[HN Gopher] Thoughts on Evals
       ___________________________________________________________________
        
       Thoughts on Evals
        
       Author : Nischalj10
       Score  : 26 points
       Date   : 2025-11-14 07:30 UTC (4 days ago)
        
 (HTM) web link (www.raindrop.ai)
 (TXT) w3m dump (www.raindrop.ai)
        
       | CharlieDigital wrote:
       | Both of these are kind of silly and vendors trying to sell you
       | tooling you probably don't need.
       | 
       | In a gold rush, each is trying to sell you a different kind of
       | shovel claiming theirs to be the best when you really should go
       | find a geologist and and figure out where the vein is.
        
       | anonymoushn wrote:
       | The framing in this post is really weird. Automated evals can be
       | much more informative than unit tests because the results can be
       | much more fine grained. A/B testing in production is not suitable
       | for determining whether all of one's internal experiments are
       | successful or not.
       | 
       | I don't doubt that Raindrop's product is worthwhile to model
       | vendors, but the post seems like its audience is C suite folks
       | who have no clue how anything works. Do their most important
       | customers even have any of these?
        
         | CharlieDigital wrote:
         | I think in most cases, outside of pure AI providers or think AI
         | wrappers, almost every team will realize more gains from
         | focusing on their user domains and solving business problems
         | versus fine tuning their prompts to eek out a 5% improvement
         | here and there.
        
           | basket_horse wrote:
           | I don't think you can use this as a blanket statement. For
           | many use cases the last 5-10% is the difference between
           | demoware and production.
        
             | CharlieDigital wrote:
             | If that were true, just switching to TOON would make your
             | startup take off.
             | 
             | That is obviously not true because a 5% gain in LLM
             | performance isn't going to make up for a bad product.
        
       | gk1 wrote:
       | Founder of a/b testing company accuses founder of evals company
       | of misrepresenting how a/b tests are used in practice, then
       | concludes by misrepresenting how evals are used in practice: "Or
       | you can write 10,000,000 evals."
       | 
       | Could've _easily_ been framed as  "you need both evals and a/b
       | testing," but instead they chose this route which comes across as
       | defensive, disingenuous, and desperate.
       | 
       | BTW, if a competitor ever writes a whole post to refute something
       | you barely alluded to without even mentioning their name...
       | congratulations, you've won.
        
         | basket_horse wrote:
         | Agree. This whole post comes across as sales rather than the
         | truth that both are useful for different things
        
       | eitland wrote:
       | This scared me until I realized it is about raindrop.ai, not
       | raindrop.io.
       | 
       | (Raindrop.io is a bookmark service that AFAIK has "take money
       | from people and stores their bookmarks" as its complete business
       | model.)
        
       | koakuma-chan wrote:
       | > Intentionally or not, the word "eval" has become increasingly
       | vague. I've seen at least 6 distinct definitions of evals
       | 
       | This. I am so tired of people saying evals without defining what
       | they mean. And now even management is asking me for evals and why
       | we are not fine tuning.
        
       | esafak wrote:
       | No, thanks. Just use evals with error bars. If you can't get
       | error bars, use an A/B test to detect spuriousness _and_ evals.
        
       | gregsadetsky wrote:
       | I'm new/uninformed in this world, but I have an idea for an eval
       | that I think has not been tried yet.
       | 
       | Can anyone direct me towards how to ... make one? At the most
       | fundamental level, is it about having test questions with known,
       | golden (verified, valid) answers, and asking different LLM models
       | to find the answer, and comparing scores (how many were found to
       | be correct)?
       | 
       | What are "obvious" things that are important to get right -
       | temperature set to 0? At least ~10 or 20 attempts at the same
       | problem for each llm? What are non-obvious gotchas?
       | 
       | Finally, any known/commonly used frameworks to do this, or any
       | tooling that can call different LLMs would be enough?
       | 
       | Thanks!
        
         | koakuma-chan wrote:
         | > Can anyone direct me towards how to ... make one?
         | 
         | https://hamel.dev/blog/posts/evals/
         | 
         | > What are "obvious" things that are important to get right -
         | temperature set to 0? At least ~10 or 20 attempts at the same
         | problem for each llm?
         | 
         | LLMs are actually pretty deterministic, so there is no need to
         | do more than one attempt with _the exact same_ data.
         | 
         | > Finally, any known/commonly used frameworks to do this, or
         | any tooling that can call different LLMs would be enough?
         | 
         | https://github.com/vercel/ai
         | 
         | https://github.com/mattpocock/evalite
        
           | gregsadetsky wrote:
           | I'm very grateful! Thanks a lot
        
           | ncgl wrote:
           | "LLMs are actually pretty deterministic, so there is no need
           | to do more than one attempt with the exact same data."
           | 
           | Is this true? I remember there being a randomization factor
           | in weighing tokens to make the output more _something_ , dont
           | recall what
           | 
           | Obviously I'm not an Ai dev
        
         | moltar wrote:
         | Take a look at promptfoo
        
       ___________________________________________________________________
       (page generated 2025-11-18 23:02 UTC)