AI is a tool and using it well takes skill
The article argues LLMs are tools whose value depends on skilled use, not automation; the lone comment adds that prompting skills shift quickly, so today's advantage may not last.
The post responds to Rich Hickey's argument that LLMs are automation, not tools, because they cannot be wielded. It contends that LLMs can be directed through iterative feedback—asking, challenging, testing, rejecting—and that difficulty, unpredictability and misuse do not disqualify something from being a tool.
It also criticizes anthropomorphic language like 'go rogue' and argues the practical distinction is skilled wielding versus uninformed delegation.
Key takeaways
- Hickey defines tools as wielded instruments and automation as something else; the author says LLMs can be wielded by specifying inputs, inspecting results and correcting course.
- The author compares LLMs to compilers and spreadsheets, arguing automation can be a feature of a tool rather than a separate category.
- A programmer can ask an LLM to explain code, challenge the explanation, inspect source, test a change, reject it and refine the request—a human-directed feedback loop.
- Accepting a summary instead of reading a paper can sacrifice learning, but the author says LLMs can also help readers question interpretations and explore objections.
- Anthropomorphic terms such as 'go rogue' and 'escape containment' risk implying independent agency; the real engineering issues are unexpected outputs, instruction-following failures and harmful access.
- The useful distinction, the author argues, is between wielding an instrument with skill and delegating work without adequate understanding or oversight.
What people think
Analysis details
20 comments analyzed. Analyzed .
The sample selects active discussion branches and includes available parent comments. It may omit parts of the full thread. The analysis included 20 of 46 usable stored comments because of the input limit. Selected themes do not measure community opinion or how common a view is.
Debate over AI's impact on human cognition and education
jgord warns that rapid LLM adoption may cause a decades-long dip in human cognitive ability and education, comparing it to smartphones and social media; Rudybeka predicts future humans will be more capable due to biology, education, drugs, nutrition, and access; oblio counters with PISA declines since 2015; jansport123 foresees a bimodal society and eventual loss of human cognitive superiority; Stevvo reports learning more with LLMs but notes many still share apocalyptic concerns.
Durable skill in using AI: knowing what to ask and domain expertise
extr says workflow tweaks and context-management recipes become obsolete, leaving 'knowing what to ask' and clear communication as the durable meta-skill; misterderpie compares this to Google Search and argues domain expertise helps articulate problems; whatever1 claims AI skill is short-lived because simple prompts can now root-cause logs; onlyrealcuzzo counters that knowing how to ask is still a skill; GuB-42 and dv_dt use tool, tutor, and textbook analogies to argue users should learn the tool's limits rather than avoid it.
Sycophancy and 'AI psychosis' as model limitations
nfw2 describes 'AI psychosis' and says frontier models can be lobotomized when sycophancy is rewarded and builds up in context, requiring persistent intention to resist; jgord asks for symptoms and notes the difficulty of LLM-dieting when the model is correct 92% of the time.
Token costs and economic limits of advanced AI use
Animats warns that multi-agent systems consume many tokens and AI may hit economic limits; progress on hard mathematical problems came at high token cost exceeding some college math department operating budgets.
Dependence on proprietary LLMs and loss of retained skills
0x696C6961 and qmmmur argue that removing the LLM reveals lost skills and that users may confuse familiarity with understanding; sixtyj notes dependence on something owned by few, no big productivity progress despite daily use, and projects not remembered; jansport123 calls for an equitable society so AI benefits the many over the few.
Evidence and verifiability of AI skill and cognitive claims
troupo asks for concrete, verifiable, replicable examples of AI skill and contrasts LLM unpredictability with deterministic compilers; oblio cites PISA declines since 2015 as evidence against optimism about LLMs; jgord asks nfw2 to describe AI psychosis symptoms, questioning the evidence base.