Il prossimoProssima guida
Asking AI to Explain Things at Your Level
Intelligenza artificiale linguistica
GUIDA alle applicazioni
Asking AI to Critique Your Work
A critique prompt works better when it names the work’s audience, purpose, standards, and the kind of feedback requested.
In questa pagina3 minuti di lettura
Panoramica
Asking for specific weaknesses and evidence can reduce vague praise, but AI feedback is not guaranteed to be candid, correct, or complete and should be checked against the work and relevant expertise.
Immersione profonda
A broad request such as “What do you think?” can yield general reactions. A critique prompt can instead define the intended reader, purpose, evaluation criteria, and scope—for example, ask the model to identify unsupported claims, unclear structure, missing counterarguments, or likely reader questions. Request prioritized findings and concrete passages so feedback is actionable. You can ask for a skeptical review or for the reviewer to look only for weaknesses, but this changes the requested stance rather than guaranteeing accuracy. Research on sycophancy has found that some assistant models may favor responses that align with user beliefs, and humans may sometimes prefer agreeable wording over correct criticism. The effect varies by model, task, and study setup. Separate diagnosis from rewriting. First ask the model to list issues and explain why they matter; then decide which critiques are valid before requesting revisions. Ask it to quote or point to the relevant text, distinguish factual questions from style preferences, and state uncertainty when it lacks evidence. A second review using a different rubric can reveal omissions, but multiple model opinions are not independent ground truth. For important work, compare feedback with a rubric, subject-matter expert, editor, or intended readers. Protect confidential material when uploading drafts. Treat AI critique as one source of suggestions, not an authority on truth, originality, or professional standards. The author remains responsible for the final revision.
Impatto strategico
Scelte di build
La progettazione a livello di applicazione determina se l’intelligenza artificiale migliora i risultati reali.
Team e flusso di lavoro
Una buona integrazione del flusso di lavoro crea guadagni di produttività di cui gli utenti possono fidarsi.
Rischio e sicurezza
I casi d'uso ben definiti riducono l'affaticamento dovuto al cambiamento e il rischio di implementazione.
The Future of Asking AI to Critique Your Work
Critique tools may integrate rubric-based review, citations to source text, and multi-pass editing. They will still need evaluation for factual accuracy, bias, and agreement with expert or audience judgments. Research on sycophancy and critique quality spans different models and tasks, so findings should not be generalized without testing. Future workflows should make the evidence behind each suggested criticism easier to inspect. Human feedback and subject expertise will remain important in deciding which comments are useful in practice across different fields.
Implementazione nel mondo reale
A researcher asks for unsupported claims and missing evidence in a draft, with the exact sentences identified.
A job applicant asks whether a cover letter addresses the role criteria rather than asking if it is “good.”
A writer first requests a prioritized critique, then chooses which suggestions to incorporate.
A team compares AI feedback with an editor’s rubric before revising a public report.
Rischi e guardrail
Automatizzare un processo interrotto può amplificare i problemi esistenti.
I team potrebbero automatizzare eccessivamente e rimuovere il necessario giudizio umano.
La qualità può variare se i risultati non vengono valutati continuamente.
Tabella di marcia per l'implementazione
Mappa il flusso di lavoro corrente e identifica la fase di maggiore attrito.
Definisci checkpoint umani prima dell'automazione completa.
Formare gli utenti su prompt, percorsi di escalation e standard di qualità.
Tieni traccia dei risultati a livello di attività per confermare il valore duraturo.
Continua a esplorare
Free newsletter
Get the daily AI briefing
Three verified AI stories every weekday morning, written in plain English. Free forever, no ads.
One email each weekday. Unsubscribe in one click. We never sell or share your address.
Test yourself
Take the Asking AI to Critique Your Work quiz
Instant feedback on every answer, and a shareable certificate with a verifiable ID once you pass a course.
Support free AI education. AI Understanding is a 501(c)(3) nonprofit — no ads, no paywall, ever. Make a donation
Domande frequenti
What is Asking AI to Critique Your Work?
A critique prompt works better when it names the work’s audience, purpose, standards, and the kind of feedback requested. Asking for specific weaknesses and evidence can reduce vague praise, but AI feedback is not guaranteed to be candid, correct, or complete and should be checked against the work and relevant expertise.
How can a reviewer make criticism more actionable?
Specific examples and rationale help the author assess feedback.
What does the cited sycophancy research suggest?
The study observed this tendency across particular models and tasks.
Why separate diagnosis from rewriting?
Reviewing critique first avoids automatically accepting unsupported revisions.
Are two model critiques independent ground truth?
Multiple generations do not replace expert or source validation.
Continua a imparare
Guide correlate
Altre guide selezionate per questo argomento