[HN Gopher] AI agent benchmarks are broken
       ___________________________________________________________________
        
       AI agent benchmarks are broken
        
       Author : neehao
       Score  : 168 points
       Date   : 2025-07-11 13:06 UTC (9 hours ago)
        
 (HTM) web link (ddkang.substack.com)
 (TXT) w3m dump (ddkang.substack.com)
        
       | anupj wrote:
       | AI agent benchmarks are starting to feel like the self-driving
       | car demos of 2016: impressive until you realize the test track
       | has speed bumps labeled "success"
        
       | RansomStark wrote:
       | I really like the CMU Agents Company approach of simulating a
       | real world environment [0]. Is it perfect, no. Does it show you
       | want to expect in production, not really, but it's much closer
       | than anything else I've seen.
       | 
       | [0] https://the-agent-company.com/
        
         | yeahyeahok wrote:
         | Damn. Super bullish on CMU. Somehow, they seem routinely left
         | out of the top CS schools discussion at least in mainstream
         | discourse: MIT, Stanford, Cal, .... Seen a disproportionate
         | amount of stellar research come from there. Also,
         | interestingly, I have met really incompetent people from all
         | the other top 3 schools but am yet to meet an incompetent CMU
         | SCS alum -- wtf are they feeding them in pitsburgh??
        
       | deepdarkforest wrote:
       | It's very funny how many layers of abstraction we are going
       | through. We have limited understanding of how LLM's work exactly
       | and why. We now do post training with RL, which again, we don't
       | have perfect understanding of it either. Then you stack LLMs
       | calls and random tools, and you have agents, and you are
       | attempting to benchmark _those_. (and this exclude voice,
       | computer use agents etc).
       | 
       | It's all just vibes,there is no good general benchmark for agents
       | and i think it's just impossible, there are just way too many
       | degrees of freedom to achieve anything useful. They're just a
       | complicated tool to achieve things. It's like trying to make a
       | general use benchmark of a stack of 10 microservices together. It
       | does not make sense, it just depends on your usecase and your own
       | metrics
        
         | bwfan123 wrote:
         | I can hear echos of an earlier era.
         | 
         | There was yahoo-pipes and web-services frameworks which rhyme
         | with MCP and agentic.
        
           | th0ma5 wrote:
           | Pipes and services in general are reliable but the issues
           | were social and economic. Getting everyone to agree was seen
           | as a great way to poach users and give up control, plus the
           | usual problems with open world vs. closed world assumptions.
           | Thanks for mentioning this!
        
         | rf15 wrote:
         | > We have limited understanding of how LLM's work exactly and
         | why.
         | 
         | blatantly untrue, and as a concept only useful to those who
         | want to sell AI as this "magical thing" that "just works"
        
       | xnx wrote:
       | All benchmarks are flawed. Some benchmarks are useful.
        
         | yifanl wrote:
         | Here's a third sentence fragment: These benchmarks are not.
        
           | suddenlybananas wrote:
           | It's nearly a haiku!
        
             | layer8 wrote:
             | All benchmarks are flawed.       Not all benchmarks are
             | useless.       But these benchmarks are.
        
           | lcnPylGDnU4H9OF wrote:
           | Just want to nit: none of those are sentence fragments, they
           | are complete thoughts with a subject and a predicate. Yours
           | kinda comes close to being a fragment but it really just
           | omits what "are not" (the predicate) is referring to, which
           | is included in prior context.
           | 
           | For example, a fragment with a missing predicate.
        
       | greatpostman wrote:
       | Benchmarks aren't broken, the models can learn anything. If we
       | give them true real world data (physics engine), they will learn
       | the real world. We are going to see artificial general
       | intelligence in our lifetime
        
