3 ms·

LLMs are statistical learning. GOFAI is symbolic, rules-based stuff, expert systems and that.

by bitwize 3mo ago

kennywinker 3mo ago
There is still a categorical difference between how they are being used. Specifically analytic vs generative. Generative AI (LLMs and image generators) are the ones people have issues with - pretty much nobody cares about ML processing for analysis.
- woodson 3mo ago
  There’s a bit of a grey area, for example speech recognition. Would you classify that as analytic or generative? Whisper and speech LLMs work pretty well, but can completely make up stuff that wasn’t in the audio at all (see e.g. “thank you for watching” transcribed during silence). Other approaches are closer to the acoustic evidence but may make other mistakes (especially wrongly transcribing long tail, low frequency terms). Pick your poison.
  - kennywinker 3mo ago
    Analytic, but badly made.
- tyfon 3mo ago
  > pretty much nobody cares about ML processing for analysis. I work in a bank and a can tell you that the customers absolutely hate ML when it rejects their loan application. Over the pond in the US, I have an impression that the fico score is not exactly popular either, but I have no first hand experience.
  - disgruntledphd2 3mo ago
    As long as you can get a reason from the model, it's not that bad. Black box automatic decision making is much more problematic.
  - nradov 3mo ago
    In the US, FICO scores are mainly unpopular with "credit criminals" who have low scores. The score performs quite well in predicting how likely a borrower is to repay a loan. The problems that arise with credit scores in general are when they're used for purposes they were never designed to serve, like screening job applicants. For home mortgage lending, FICO scores are now being partially replaced by VantageScore. https://vantagescore.com/ https://vantagescore.com/
  - kennywinker 3mo ago
    People have absolutely hated their loan applications being rejected since before ML was being used anywhere near it. As evidence, let me cite the “computer says no.” sketch from 2004
    - bitwize 3mo ago
      In 2004, the logic driving such decisions was literally GOFAI: rules engines that fire off responses when certain criteria are met.
AussieWog93 3mo ago
Who downvoted this person for correctly defining GOFAI on an tech forum?

LLMs are statistical learning. GOFAI is symbolic, rules-based stuff, expert systems and that.

by bitwize 3mo ago

LLMs are statistical learning. GOFAI is symbolic, rules-based stuff, expert systems and that.

There is still a categorical difference between how they are being used. Specifically analytic vs generative. Generative AI (LLMs and image generators) are the ones people have issues with - pretty much nobody cares about ML processing for analysis.

woodson 3mo ago
There’s a bit of a grey area, for example speech recognition. Would you classify that as analytic or generative? Whisper and speech LLMs work pretty well, but can completely make up stuff that wasn’t in the audio at all (see e.g. “thank you for watching” transcribed during silence). Other approaches are closer to the acoustic evidence but may make other mistakes (especially wrongly transcribing long tail, low frequency terms). Pick your poison.
- kennywinker 3mo ago
  Analytic, but badly made.
tyfon 3mo ago
> pretty much nobody cares about ML processing for analysis. I work in a bank and a can tell you that the customers absolutely hate ML when it rejects their loan application. Over the pond in the US, I have an impression that the fico score is not exactly popular either, but I have no first hand experience.
- disgruntledphd2 3mo ago
  As long as you can get a reason from the model, it's not that bad. Black box automatic decision making is much more problematic.
- nradov 3mo ago
  In the US, FICO scores are mainly unpopular with "credit criminals" who have low scores. The score performs quite well in predicting how likely a borrower is to repay a loan. The problems that arise with credit scores in general are when they're used for purposes they were never designed to serve, like screening job applicants. For home mortgage lending, FICO scores are now being partially replaced by VantageScore. https://vantagescore.com/ https://vantagescore.com/
- kennywinker 3mo ago
  People have absolutely hated their loan applications being rejected since before ML was being used anywhere near it. As evidence, let me cite the “computer says no.” sketch from 2004
  - bitwize 3mo ago
    In 2004, the logic driving such decisions was literally GOFAI: rules engines that fire off responses when certain criteria are met.

AussieWog93 3mo ago

Who downvoted this person for correctly defining GOFAI on an tech forum?