[HN Gopher] Average is all you need
___________________________________________________________________
Average is all you need
Author : AlexC04
Score : 85 points
Date : 2026-04-13 16:36 UTC (4 days ago)
(HTM) web link (rawquery.dev)
(TXT) w3m dump (rawquery.dev)
| drfloyd51 wrote:
| If average is all we need, then anyone can do it. What value do I
| add? How does an employee differentiate themselves?
|
| Why didn't the boss ask the AI for the charts to begin with?
|
| Everyone's income is going to be below average, because they got
| fired.
| CodeyWhizzBang wrote:
| Not everyone can be average. Half of people will be below
| average.
|
| I might not agree with the point, but I can see that idea that
| many things just need to be "good enough" (which we might
| define as "average") and we save our real expertise for the
| things that really matter.
| sva_ wrote:
| > Half of people will be below average.
|
| s/average/median
| jagged-chisel wrote:
| I don't believe this is a meaningful distinction when we're
| not going to agree on how to judge performance of software
| engineers. If this were solely about income, it might be an
| important distinction.
| wongarsu wrote:
| The article assumes a normal distribution, making the
| distinction moot
|
| But it is useful to question whether that is true in all
| cases. The cases that aren't normal-distributed might be
| exactly the cases where it pays off to be neither average
| or median
| programjames wrote:
| It is almost never true. If you filter people you're
| going to get a Pareto distribution.
| skeeter2020 wrote:
| there is a major shortcoming in this assumption;
| everything we've seen related to the internet and
| technology in general suggests there is rarely a normal
| distribution. I think it's way more valuable ato frame
| the questions as a long tail (pareto) distribution and a
| "good enough" cut-off point.
| paulddraper wrote:
| Median is a type of average.
|
| Though usually "average" implies arithmetic mean.
| jerf wrote:
| Reducing the amount of time I spend on the average code has
| meant I'm spending more time adding my above-average
| contributions to the code base. Amdahl's law, basically.
| Reducing the amount of time spent on one task means the
| percentage of time spent on the others increases.
|
| How stable that is on the long term, I don't know any more than
| the next guy, but it is where I'm contributing now.
| bluegatty wrote:
| The power saw makes average cuts, it didn't disemploy
| carpenters, we just made better homes.
| HWR_14 wrote:
| We make more homes, but I would say the construction of the
| average home is worse after the invention of the power saw
| than before it.
| bluegatty wrote:
| Good gosh no.
|
| That's like saying 'cars were better made in the 1950's
| because they used tons of steel'. Like they were 'heavier
| and more robust' - but that doesn't mean better.
|
| Foundations are way better, more robust, especially
| weatherized. Windows today are like magic compared to
| windows 100 years ago.
|
| What we do more poorly now is we don't use wood everywhere,
| aka doors, and certain kinds of workmanship are not there -
| like winding staircases, mouldings - but you can easily
| have that if you want to pay for it. That's a choice.
|
| AI is power and leverage, it will make better things as
| long as it's directed by skilled operators.
| HWR_14 wrote:
| Yes, houses got better because materials got better.
| Windows are better. But the construction of the houses is
| worse.
|
| The precision of how the wood or material meets is worse
| (when cut at the site). There is a huge amount of sloppy
| work in modern construction.
| kaashif wrote:
| I'm interested in how one would prove that one way or
| another.
|
| It seems to me that in the past there probably was lots
| of shoddy workmanship and just no-one paid attention to
| it.
|
| But I have no proof of that.
| motoroco wrote:
| this is not true in my experience. prefab kits of all
| sizes (from sheds to houses to barns, like were once
| possible to order from a Sears catalog) have worse
| tolerances than a carpenter working on site. you can
| measure 3 times and cut perfectly, and still end up with
| a few mm gap (or sometimes worse) after tiny errors
| accumulate as you assemble piece after piece. it
| _requires_ measuring as you go and cutting on site to
| handle this small amount of drift and to really produce
| something of high quality. it doesn't come in a box
| marcosdumay wrote:
| No, it doesn't. The power saw makes perfect cuts. That's why
| carpenters use them.