         | hddbbdbfnfdk wrote:
         | more like in the next two weeks methinks
        
       | jerf wrote:
       | When I was being a bad HN reader and just reacting to the title,
       | my initial impulse was to be placating, and observe that they are
       | probably just immature. After all, for all that has happened,
       | this is still only a couple year's worth of development, and it
       | does tend to take a long time to develop good benchmarks.
       | 
       | However the article does seem to be pointing out some fundamental
       | issues. I'm particularly annoyed by using LLMs to evaluate the
       | output of LLMs. Anyone with enough experience to be writing
       | benchmarks of this sort in the first place ought to know that's a
       | no-go. It isn't even just using "AI to evaluate AI" _per se_ ,
       | but using a judge of the same architecture as the thing being
       | judged maximizes the probability of fundamental failure of the
       | benchmark to be valid due to the judge having the exact same
       | blind spots as the thing under test. As we, at the moment, lack a
       | diversity of AI architectures that can play on the same level as
       | LLMs, it is simply necessary for the only other known
       | intelligence architecture, human brains, to be in the loop for
       | now, however many other difficulties that may introduce into the
       | testing procedures.
       | 
       | Tests that a "do nothing" AI can pass aren't intrinsically
       | invalid but they should certainly be only a very small number of
       | the tests. I'd go with low-single-digit percentage, not 38%. But
       | I would say it should be above zero; we do want to test for the
       | AI being excessively biased in the direction of "doing
       | something", which is a valid failure state.
        
         | potatolicious wrote:
         | > _" I'm particularly annoyed by using LLMs to evaluate the
         | output of LLMs."_
         | 
         | +1, and IMO part of a general trend where we're just not
         | serious about making sure _this shit works_. Higher scores make
         | stonks go up, who cares if it actually leads to reliably
         | working products.
         | 
         | But also more importantly it's starting to expose the fact that
         | we haven't solved one of ML's core challenges: data collection
         | and curation. On the training side we have obviated this
         | somewhat (by ingesting the whole internet, for example), but on
         | the eval side it feels like we're increasing just going
         | "actually constructing rigorous evaluation data, especially at
         | this scale, would be very expensive... so let's not".
         | 
         | I was at a local tech meetup recently where a recruiting firm
         | was proudly showing off the LLM-based system they're using to
         | screen candidates. They... did not evaluate the end-to-end
         | efficacy of their system. At all. This seems like a theme
         | within our industry - we're deploying these systems based
         | purely on vibes without any real quantification of efficacy.
         | 
         | Or in this case, we're quantifying efficacy... poorly.
        
           | rsynnott wrote:
           | > +1, and IMO part of a general trend where we're just not
           | serious about making sure this shit works.
           | 
           | I suspect quite a lot of the industry is actively _opposed_
           | to that, because it could be damaging for the "this changes
           | everything" narrative.
        
         | alextheparrot wrote:
         | LLMs evaluating LLM outputs really isn't that dire...
         | 
         | Discriminating good answers is easier than generating them.
         | Good evaluations write test sets for the discriminators to show
         | when this is or isn't true. Evaluating the outputs as the user
         | might see them are more representative than having your
         | generator do multiple tasks (e.g. solve a math query and format
         | the output as a multiple choice answer).
         | 
         | Also, human labels are good but have problems of their own, it
         | isn't like by using a "different intelligence architecture" we
         | elide all the possible errors. Good instructions to the
         | evaluation model often translate directly to better human
         | results, showing a correlation between these two sources of
         | sampling intelligence.
        
           | suddenlybananas wrote:
           | What's 45+8? Is it 63?
        
             | alextheparrot wrote:
             | If this sort of error isn't acceptable, it should be part
             | of an evaluation set for your discriminator
             | 
             | Fundamentally I'm not disagreeing with the article, but
             | also think most people who care take the above approach
             | because if you do care you read samples, find the issues,
             | and patch them to hill climb better
        
           | e1g wrote:
           | Agree, current "thinking" models are effectively "re-run this
           | question N times, and determine the best answer", and this
           | LLM-evaluating-LLM loop demonstrably leads to higher quality
           | answers against objective metrics (in math, etc).
        
             | brookst wrote:
             | That's... not how thinking models work. They tend to be
             | iterative and serial, not parallel and then pick-one.
        
