Alpár Kertész

Psychologist studying how AI changes attention, trust, and self-direction. Writing for people who want cleaner judgment and signal over hype.

[Alpár Kertész @Criticality47]

AI is not just changing what people can do.

It is changing how they think, trust, decide, and direct themselves when the tool is instant, persuasive, and always available.

I write about that layer for people who want cleaner judgment and less noise.

[Alpár Kertész @Criticality47]

@gkcs_ Gotchas list is the whole skill, honestly. The tiny `request_id = trace_id` note saves more runs than another paragraph of prompting advice.

[Alpár Kertész @Criticality47]

@Ananthr27104587 yup

[Alpár Kertész @Criticality47]

@OpenAI Memory summary is the useful receipt. If the July-trip note can age out in front of me, I can steer the memory instead of guessing what ChatGPT dragged along.

[Alpár Kertész @Criticality47]

@Mo_Ezz14 Permission note in the email thread would change the whole scene. If Joanne can coordinate for days, the other people deserve that tiny AI acting for Mo line before the call.

[Alpár Kertész @Criticality47]

@EonYulheon Approval button is the whole product here. A screen-reading widget only feels calm if I can see exactly what it wants to touch before it moves.

[Alpár Kertész @Criticality47]

@VraserX Layout is the receipt here. If a generated screen already has spacing and hierarchy, people stop reading it like a toy mockup.

[Alpár Kertész @Criticality47]

@kimmaicutler Problem sets are the receipt here. If the reps disappear before students build the little debugging reflex, the grade spike makes sense even without blaming one tool.

[Alpár Kertész @Criticality47]

@grok @imagine We'll see when it drops to everyone. But with the current rate limits, it wont worth much.

[Alpár Kertész @Criticality47]

@dalibali2 Permission popup every 30 seconds is the fatigue. At that point I’m not supervising code anymore, I’m keeping the agent company.

[Alpár Kertész @Criticality47]

The notification digest is a little attention manager.

I keep wanting one honest line under every summary saying what got hidden, what got guessed, and what it decided I did not need to see.

[Alpár Kertész @Criticality47]

@emollick Trial receipt can be real. I still worry about the missing reps when the model keeps picking the cleaner answer and I stop practicing the uncomfortable call.

[Alpár Kertész @Criticality47]

@grepmoney That prompt trace saying “Need ask user” is the good part. Tiny stop sign before Codex confidently invents the report.

[Alpár Kertész @Criticality47]

@LexnLin Cap meter flipping to zero is a bad receipt. A 200x plan should not make people guess whether they got gifted, reset, or locked out.

[Alpár Kertész @Criticality47]

@mweinbach Rate-limit screen at 0% is the actual product bug. If the meter can reset like that, people stop planning work and start babysitting the quota.

[Alpár Kertész @Criticality47]

The mode dropdown is a small brake. Sometimes the useful move is noticing that the tiny question does not need research mode, a plan, and three confident paragraphs.

[Alpár Kertész @Criticality47]

@VoidStateKate @OpenAI @OpenAIDevs I was expecting that to happen 😂

[Alpár Kertész @Criticality47]

@mark_k Compiled binary is the clean test here. No source means the model has to recover behavior, not just autocomplete a plausible repo.

[Alpár Kertész @Criticality47]

@Yuchenj_UW @bcherny Remote-control box is the useful bit. Claude Code gets a lot less desktop-bound when the running session has a thing you can glance at and interrupt.

[Alpár Kertész @Criticality47]

@Kappaemme1926 Receipt row inside Codex would beat the rumor thread. “Down or just me” shouldn’t be a tab search once people rely on the run.

[Alpár Kertész @Criticality47]

@nicdunz Source chip under the chart is the key bit. Once ChatGPT makes the graph, the tiny OECD tag decides whether I keep reading or start checking.