| roenxi wrote:
| That isn't a sane starting point; if a corporation's strategy
| is to only hire above average employees they're going to fail.
| Enron springs to mind. Corporations generally take average
| people and give them a reasonably well defined scopes of simple
| work to complete that adds value. The bigger the corporation
| the more difficulty they have handling even the standard
| deviation above average differently to the one below; almost
| everyone just becomes a human resource to be swapped around
| based on social factors.
|
| The people who need to be above average and exceptionally are
| senior management and maybe a few bright sparks in middle
| management. Most of the value-add happens there that builds
| social machines that then do the work.
|
| > If average is all we need, then anyone can do it.
|
| Pretty much, yes. That is why the range of salaries on offer is
| pretty compressed compared to the range of returns capitalists
| get.
| drfloyd51 wrote:
| > The people who need to be above average and exceptionally
| are senior management and maybe a few bright sparks in middle
| management. Most of the value-add happens there that builds
| social machines that then do the work.
|
| That is the dream. Upper management can get software made
| without talent.
|
| But is seems to be the greatest ideas in the last 30 years
| didn't start in board rooms. They started with a couple
| coders creating a new idea.
|
| No boardroom could have invented Google. It was so
| fundamentally different than what other search engines were
| doing.
|
| We have this myth that upper management is so important. It
| is as the business grows in size, they are excellent for
| coordination. But ideas come from people closer to the
| problems.
| j45 wrote:
| The average of quality isn't always available in all people.
| analog31 wrote:
| For that matter, how does a business differentiate themselves,
| if people can write their own software? While we're busy trying
| to replace our employees with AI, our customers are trying to
| replace our products with AI.
| raw_anon_1111 wrote:
| At any tech company with leveling guidelines that I have seen,
| promotions above mid level have never been based on "I codez
| real gud". It's always been based on scope, impact and dealing
| with ambiguity. It's stated differently in different companies.
|
| No one has ever differentiated themselves based on how good of
| a ticket taker they are. Coding especially on the enterprise
| dev side where most developers work has been being commoditized
| since 2016 at least and compensation has stagnated since then
| and hasn't come near keeping up with inflation.
|
| In 2016, a good solid full stack, mobile or web developer
| working in the enterprise could make $135K working in a second
| tier city. That's $185K inflation adjusted today. Those same
| companies aren't paying $185K for the same position.
|
| My one anecdote is that the same company I worked for back then
| making $125K and some of my coworkers were making $135K just
| posted a position on LinkedIn with the same requirements (SQL
| Server + C#) offering $145K fully remote.
| Ancapistani wrote:
| > At any tech company with leveling guidelines that I have
| seen, promotions above mid level have never been based on "I
| codez real gud". It's always been based on scope, impact and
| dealing with ambiguity. It's stated differently in different
| companies.
|
| I 100% agree here.
|
| AI has been a huge boon for me personally, because I stopped
| spending most of my writing code years ago. I was reviewing
| code, writing procedures, handling incidents, and generally
| just looking for pain points across the entire company and
| solving them before they became critical.
|
| Those skills have transferred directly to working with AI.
| JackSlateur wrote:
| Average is all we need ! I mean, working 50% is enough, right ?
|
| A car that starts 50% of the time ?
|
| A plane that stops on 50% of the flights ?
|
| A pacemaker that beats only 50% of the time ?
|
| David Goodenought said that average is enough ..
| CodeyWhizzBang wrote:
| A car that starts 50% of the time isn't "average". The average
| new car starts more or less every time. (And if you said 'modal
| average', I'd say the modal average new car starts every time).
| JackSlateur wrote:
| It is not average today because people in the past tried to
| do better, not average things
| CodeyWhizzBang wrote:
| I think that's maybe the point of the article:
|
| "Whereas before, average was expensive in terms of both
| time and effort, average became cheap."
| montroser wrote:
| Average is all you need, if your needs are average.
| jagged-chisel wrote:
| But, you see, our needs are above average because we target
| above average exists so we only hire from the top 1% of
| software engineers, blah, blah, yadda, yadda, etc.
|
| The Business simply cannot admit that it's really doing nothing
| above average. If they did, investment dries up.