           | majormajor wrote:
           | > Discriminating good answers is easier than generating them.
           | 
           | I don't think this is true for many fields - especially
           | outside of math/programming. Let's say the task is "find the
           | ten most promising energy startups in Europe." (This is
           | essentially the sort of work I see people frequently talk
           | about using research modes of models for here or on
           | LinkedIn.)
           | 
           | In ye olden days pre-LLM you'd be able to easily filter out a
           | bunch of bad answers from lazy humans since they'd be short,
           | contain no detail, have a bunch of typos, formatting
           | inconsistencies from copy-paste, etc. You can't do that for
           | LLM output.
           | 
           | So unless you're a domain expert on European energy startups
           | you can't check for a good answer without doing a LOT of
           | homework. And if you're using a model that usually only looks
           | at, say, the top two pages of Google results to try to figure
           | this out, how is the validator going to do better than the
           | original generator?
           | 
           | And what about when the top two pages of Google results start
           | turning into model-generated blogspam?
           | 
           | If your benchmark can't evaluate prospective real-world tasks
           | like this, it's of limited use.
           | 
           | A larger issue is that once your benchmark, that used this
           | task as a criteria, based on an expert's knowledge, is
           | published, anyone making an AI Agent is incredibly
           | incentivized to (intentionally or not!) to train
           | _specifically_ on _this_ answer without necessarily actually
           | getting better at the fundamental steps in the task.
           | 
           | IMO you can never use an AI agent benchmark that is published
           | on the internet more than once.
        
             | jgraettinger1 wrote:
             | > You can't do that for LLM output.
             | 
             | That's true if you're just evaluating the final answer.
             | However, wouldn't you evaluate the context -- including
             | internal tokens -- built by the LLM under test ?
             | 
             | In essence, the evaluator's job isn't to do separate fact-
             | finding, but to evaluate whether the under-test LLM made
             | good decisions given the facts at hand.
        
               | majormajor wrote:
               | I would if I was the developer, but if I'm the user being
               | sold the product, or a third-party benchmarker, I don't
               | think I'd have full access to that if most of that is
               | happening in the vendor's internal services.
        
             | brookst wrote:
             | > IMO you can never use an AI agent benchmark that is
             | published on the internet more than once.
             | 
             | This is a long-solved problem far predating AI.
             | 
             | You do it by releasing 90% of the benchmark publicly and
             | holding back 10% for yourself or closely trusted partners.
             | 
             | Then benchmark performance can be independently evaluated
             | to determine if performance on the 10% holdback matches the
             | 90% public.
        
             | alextheparrot wrote:
             | > Good evaluations write test sets for the discriminators
             | to show when this is or isn't true.
             | 
             | If they can't write an evaluation for the discriminator I
             | agree. All the input data issues you highlight also apply
             | to generators.
        
           | tempfile wrote:
           | > Discriminating good answers is easier than generating them.
           | 
           | This is actually very wrong. Consider for instance the fact
           | that people who grade your tests in school are typically more
           | talented, capable, trained than the people taking the test.
           | This is true even when an answer key exists.
           | 
           | > Also, human labels are good but have problems of their own,
           | 
           | Granted, but...
           | 
           | > it isn't like by using a "different intelligence
           | architecture" we elide all the possible errors
           | 
           | nobody is claiming this. We elide the _specific, obvious_
           | problem that using a system to test itself gives you no
           | reliable information. You need a control.
        
             | rf15 wrote:
             | Trading control for convenience has _always_ been the
             | tradeoff in the recent AI hype cycle and the reason why so
             | many people like to use ChatGPT.
        
             | alextheparrot wrote:
             | It isn't actually very wrong. Your example is tangential as
             | graders in school have multiple roles -- teaching the
             | content and grading. That's an implementation detail, not a
             | counter to the premise.
             | 
             | I don't think we should assume answering a test would be
             | easy for a Scantron machine just because it is very good at
             | grading them, either.
        
           | diggan wrote:
           | > Discriminating good answers is easier than generating them.
           | 
           | Lots of other good replies to this specific part, but also,
           | lots of developers are struggling with the feeling that
           | reviewing code is harder than writing code (something I
           | personally not sure I agree with), seen that sentiment being
           | shared here on HN a lot, and would directly go against that
           | particular idea.
        
             | alextheparrot wrote:
             | I wish the other replies and this would engage with the
             | sentence right after it indicating that you should test
             | this premise empirically.
        
         | sdenton4 wrote:
         | When I was working in audio compression, evaluation was very
         | painful because we had no programmatic way to measure how good
         | some reconstructed audio sounds to a human. Any metric you
         | could come up with was gameable, and direct optimization would
         | lead to artifacts.
         | 
         | As a result, we always had a two-step evaluation process. We
         | would use a suite of metrics to guide development progress
         | (validation), but the final evaluation reported in a paper
         | always involved subjective human listening experiments. This
         | was expensive, but the only way to show that the codecs were
         | actually improving.
         | 
         | Similarly, here it seems fine to use LLMs to judge your work in
         | progress, but we should be requiring human evaluation for
         | 'final' results.
        
