Research taste is the capacity to recognise what matters before you can fully articulate why
Research taste is a cultivated form of discernment applied to the decisions that shape a research career: which problems are worth pursuing, which collaborators will amplify your work, and which under-explored areas have genuine leverage. It’s refined judgement, built through practice, critical reflection, and honest assessment of outcomes. A researcher with good taste consistently works on consequential problems; one without it can be technically capable and still directionless.
Research taste
One-sentence definition: Research taste is a cultivated form of discernment: the capacity to identify which problems, collaborators, and opportunities are worth pursuing, developed through iterative practice and critical reflection on your own judgements.
The AI researcher Chris Olah divides being a good researcher into taste, the ability to pick good problems and good ways of attacking them, and technique, the ability to carry the work out (Wiblin, 2021). He treats taste as something you can deliberately develop, and his notes on it offer concrete exercises for doing so (Olah, 2021). I find it useful to think of taste working in three places:
- identifying interesting problems
- recognising good people to work with
- spotting neglected but high-leverage opportunities
What these share is an act of recognition, discerning value before it has been fully demonstrated. That’s what separates taste from technical skill: skill gets better at doing what you’ve already decided to do, and taste shapes the decisions themselves.
The difference between raw and refined intuition matters here. Research taste includes critical reflection on the impulse to pursue certain problems, asking which impulses come from a problem’s importance and which from less useful drives: status, novelty, or the safety of well-trodden ground. Developing taste means learning to interrogate your own judgements while still trusting them enough to act. It improves through making choices, watching what happens, and revising the implicit criteria behind those choices. Olah’s point is that this feedback is slow, since testing a research idea can take months, which is why taste takes years to build and why watching researchers whose taste you can observe is part of how it’s acquired.
Taste in doctoral supervision
Research taste is probably most visible in doctoral supervision. A supervisor with well-developed taste can help a student tell problems that are tractable and important from problems that are merely researchable: the difference between asking “can this be done?” and “is this worth doing?” Modelling that distinction, and making room for students to develop their own sense of it, is part of what supervision is for.
The same capacity matters in research mentorship more broadly. Helping early-career researchers develop independent judgement about what matters means they do more than replicate existing approaches or follow the field’s current preoccupations.
Taste in AI-assisted research
How a researcher directs AI is an expression of their taste: knowing what to ask, which threads to pursue, and what to set aside. A researcher with well-developed taste will use AI to amplify consequential work, and one without it risks producing volume without direction. AI tools make taste more visible, because the choices that used to be hidden inside slow work now show up in what the researcher asks the tool to do. The post on the PhD as a process of becoming treats how a researcher uses AI as developmental evidence, and research taste is what that evidence is evidence of.
Whether taste transfers, and whose values it encodes
Research taste is partly domain-specific: what counts as an important problem in occupational therapy research differs from what counts as important in epidemiology. Whether the underlying capacity (interrogating your own motivations, recognising genuine novelty, identifying high-leverage gaps) transfers across domains isn’t settled.
There’s also a structural tension. Research taste is partly cultivated by internalising a field’s value system, and those value systems aren’t neutral. Academic incentive structures, including publication metrics, grant criteria, and disciplinary prestige hierarchies, shape what “important problems” look like in practice (see the research industrial complex). Taste developed within a field can systematically undervalue certain kinds of question, and recognising that is itself part of what good taste requires.
Sources
- Olah, C. (2021, January 9). Research taste exercises. colah’s blog. https://colah.github.io/notes/taste/
- Rowe, M. (2025). Taste and judgement in human-AI systems.
- Wiblin, R. (Host). (2021, August 11). Chris Olah on working at top AI labs without an undergrad degree [Audio podcast episode]. In 80,000 Hours Podcast. https://80000hours.org/podcast/episodes/chris-olah-unconventional-career-path/
Notes
- The broader concept of taste as discernment, aesthetic and cultural, is covered in the taste and judgement essay; this note applies it specifically to research practice.