| Axel2Sikov wrote:
| That is correct. And if you need more you can get it as well.
| ljhsiung wrote:
| I liken it to the Ikeaficiation of furniture. To a great
| majority, such as my college self, it was preferable and
| desirable. As I've made more money, I've wanted something
| better.
|
| There's a market for both, but the furniture slop of Ikea is
| dominant.
| mpalmer wrote:
| This is not only average. This is actual magic. So
| let's be real: the SQL is average. The joins are average. The
| chart is average. And that took us less than 5 minutes and that
| was amazing, that is the entire point. You did not
| need a data engineer to model your HubSpot data, or a meeting to
| agree on whether it should be last-click or first-click or linear
| or time-decay or whatever. You needed a query,
| written fast, on data you already own. Your LLM wrote it. You
| confirmed it made sense. Your manager got a link.
| Honestly, average is clearly magic; prove me wrong.
|
| I'll give it a go. This is generated slop, and the poor, factory-
| made quality of the writing undercuts every aspect of the
| argument.
|
| It is like nails on a chalkboard.
| Axel2Sikov wrote:
| Author here; I suppose the... side eye awkward monkey meme was
| a bit lost on you; it was written on purpose. Funnily enough.
| Everything is slop if you want it to be slop. This however, was
| written by hand my little hands. Now I might be a bad writter -
| that is indeed another subject.
| montroser wrote:
| This tracks. Tasks that used to be a day or two of grunt work are
| now an hour with Claude.
|
| And there is a lot of that type of work to do if you're trying to
| grow a business. But, _something_ in there should be trying to be
| _exceptional_ or else you have no moat. Claude will probably not
| be able to breeze through that part with the same amount of
| ease...
| bashwizard wrote:
| The majority of devs are average. What a shocker.
| programjames wrote:
| The majority of any filtered group are below average. Imposter
| syndrome isn't a thing, 80% of people really did just barely
| make the cutoff.
| tsimionescu wrote:
| > But this is a pain, first because, if you do anything that is
| not selling a product online that people can buy right when they
| click a button, it is a drag to create those attribution models
| effectively: is it last click, first click, weighted
| attribution... who knows. Nobody knows. Everybody gives up and
| just adds it to a dashboard and pretends it makes sense.
|
| Yes, thinking about your data and how to check it is so annoying.
| Much better to do something average, see if the result puts you
| in a good light, and share that insight into your company's
| working with ~~everyone on the internet~~ your boss.
|
| Rarely have I seen "we help you create meaningless slop more
| easily" advertised so explicitly. Or is this also average?
| winterbloom wrote:
| how do you know those queries are actually correct without domain
| knowledge?
|
| Do you know enough about JOINs and how they work to be able to
| break those big queries down and figure out whether they are
| doing exactly what you're asking for in English?
| antihipocrat wrote:
| You don't, and if businesses start using vibed reports for
| regulated reporting then I guess we'll see soon what the courts
| say about that
| Ancapistani wrote:
| Litigation aside for a moment - I'm not sure vide-coded
| reporting could be much worse than what I've seen from early-
| career analysts in past companies.
| Axel2Sikov wrote:
| You don't. But you can still check ? -\\_(tsu)_/-
| throwaway98797 wrote:
| adding LLMs to the incompetent doesn't transform them
|
| if anything it makes the world more dangerous
|
| a reckoning is coming
|
| the top decile will be janitors for the rest
| segh wrote:
| Being average is a just stage LLMs pass through as AI makes its
| way towards 'expert' and 'super human' levels.
| marginalia_nu wrote:
| LLMs are trained to predict tokens on highly mediocre code
| though. How will it exceed its training data?
| Retr0id wrote:
| Humans can decide to write above-average code by putting in
| more effort, writing comprehensive tests, iteratively
| refactoring, profile-informed optimization, etc.
|
| I think you can have LLMs do that too, and then generate
| synthetic training data for "high-effort code".
| marginalia_nu wrote:
| Well state of the art LLMs sure can't consistently produce
| high quality code outside of small greenfield projects or
| tiny demos, which is a domain that was always easy even for
| humans as there are very few constraints to consider, and
| the context is very small.