           | ttoinou wrote:
           | Wouldn't that process avoid you finding a better subjective
           | audio codec that doesn't reduce typical metrics (PSNR etc.) ?
           | another process would rather be to first construct a metric
           | software that tries to be similar to the subjective
           | experience of humans, then use that to create audio codecs
           | optimizing this metric
        
             | layer8 wrote:
             | You are describing psychoacoustic models, which work to a
             | reasonable extent for lossy compression of audio (MP3 and
             | successors are based on them), but I can see how it would
             | be much more difficult/less helpful for reconstructing
             | audio.
        
             | sdenton4 wrote:
             | There's two answers to that....
             | 
             | The first is, how do you know the subjective optimization
             | your making is actually any good? You're just moving the
             | problem back one layer of abstraction.
             | 
             | The second is, we did that, eventually, by training models
             | to predict subjective listening scores from the giant pile
             | of subjective test data we had collected over the years.
             | (ViSQoL) It's great, but we still don't trust it for end-
             | of-the-day, cross codec comparison, because we don't want
             | to reward overfit on the trained model.
             | 
             | https://arxiv.org/abs/2004.09584
        
               | ttoinou wrote:
               | Nice
               | 
               | Well yeah you would still need human testing
        
           | DonHopkins wrote:
           | You gotta snag yourself one of those awesome KEMAR dummy head
           | and torso simulators, preferably the fully accessorized
           | luxury edition that comes with the heavy duty portable travel
           | case with lots of room for extra ears and microphones and
           | wigs, which is so much fun to take through airport security.
           | 
           | They were great for taking to Grateful Dead concerts to
           | record the music directly in front of the Wall of Sound, and
           | to measure the response so you can play back all your Dead
           | tapes with that same front row psychoacoustic perspective. ;)
           | 
           | https://www.grasacoustics.com/industries/kemar/applications-.
           | ..
           | 
           | https://www.grasacoustics.com/products/accessories/product/4.
           | ..
        
         | BoiledCabbage wrote:
         | > Tests that a "do nothing" AI can pass aren't intrinsically
         | invalid but they should certainly be only a very small number
         | of the tests. I'd go with low-single-digit percentage, not 38%.
         | But I would say it should be above zero; we do want to test for
         | the AI being excessively biased in the direction of "doing
         | something", which is a valid failure state.
         | 
         | There is a simple improvement here: give the agent a "do
         | nothing" button. That way it at least needs to understand the
         | task well enough to know it should press the do nothing button.
         | 
         | Now a default agent that always presses it still shouldn't
         | score 38%, but that's better than a NOP agent scoring 38%.
        
         | jstummbillig wrote:
         | > using a judge of the same architecture as the thing being
         | judged maximizes the probability of fundamental failure of the
         | benchmark to be valid due to the judge having the exact same
         | blind spots as the thing under test.
         | 
         | That's what humans do all the time. What's the fundamental
         | difference? Or are you saying that's also broken?
        
           | qsort wrote:
           | We want machines that are better than humans, otherwise what
           | purpose do they serve?
        
             | xnx wrote:
             | A machine with human level "AI" is still useful if it can
             | run 24/7 and you can spin up 1M instances.
        
               | einrealist wrote:
               | And boil the planet.
        
               | fragmede wrote:
               | and they don't have family that gets sick or dies or come
               | into work hungover or go off on political tangents and
               | cause HR issues or want to take vacations or complain
               | about bad working conditions.
        
           | rsynnott wrote:
           | ... I mean, when evaluating "45 + 8 minutes" where the
           | expected answer was "63 minutes", as in the article, a
           | competent human reviewer does not go "hmm, yes, that seems
           | plausible, it probably succeeded, give it the points".
           | 
           | I know LLM evangelists love this "humans make mistakes too"
           | line, but, really, only an _exceptionally_ incompetent human
           | evaluator would fall for that one.
        
             | brookst wrote:
             | have you ever hired human evaluators at scale? They make
             | all sorts of mistakes. Relatively low probability, so it's
             | a noise factor in, but I have yet to meet the human who is
             | 100% accurate at simple tasks done thousands of times.
        
