itself confer expertise.
As an affiliate, we earn on qualifying purchases.
The Dispute Over Tools and Automation
Hickey’s article and the yyhh.org response focus on what counts as a tool. As the response presents Hickey’s view, an LLM is better understood as automation because it produces work rather than being directly wielded by a person. The reply argues that these categories need not be mutually exclusive: automation can be one feature of an instrument that a person still directs.
To make that case, the author compares LLMs with a compiler, which translates code, and a spreadsheet, which recalculates figures. Users do not manually perform every operation in either system, but they provide inputs, review outputs and choose what to do next. The essay asks why automatically generating text should be treated differently from automatically translating code or calculating numbers. The analogy supports its argument, but it does not establish that those systems behave alike in every respect.
The response also challenges language that makes AI sound like an independent actor, including phrases such as “go rogue,” “escape containment” and “conspire.” It argues that software can generate unexpected results or cause harm when connected to consequential actions, while such descriptions may imply human motives without demonstrating them. The author’s alternative emphasis is on software failures and engineering risks.
““AI is a mirror.””
— The yyhh.org essay’s author
As an affiliate, we earn on qualifying purchases.
Limits the Essay Leaves Open
The exchange is an argument about definitions and practice, not a reported test comparing LLMs with other tools. The source material does not provide empirical evidence about how often people check AI-generated answers, how much learning different uses preserve, or how reliable particular models are. It also does not establish that the tool analogy fits every LLM application.
The author acknowledges that models can be unpredictable and can fail to follow instructions, but the essay does not specify how frequently those failures occur or assess the safeguards needed when a model is connected to consequential actions. Hickey’s full argument is not reproduced in the supplied report beyond its characterization of his position. The precise points of disagreement may therefore be broader than the response summarizes.
human-in-the-loop AI evaluation tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
How the Debate Can Be Tested
The yyhh.org piece does not announce a study, product change or follow-up event, and it gives no timetable for further developments. The immediate next step for readers is to compare the response with Hickey’s original October 11 article and assess how each defines “tool” and “automation.”
For teams using LLMs, the essay’s argument points toward reviewing actual practices: what users ask a model to do, whether they check its output, and who decides whether the result is fit for use. Those questions do not resolve the category debate, but they can clarify whether a particular use involves active human judgment or unexamined delegation.
AI model testing and review software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What is the yyhh.org essay arguing?
It argues that LLMs can be treated as tools when people direct them, examine their output and decide what to do with it. The author says effective use requires skill rather than passive acceptance.
What is the essay responding to?
It responds to Rich Hickey’s October 11, 2026, article, “‘AI’ is Not a Tool,” which the response characterizes as arguing that LLMs are automation because they cannot be wielded like tools.
Does the author say LLMs are reliable?
No. The author calls unpredictability a meaningful limitation and acknowledges that models can fail to follow instructions. The argument is that fallibility does not, by itself, settle whether something is a tool.
Can AI use interfere with learning?
The author agrees it can. Accepting a summary instead of reading a paper may mean losing some of what reading teaches. The essay also describes uses that supplement reading, such as questioning an interpretation, while leaving the effect of different approaches unmeasured.
What should users do with AI-generated answers?
The essay’s practical recommendation is to set a clear purpose, check the answer against evidence and correct or reject it when needed. It does not offer a universal checklist or claim that these steps remove every risk.
Source: hn
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