|
| Part of the problem is that better code is almost always
| less code. Where a skilled programmer will introduce a
| surgical 1-3 LOC diff, an incompetent programmer will
| introduce 100 LOC. So you'll almost always have a case
| where the bad code outnumbers the good.
| Retr0id wrote:
| Current LLMs do tend to explode complexity if left to
| their own devices but I don't think that's an inherent
| limitation. Mediocre programmers _can_ write good code if
| they try hard enough and spend enough time on it.
| monkaiju wrote:
| Thats because humans have "understanding" they can use to
| assess quality, without understanding "trying harder" just
| means spending more "effort" distilling an average result,
| at best over a larger sample size.
| utopiah wrote:
| Who are you to question our faith? /s
| bluGill wrote:
| Because you ask it to improve things and so it produces
| slightly better than average results - the average person can
| find things wrong with something, and fix it as well. Then
| you feed that improved result back in and generate a model
| where the average is better.
|
| /end extreme over optimism.
| movedx01 wrote:
| Probably the same way other models learned to surpass human
| ability while being bootstrapped from human-level data -
| using reinforcement learning.
|
| The question is, do we have good enough feedback loops for
| that, and if not, are we going to find them? I would bet they
| will be found for a lot of use cases.
| utopiah wrote:
| This is yet another ad, it's tiring.
|
| It's a post claiming average AI is useful... by a for-profit
| "data platform with a CLI that LLM agents can use directly". What
| are they going to do? Criticize the whole industry they are
| selling to?
| Axel2Sikov wrote:
| I did not post it. I did not intend for it to be posted here -
| really. It just happened someone did see it and posted it. So I
| did not advertise anything :-)
| antisthenes wrote:
| Another writer trying to redefine a common english word to mean
| whatever they want it to mean at the time.
|
| Pass.
| jihadjihad wrote:
| > You did not write a single line of SQL. You did not set up an
| attribution model. You asked a question, in English, and got a
| table.
|
| But nobody bothered to check if it was correct. It might _seem_
| correct, but I 've been burned by queries exactly like these
| many, many times. What can often happen is that you end up with
| multiplied rows, and the answer isn't "let's just add a DISTINCT
| somewhere".
|
| The answer is to look at the base table and the joins. You're
| joining _customers_ to two (implied) one-to-many tables,
| _charges_ and _email_events_. If there are multiple _charges_
| rows per customer, or an email can match multiple _email_events_
| rows, it can lead to a Cartesian multiplication of the rows since
| any combination of matches from the base table to the joined
| tables will be included.
|
| If that's the case, the transactions and revenue values are
| likely to be inflated, and therefore the pretty pictures you
| passed along to your boss are wrong.
|
| Further reading, and a terrific resource:
|
| https://kb.databasedesignbook.com/posts/sql-joins/#understan...
| Axel2Sikov wrote:
| Ok but... nobody said you didn't had to check either(?).
| paulryanrogers wrote:
| How do you check if you don't have any other view into the
| data but SQL and you don't know SQL?
| Axel2Sikov wrote:
| Same way you do today; you trust whoever wrote the query.
|
| I do not sell a wrapper on top of some LLM; you can
| absolutely write your SQL directly. There is an engine,
| there are iceberg tables. You can just live your best life
| doing your own SQL by hand.
|
| Now if you couldnt do it before and you have a sensible
| understanding, you can likely do a bit more with the CLI
| tooling. And if you know a lot more, you can still do that.
| The queries are not hidden, or abstracted, If you need them
| they will be saved - transparently in SQL.
|
| So I dont know what is the answer to the question "how do
| people do things they don't know how to do" ?
| paulryanrogers wrote:
| > So I dont know what is the answer to the question "how
| do people do things they don't know how to do" ?
|
| The statue quo had been to learn SQL or ask a human you
| trust to check their own work, which hopefully you can
| reuse.
|
| Now it's ask AIs that are intentionally a bit random, and
| less likely to (or incapable of) check(ing) their work.
| Perhaps without seeing the SQL at all, requiring to trust
| it for every interaction. And in a culture that moves so
| fast that there is no checking by any(one|thing).
