Comments by "" (@rumfordc) on "Is AI really getting dumber? Llama2 vs GPT-4" video.

  1. that depends on how much compute power they decide to lend you. the fewer users, the less income, the less they'll be willing to spend on computation.
    10
  2. if i have to do manual work to make it smart, that means its dumb.
    2
  3.  @cashmoneybanks8442  in all cases. you could argue whatever you want. what's important is that the same test was fed to 2 different versions and the newer one performed worse.
    2
  4.  @cashmoneybanks8442  that doesn't address the 94% drop in the math section
    2
  5.  @getignored  because good things make us look bad
    2
  6. nah, just read the study. the tests were varied and fair. it's just objectively worse at reasoning than it used to be.
    2
  7. CS is by far the better option. AI hype trend will be dead by the time you graduate, and if you are still interested in it then CS will equip you well enough to learn AI on your own.
    2
  8.  @2Damnsmartt  AI/machine learning will stick around, its not temporary, but the recent large influx of machine learning jobs will not last long. Right now thousands of companies are trying to figure out how they can get on the hype train without realizing that their product doesn't actually have any use for machine learning. After a few years of wasting their time, they're going to correct their mistake and the ML jobs they were offering will revert back to CS jobs. Only a small portion of the ML jobs will remain and the competition for them will be very steep. It's very similar to the crypto space ~5 years ago. Of course whatever you're interested in is what you should pursue, but I think CS will give you far more options. You can learn ML quickly when you already have a CS background, but the reverse is not true.
    2
  9.  @cashmoneybanks8442  calculators can't answer that question. its not the question that was dumb, its the LLM. the previous version answered the question, the new one didn't.
    1
  10.  @cashmoneybanks8442  irrelevant. the march model scored 97% accuracy, the june model scored 2.4% accuracy. there's no excuse you can make for that.
    1
  11.  @cashmoneybanks8442  i read your comment but it was just irrelevant subjective excuses for an objective drop in performance.
    1
  12. ​ @cashmoneybanks8442  you say their "skewed" and "bad" because you're projecting your own subjective desires onto the results. nothing was skewed. the same test was ran twice and the newer version failed to perform as well as the older version. that's it. did some other categories improve? of course, but that doesn't change the categories that tanked. at the end of the day, the test was ran with full consistency and the results entirely honest and accurate. everything you've said is tangential to the actual tests performed and results shown.
    1
  13.  @cashmoneybanks8442  Nothing you said discredited the 97% accuracy of the march results. Your stories about parrots don't change that result. Your "its not really doing math" semantic brainfart does not change that result. Your tangent about composite numbers does not change that result.
    1
  14. so the AI isn't as smart as a 2 year old?
    1
  15.  @edwardwhitehead8800  you realize its constantly changing, right? that's literally the point of this video
    1
  16. "it can only get better" they said
    1