               | Jensson wrote:
               | Which is why you hire them at scale as you say, then they
               | are very reliable. LLM at scale are not.
               | 
               | The problem with these AI models is there is no such
               | point where you can just scale them up and they can solve
               | problems as accurately as a group of humans. They add too
               | much noise and eventually go haywire when left to their
               | own devices.
        
           | jerf wrote:
           | Yes, humans evaluating humans also causes human foibles to be
           | magnified.
           | 
           | I cite the entire current education system. Substantiating
           | that claim would take more than an HN comment allows, though
           | I think most people can probably get the drift of what I'm
           | talking about, even if we'd disagree about the details.
           | Absolutely humans are not immune to this.
           | 
           | I also cite the entire concept of "fallacies", many of which
           | are things that both human brains tend to produce and then
           | tend to evaluate poorly. An alien species might find some of
           | our fallacies absolutely transparent, and have entirely
           | different fallacies of their own that none of us would find
           | convincing in the slightest, because of fundamentally
           | different brain architectures.
           | 
           | I don't think AIs are ready for this yet and I don't expect
           | LLMs ever will be, but in the future getting an outsider
           | perspective from them in a sort of Mixture of Experts
           | architecture could be valuable for life decisions. (I look to
           | the future AI architectures in which LLMs are just a
           | component but not the whole.)
        
           | jacobr1 wrote:
           | The equivalent would be having the _same human_ review their
           | own work. We require others with different experience and
           | fresh eyes for secondary review and for the most important
           | task multiple people.
           | 
           | To some extent the same llm with a new context history and
           | different prompt is sorta like that ... but still is much
           | weaker than using a different system entirely.
        
             | brookst wrote:
             | How do you feel about o3 reviewing 4o-mini?
        
         | datpuz wrote:
         | Benchmarks in software have always been bullshit. AI benchmarks
         | are just even more bullshit since they're trying to measure
         | something significantly more subjective and nuanced than most.
        
         | xnx wrote:
         | > I'm particularly annoyed by using LLMs to evaluate the output
         | of LLMs
         | 
         | This does seem a little crazy on its face, but it is yielding
         | useful and improving tools.
        
           | jerf wrote:
           | It's not about it being crazy and it's not about personal
           | opinions about AI. It's about chaos mathematics. Iterating
           | with the same system like that has certain easy-to-understand
           | failure states. It's why I phrased it specifically in terms
           | of using the same architecture to validate itself. If we had
           | two radically different AI architectures that were capable of
           | evaluating each other, firing them at _each other_ for
           | evaluation purposes would be much, _much_ less susceptible to
           | this sort of problem than firing either of them at
           | themselves. That will never be a good idea.
           | 
           | See also a cousin comment of mine observing that human brains
           | are absolutely susceptible to the same effect. We're just so
           | used to it that it is the water we swim through. (And
           | arguably human brains are more diverse than current AI
           | systems functioning at this level. No bet on how long that
           | will be true for, though.)
           | 
           | Such composite systems would still have their own
           | characteristics and certainly wouldn't be guaranteed to be
           | perfect or anything, but at least they would not tend to
           | iteratively magnify their own individual flaws.
           | 
           | Perhaps someday we will have such diverse architectures. We
           | don't today have anything that can evaluate LLMs other than
           | human brains, though.
        
         | DonHopkins wrote:
         | It's like using steel to produce steel. What else are you going
         | to use? Bamboo?
        
           | dmbche wrote:
           | I'm not sure if I'm dense, but we don't use steel to make
           | steel (whether crucibles or "feed material").
           | 
           | The first person to make steel made it without steel didn't
           | they?
           | 
           | Did I miss something?
           | 
           | Edit0: fun tidbit - Wootz steel was made with crucibles of
           | clay with rice husks mixed in (husks would carbonize quickly
           | and introduce air layers to better isolate) and many
           | seemingly random objects (fruits, vegetation) were added to
           | the crucible to control carbon content.
           | 
           | I higly recommend A Collection of Unmitigated Pedantry's
           | series on steel (it's a blog, just search "ACOUP steel".
        
             | dmbche wrote:
             | Second fun tidbit : Bamboo was used as the fuel source in
             | some furnaces - they did indeed use bamboo like the parent
             | comment mentionned.
        