| XenophileJKO wrote:
| If you think a language model can't check their work,
| then you are using the tools wrong. Plain and simple.
|
| Modern models are quite capable at surfacing and
| validating their assumptions and checking correctness of
| solutions.
|
| Oversight helps you build confidence in the solutions. Is
| it perfect, no.. but way better then most engineers I
| also ask to check things.
| Bridged7756 wrote:
| No they don't. To be able to "check one's work", implies
| that they can be held accountable, that they can tell
| apart right from wrong, when in reality they're merely
| text predictors.
|
| If you think an LLMs can check their work, then you are
| doing a terrible job at writing software. Plain and
| simple.
|
| They even go as far as "cheating", so tests fail, writing
| incorrect tests, or straight out leaking code (lol) like
| the latest Claude Code blunder. Is this the tool the
| original comment "is using wrong, plain and simple"? Or
| do you have access to some other model that works in a
| wildly different way than generating text predictions?
| chairmansteve wrote:
| Yes. The correct answer is to ask an SQL expert to write the
| query. An LLM will make the expert much faster at writing the
| script, maybe.
| mattmanser wrote:
| It doesn't.
|
| I can write that script faster than I can write the text
| asking the AI to write the script as SQL is concise and my
| IDE has auto-complete.
| busfahrer wrote:
| This seems like a nice context to mention Sturgeon's law:
|
| > ninety percent of everything is crud
|
| https://en.wikipedia.org/wiki/Sturgeon%27s_law
| throw310822 wrote:
| Why average? I've always taken pride in my work and developed
| things that went beyond the expectations of the management and of
| the final users. Now I'm using LLMs a lot and I've been able to
| do much more than I used to- I find them great coworkers,
| technically very knowledgeable, patient and fast. I provide the
| big picture, keep an eye on the architectural soundness and code
| quality, and design the features. The LLM does the rest. The
| results are way above average.
| sdevonoes wrote:
| Nobody cares
| llmssuck wrote:
| Just want to add this is my experience as well. Just solid
| coworkers. Of course they mess up sometimes, but easier to fix
| up than with humans and their politic and egos. I find I can
| actually reason for once instead of always fighting and
| deferring to whomever has The Biggest Opinion and not rarely
| just the loudest voice.
|
| I think many people here work at nice, large places with
| reasonable and knowledgeable colleagues that are cooperative
| and mostly rational and try to do the right thing. In my
| experience that is not a common or widespread thing. Of course
| I only have small to medium business experience, but that's
| still a pretty good chunk of the economy. LLMs are an absurd,
| ridiculous win in those kinds of environments.
| movedx01 wrote:
| Average is only a tombstone of someone having failed to do
| better. And settling for average means pulling down.
|
| When it comes to bs dashboard where "average is all you need",
| maybe the "better than average" result would be asking yourself
| if it's even worth doing in the first place?
| chriswait wrote:
| I always find it a bit weird to see posts on the front page where
| all the comments disagree with the central premise of the
| article. In this case the post is an ad advocating for executing
| code you didn't write and handing the results to your manager.
|
| It makes me wonder if Hacker News has a silent majority of people
| who would actually use AI in this way without wanting to admit
| it, and a vocal minority of people who wouldn't.
| Ancapistani wrote:
| I'll admit that there are definitely times where I decide it's
| fine to roll with it blind. It's not often, not for critical
| paths, and definitely not where I don't have a good
| understanding of the blast radius if it fails spectacularly -
| but you'd be surprised how often it's easier and faster to fix
| it if it breaks than it would be to make sure it's not broken.
| fedeb95 wrote:
| yes. Most people are upset and fear losing their job because they
| feel their job is sub-par. In reality, that's for most of them
| impostor syndrome, for some could be a wake up call.
| kfk wrote:
| This is all fun and games when you work with toy data samples.
| But most organizations are more complex, they have to match
| invoices from SAP with opportunities in Hubspot; or they have to
| consider that little sales territory exception for the sales guy
| in Munich to calculate the proper commission projection; or they
| have custom tables in Salesforce with 0 documentation; or... you
| get my point.
|
| Not all context is documented, and some context has to even be
| changed because it doesn't make sense.
