Everything is the right question away

  • Comments posted to this topic are about the item Everything is the right question away

  • A few days ago I read an interesting view that said:

    "AI is not predictive.  It is just statistical echoes of past data".

     

    Rick
    If you do a half-assed job of things, folks will ask 'why did this ass only do half the job?'

  • It's a predictive engine, which can feel like a human or an oracle. However, the prediction is a statistical likelihood of what answers the question.  and it's not always past data. sometimes it's combinations of past data (words) that never existed together.

  • How do you measure the effectiveness of the prediction?

    Does running the prompt in multiple LLMs help increase accuracy?

    412-977-3526 call/text

  • There isn't a good way to measure the effectiveness, mostly because the results from LLMs are not deterministic. Using multiple models is mostly to try and understand how different they are from each other and does one tend to produce more useful results.

    I guess how often you feel you have success might be the measure of "accuracy"

  • I have been using known lists for comparison to what the LLM is out putting. The biggest challenge is that they don't let you return enough records to be a meaningful test.

    I think the idea that geocoding isn't deterministic is challenging

    Coding is an entirely different story

    412-977-3526 call/text

  • I think things are likely deterministic-ish. In the same way that

    select * from lookuptable

    is often deterministic. It's not unless it's

    select * from lookuptable
    order by modifieddate

    but it can appear to be, or be close enough to be useful. Often the same, or very similar results come back from an LLM, but it's not guaranteed.

Viewing 7 posts - 1 through 7 (of 7 total)

You must be logged in to reply to this topic. Login to reply