           | AIPedant wrote:
           | It's more like using a faulty and dangerous automated foundry
           | to make steel when you could just hire steelworkers.
           | 
           | That's the real problem here - these companies are swimming
           | in money and have armies of humans working around the clock
           | training LLMs, there is no honest reason to nickel-and-dime
           | the actual evaluation of benchmarks. It's like OpenAI using
           | exact text search to identify benchmark contamination for the
           | GPT-4 technical report. I am quite certain they had more
           | sophisticated tools available.
        
         | szvsw wrote:
         | > I'm particularly annoyed by using LLMs to evaluate the output
         | of LLMs.
         | 
         | Even though I largely agree with parts of what you wrote, if
         | you squint your eyes enough you can _kind of_ see an argument
         | along the lines of "difficult to solve but easy to verify."
        
       | mycall wrote:
       | SnitchBench [0] is unique benchmark which shows how aggressively
       | models will snitch on you via email and CLI tools when they are
       | presented with evidence of corporate wrongdoing - measuring their
       | likelihood to "snitch" to authorities. I don't believe they were
       | trained to do this, so it seems to be an emergent ability.
       | 
       | [0] https://snitchbench.t3.gg/
        
         | ggregoryarms wrote:
         | Seems like more of a subtextual/accidental ability than an
         | emergent ability.
        
       | camdenreslink wrote:
       | The current benchmarks are good for comparing between models, but
       | not for measuring absolute ability.
        
         | qsort wrote:
         | Not even that, see LMArena. They vaguely gesture in the general
         | direction of the model being good, but between contamination
         | and issues with scoring they're little more than a vibe check.
        
         | fourside wrote:
         | But if the test metrics are fundamentally flawed they might not
         | be useful even for relative comparisons. Like if I told you
         | that Model A scores 10x as many blorks points as model B, I
         | don't know how you translate that into insights about
         | performance on real world scenarios.
        
         | rsynnott wrote:
         | I don't really buy that they're even necessarily useful for
         | comparing models. In the example from the article, if model A
         | says "48 + 6 minutes" and gets marked correct, and model B says
         | "63 minutes" (the correct answer) and gets marked correct, the
         | test will say that they're equivalent on that axis when in fact
         | one gave a completely nonsense answer.
        
       | TheOtherHobbes wrote:
       | Any sufficiently hyped technology is indistinguishable from
       | magic.
        
       | rsynnott wrote:
       | > 45 + 8 = 63
       | 
       | > Pass
       | 
       | Yeah, this generally feels like about the quality one would
       | expect from the industry.
        
       | let_tim_cook_ wrote:
       | Are any authors here? Have you looked at AppWorld?
       | https://appworld.dev
        
       | ttoinou wrote:
       | What makes LLMs amazing (fuzzy input, fuzzy output) is exactly
       | why they are hard to benchmark. If they could be benchmarked
       | easily, they wouldn't be powerful by definition. I have no idea
       | what's going on in the minds of people benchmarking LLMs for
       | fuzzy tasks, and in the minds of people relying on benchmarks to
       | make decisions about LLMs, I never looked at them. People doing
       | benchmarks have to prove what they do is useful, not us public
       | proving them they're doing it wrong.
       | 
       | Of course, for such tasks we could benchmark them :
       | 
       | * arithmetic (why would use LLM for that ?)
       | 
       | * correct JSON syntax, correct command lines etc.
       | 
       | * looking for specific information in a text
       | 
       | * looking for a missing information in a text
       | 
       | * language logic (ifs then elses where we know the answer in
       | advance)
       | 
       | But by Goodhart's Law, LLMs that have been trained to succeed in
       | those benchmarks might loose powerfulness in others tasks where
       | we really need them (fuzzy inputs, fuzzy outputs)
        
         | meroes wrote:
         | > arithmetic (why would use LLM for that ?)
         | 
         | Because people ask LLMs all of these things, including
         | arithmetic. People were saying the same about the number of r's
         | in strawberry. Why ask and LLM that!?!? But the big AI
         | companies want LLMs to be better at these questions, probably
         | because people ask them to LLMs. The big AI companies want this
         | because there is no other explanation for the money poured into
         | RLHF'ing these types of problems.
        