|
| I find AI very useful, but I think a lot of this AI SQL products
| are misleading.
| xg15 wrote:
| I think the author should be introduced to (or reminded of) the
| tale of the average from the US Air Force [1]. Social reality is
| high-dimensional and the "normal" thing is actually to be average
| in some dimensions, but strongly non-average in many others. So a
| "perfectly average" family would paradoxically be an outlier
| themselves.
|
| I think this is important, because if his hypothesis is right,
| then LLMs behave differently here: They really _are_ average in
| all dimensions. They are the pilots the Air Force _thought_ they
| had before Daniels made the study.
|
| So if he is right, we'd be changing from a mostly-non-average to
| a mostly-average society, which would really be a massive change
| - and probably not a good one IMO.
|
| [1]
| https://noblestatman.com/uploads/6/6/7/3/66731677/cockpit.fl...
| hackncheese wrote:
| Wow incredibly interesting read, got me thinking about design
| principles and the "average user"
| codethief wrote:
| Interesting link! What was the solution? How did they end up
| redesigning the cockpit?
| xg15 wrote:
| From what I understood, it was the beginning of user-
| adjustable design. Instead of mounting the seat in a fixed
| height with fixed distance to the pedals, etc, they designed
| the cockpit so the pilot could adjust everything himself.
| Basically, what is standard in every car today.
| 9991 wrote:
| In high dimensional spaces, almost all of the volume of a
| sphere is in a thin shell close to the surface.
| localhoster wrote:
| Tbh I dont really agree with your statements. Especially with
| working with data, intention is key. By using an llm, by
| definition, you are loosing intention. And Thai puts you in a
| position where you have to 1) think of exactly what you look for.
| 2) able to understand what the llm generated.
|
| You might say it "still less work" and that's true, perhaps, only
| for the first few times. After a while you _learn_ how to do it,
| and understand how to _think_ with the language of your data.
| With LLMs, you never get this benefit, and also loose your
| ability to judge the LLM's output properly.
|
| But again, that might be enough on your case, or, you simply
| don't _know_.
| skybrian wrote:
| It seems like person receiving the report is already in that
| position. Should everyone make reports themselves?
|
| Let's say you start with a report someone else wrote. It seems
| like you still need to read it and understand what it's telling
| you. Sometimes plotting all the points helps, or drilling down
| and looking at the raw data.
| pc86 wrote:
| This says "Editorial" at the top but has no authorship
| information. Who wrote this?
| myhf wrote:
| Nobody wrote this.
| underlines wrote:
| i fail to understand how text2sql on quite simple data sources is
| anything to write home about 3 years after it came onto the
| market? can someone elaborate?
|
| where it gets interesting is when you have a custom system that
| your LLM surely never saw (custom ERP) that has 50 sometime
| cryptic tables, unclear look up tables and unexplained flags.
|
| something no text2sql solution solved for us.
|
| we built a second mcp that lets the agent look up business logic
| (generated from source code) and then does better queries. that i
| think is something i never read in a blog post about a text2sql
| solution.
| Axel2Sikov wrote:
| There is a missundertanding. This is not text2sql. This is CLI
| first; ideally the tool being operated by Claude Code and
| alike.
|
| You could use claude code for the "text2sql" kind of part, but
| this is not why this tool exists. Nor what the article
| advocates.
| leecommamichael wrote:
| I flat out disagree. If we want things that are better than
| average, we will need to be making things that are not average.
| The machine doesn't learn yet.
| cremer wrote:
| The real risk is not that "LLM SQL" is wrong, it is in fact that
| a man who asked for it can't recognize when it is wrong. Writing
| a query and knowing when JSON silently doubled your rows are
| different skills and LLMs are able to give you only one of them
| (I'm not sure if they really can)
| tqi wrote:
| > Claude doesn't ask you to define an attribution model. It
| doesn't open a whiteboard. It runs [query]"
|
| Why is that a good thing? Claude didn't ask any obvious follow up
| questions, like what determined whether a user got an email or
| not? It is using the ab test terminology in Step 3 without any
| kind of confirmation that this is, you know, a valid test.
___________________________________________________________________
(page generated 2026-04-17 23:01 UTC)