I think that would be Google's approach, before they were waylaid by ad revenue incentives: that question intent can best be inferred by knowing as much about the asker as possible.
But I feel optimizing for lazy question phrasing is infantilizing the userbase and assuming they're incapable of learning how to ask more precise questions.
Which is an endemic problem in the modern web. We should be building systems with low barriers to simple adoption, but whose power scales with a user's expertise.
Instead, we're hyperoptimizing for lowest common denominator, first interaction and as a result putting a glass ceiling on system power.
> that would be Google's approach, before they were waylaid by ad revenue incentives: that question intent can best be inferred by knowing as much about the asker as possible.
I’d rephrase it: this was googles approach because they were waylaid by ad revenue incentives. Why does everyone assume the search bubble is a well intentioned accident?
The linguistic arguments about the nature of truth above don’t cut it for me as any justification that the answer depends on who is asking. If we’re telling chemists that oil and water do mix as mentioned elsewhere in thread, we should probably tell the same to children and just make a slider for the level of additional detail. No problem. Especially if the alternative is a dystopic post truth panopticon.
> systems whose power scales with expertise
Couldn’t agree more that this is what we want/need, but disagree about infantilizing user bases and lowest common denominator.
Platform’s aren’t trying to scale user power with expertise, they want to scale revenue with users, and separately, to deliberately restrict user power/control so that platforms decide what users see. And it’s not necessarily political or linguistic or about the nature of truth.
A simple example of this that’s everywhere is filtering content by sub genre tags. Ever notice that you often see niche content tags like “time-travel” or “mind-bending” but can’t click it? Platforms want the tags for internal analytics naturally, but they want the control of not providing it as a filter, forcing users instead into a fuzzier category like “people also watched” or “top ten this week”.
Why? Because platforms can hide advertised content there, push stuff they pay less license fees for, make operations cheaper with caching, or whatever else.
Because lots of people are ignorant and imprecise.
HN selects from people who have at least a passing knowledge/interest in programming/science, a subpopulation which is already several standard deviations from the mean in specificity and debugging.
> disagree about infantilizing user bases and lowest common denominator
Platforms, with Google as exemplar, evolve in two ways.
1: Features they explicitly choose not to ship, because they're strategically dangerous. See tool-use foot dragging by OpenAi.
2: Features they deprioritize, because they aren't as revenue-impactful as other things.
To me, it feels like Google dropped the ball via the second path.
I'm sure they've been doing a ridiculous amount of cool work behind the scenes on individual context grounding... but once prod was "good enough for ads" the company as a whole wasn't incentivized to do the hard thing and ship more advanced features in search.
Which is how they ended up as legacy as they are, competing against LLM search that's by definition context-native.
But I feel optimizing for lazy question phrasing is infantilizing the userbase and assuming they're incapable of learning how to ask more precise questions.
Which is an endemic problem in the modern web. We should be building systems with low barriers to simple adoption, but whose power scales with a user's expertise.
Instead, we're hyperoptimizing for lowest common denominator, first interaction and as a result putting a glass ceiling on system power.