           | ttoinou wrote:
           | for me, that could only be solved by changing architecture
           | and/or introducing more insider tooling (like calling a
           | program to make computation). It doesnt make any sense to
           | fine tune a fuzzy input fuzzy output natural language
           | processing algorithm to add and multiply all combinations of
           | six digits numbers
        
             | potatolicious wrote:
             | This feels like a philosophical fault line in the industry.
             | 
             | For people whose purpose is to produce reliably working
             | systems yeah, training a model that calls out to
             | deterministic logic to do things like math makes total
             | sense. It will pretty much always be more reliable than
             | training a text generation model to produce correct
             | arithmetic.
             | 
             | But it feels like there's another side of the industry
             | that's more concerned with... I dunno, metaphysical aspects
             | of these models? Where the idea that the model is a
             | stochastic ball that isn't conscious, isn't thinking, and
             | does poorly at various tasks is anathema. So the effort
             | continues to try and train and fine-tune these models
             | until... something.
             | 
             | It reminds me of the great Tesla-vs-everyone-else self-
             | driving debates that raged over the past several years.
             | Lots of people unhappy that the best-functioning systems
             | fused many sensor types and a mixture of heuristic and
             | machine-learned systems in a complex architecture. These
             | folks insisted that the "best" architecture was an end-to-
             | end machine-learned system based entirely on visible light
             | cameras. Because it's "most human" or some other such
             | nonsense. As far as I can tell there was never any merit to
             | this position beyond some abstract notion of architectural
             | purity.
             | 
             | Same thing here I suppose.
        
         | th0ma5 wrote:
         | Since when do people like the fuzziness of outputs? I think you
         | make an interesting point but it also seems to imply that
         | benchmarking will never truly be possible, which I think is
         | true unless we can also make them observable which also as you
         | say gives up the mystique.
        
       | beebmam wrote:
       | I don't think "Benchmarks" are the right way to analyze AI-
       | related processes, which is probably similar to the complexity
       | surrounding human intelligence measurements and how well each
       | human can handle real-world problems.
        
       | neehao wrote:
       | And I would say, often we need effortful labels by groups of
       | humans: https://www.gojiberries.io/superhuman-level-performance/
        
       | KTibow wrote:
       | This is more or less a funnel to their Agentic Benchmark
       | Checklist: https://arxiv.org/abs/2507.02825
        
         | nerevarthelame wrote:
         | Finally, a benchmark for benchmarks. And what's great is that
         | they already benchmarked their benchmark benchmark.
         | 
         | (Apologies for the benchmark snark. I'm glad people are doing
         | this research, thanks for sharing it.)
        
       | rybosworld wrote:
       | Based on the comments, I think a lot of people are missing what
       | the AI Agent actually got wrong here. Nowhere did the agent claim
       | that 45 + 8 = 63.
       | 
       | You can see the Agent's step by step thought process here (also
       | linked in the article):
       | 
       | https://ibm-cuga.19pc1vtv090u.us-east.codeengine.appdomain.c...
       | 
       | The Agent correctly entered the starting point (MIT) and the
       | ending point (Harvard) and the mode of transport (on foot).
       | OpenStreetMap returns this as taking 45 minutes long.
       | 
       | Then the agent reversed the directions, and changed the mode of
       | transport to car. What it should have also done, is change the
       | destination to Logan Airport. This is the part that the agent
       | missed. OpenStreetMap then returns that the drive from Harvard to
       | MIT takes 8 minutes.
       | 
       | The agent then returned the answer as being 45 minutes walking
       | and 8 minutes driving. The first number is correct. The second is
       | wrong because the agent chose the wrong destination, not because
       | it did math incorrectly.
       | 
       | Seems like lots of readers are chomping at the bit to prove how
       | stupid the models are rather than focus on the real problem the
       | author is highlighting.
        
         | asadotzler wrote:
         | "champing"
        
         | suddenlybananas wrote:
         | The model's scoring was done by another model though no? That
         | was the source of the answer being mislabed as correct. So a
         | different model thought that 45+8=63.
        
       | btdmaster wrote:
       | There is a cool solution for this:
       | https://huggingface.co/spaces/Jellyfish042/UncheatableEval
       | 
       | This doesn't work for instruction-tuned models, but it's an
       | interesting alternative approach that doesn't need a complicated
       | (and thus gameable) evaluation function or human interaction.
       | Instead, predict the next word with data newer than the training
       | set.
        
       ___________________________________________________________________
       (page generated 2025-07-11 23:01 UTC)