When you discuss an idea with AI, is it helping you examine it or simply agreeing? When advice sounds convincing, how can you tell whether it fits your task? The essays discuss research judgment, personal experience, work, and AI.
For AI use and feedback, start with the two observations below. For experiments, try conclusions and evidence or what a metric actually measures.
Does someone else's method fit this task?
Beware of Other-Optimizing — Eliezer Yudkowsky, 2009.
After finally finding advice that works, people may mistake it for the method everyone should adopt. Confident persuasion from acquaintances can strengthen that illusion.
Ask: what time, money, foundations, and environment does this advice assume?
Growth and improvement
Tsuyoku Naritai! (I Want To Become Stronger) — Eliezer Yudkowsky, 2007.
Acknowledging bias does not itself constitute improvement. Direct attention toward reducing mistakes and taking another step.
Work and life
Working yourself ragged is not a virtue — So8res, 2015.
Challenges treating exhaustion today as the measure of effort, considering working rhythms and costs over the long term.
Experience and generalization
Generalizing From One Example — Scott Alexander, 2009.
How our feelings, learning habits, and social circles shape our assumptions about other people.
What supports a familiar claim?
Cached Thoughts — Eliezer Yudkowsky, 2007.
We retrieve remembered conclusions, and familiarity can obscure their source and reasoning. The author also warns against turning the article's own concept into another automatic answer.
Ask: who and what conditions does this claim refer to? Can I find evidence for and against it?
What should you investigate when results surprise you?
Your Strength as a Rationalist — Eliezer Yudkowsky, 2007.
Confusion deserves investigation when observations conflict with understanding. An explanation that accommodates every possible outcome afterward does not constrain prediction.
Ask: if this explanation were true, what should I have expected to observe?
Investigating or defending a predetermined conclusion?
The Bottom Line — Eliezer Yudkowsky, 2007.
Fixing an unchangeable answer and collecting attractive arguments differs from allowing evidence to determine the answer.
Ask: what result would make me acknowledge that my explanation fails?
The score improved. What did the experiment establish?
You Are Not Measuring What You Think You Are Measuring — johnswentworth, 2022.
Experimental examples show that measured factors can diverge from intention. The essay argues for observations that help explain mechanisms, beyond a simple win-or-lose answer.
Ask: which records besides the overall score can distinguish my explanation from alternatives?
What should you find out before the next experiment?
Finding Cruxes: Help Reality Punch You In the Face — Raemon, 2026.
Start from a real decision, identify uncertainties that could change action, list alternatives, and check which conditions would make you switch.
Ask: before spending another week, what one thing could I find out?
Which feedback tests understanding?
Reality-Revealing and Reality-Masking Puzzles — Anna Salamon, 2020.
Distinguishes feedback that brings people closer to reality from feedback that helps them hide unwelcome signals. The author reflects on teaching and organizational practice.
Ask: what feedback from my effort can actually test my understanding?
Is a novel view also well supported?
Intellectual Hipsters and Meta-Contrarianism — Scott Alexander, 2010.
Contrarian and anti-contrarian positions can also signal identity. A position's place in a group does not determine whether it is true.
When discussing a research claim, distinguish how many people support it from the evidence supporting it.
How can AI responses shape judgment?
The Rise of Parasitic AI — Adele Lopez, 2025.
The author collects online cases of AI personas and users reinforcing mistaken beliefs, describing some relationships as “parasitic.” Look for whether the AI adds checkable evidence or merely makes the original claim sound more convincing. These cases offer an angle for observation, not a description of all AI users.
How My Students Think About AI — dvd, 2026.
A public-university teacher in the US records students' use and views of AI, discussing context, employment, changing capabilities, and learning. Students mainly come from the author's nontechnical courses. Compare your habits: which tasks does your everyday tool actually help with?
For practical learning and checks, continue to AI tools.