• 0 Posts
  • 9 Comments
Joined 1 year ago
cake
Cake day: June 19th, 2023

help-circle

  • I don’t think that anyone would argue that the general public can even solve a mathematical matrix, much less that they can only comprehend a stool based on going down a row in a matrix to get the mathematical similarity between a stool, a chair, a bench, a floor, and a cat.

    LLMs rely on billions of precise calculations and yet they perform poorly when tasked with calculating numbers. Just because we don’t calculate anything consciously to get a meaning of a word doesn’t mean that no calculations are actually done as part of our thinking process.

    What’s your definition of “the actual meaning of the concept represented by a word”? How would you differentiate a system that truly understands the meaning of a word vs a system that merely mimics this understanding?




  • I don’t think your assumption holds. Corporations are not, as a rule, incompetent - in fact, they tend to be really competent at squeezing profit out of anything. They are misaligned, which is much more dangerous.

    I think the more likely scenario is also more grim:

    AI actually does continue to advance and gets better and better displacing more and more jobs. It doesn’t happen instantly so barely anything gets done. Some half-assed regulations are attempted but predictably end up either not doing anything, postponing the inevitable by a small amount of time, or causing more damage than doing nothing would. Corporations grow in power, build their own autonomous armies, and exert pressure on governments to leave them unregulated. Eventually all resources are managed by and for few rich assholes, while the rest of the world tries to survive without angering them.
    If we’re unlucky, some of those corporations end up being managed by a maximizer AGI with no human supervision and then the Earth pretty much becomes an abstract game with a scoreboard, where money (or whatever is the equivalent) is the score.

    Limitations of human body act as an important balancing factor in keeping democracies from collapsing. No human can rule a nation alone - they need armies and workers. Intellectual work is especially important (unless you have some other source of income to outsource it), but it requires good living conditions to develop and sustain. Once intellectual work is automated, infrastructure like schools, roads, hospitals, housing cease to be important for the rulers - they can give those to the army as a reward and make the rest of the population do manual work. Then if manual work and policing through force become automated, there is no need even for those slivers of decency.
    Once a single human can rule a nation, there is enough rich psychopaths for one of them to attempt it.

    There are also other AI-related pitfalls that humanity may fall into in the meantime - automated terrorism (e.g. swarms of autonomous small drones with explosive charges using face recognition to target entire ideologies by tracking social media), misaligned AGI going rogue (e.g. the famous paperclip maximizer, although probably not exactly this scenario), collapse of the internet due to propaganda bots using next-gen generative AI… I’m sure there’s more.



  • GPT3 is pretty bad at it compared to alternatives (although it’s hard to compete with calculators on that field), but if it was just repeating after the training dataset it would be way worse. From the study I’ve linked in my other comment (https://arxiv.org/pdf/2005.14165.pdf):

    On addition and subtraction, GPT-3 displays strong proficiency when the number of digits is small, achieving 100% accuracy on 2 digit addition, 98.9% at 2 digit subtraction, 80.2% at 3 digit addition, and 94.2% at 3-digit subtraction. Performance decreases as the number of digits increases, but GPT-3 still achieves 25-26% accuracy on four digit operations and 9-10% accuracy on five digit operations, suggesting at least some capacity to generalize to larger numbers of digits.

    To spot-check whether the model is simply memorizing specific arithmetic problems, we took the 3-digit arithmetic problems in our test set and searched for them in our training data in both the forms " + =" and " plus ". Out of 2,000 addition problems we found only 17 matches (0.8%) and out of 2,000 subtraction problems we found only 2 matches (0.1%), suggesting that only a trivial fraction of the correct answers could have been memorized. In addition, inspection of incorrect answers reveals that the model often makes mistakes such as not carrying a “1”, suggesting it is actually attempting to perform the relevant computation rather than memorizing a table.



  • That’s not entirely true.

    LLMs are trained to predict next word given context, yes. But in order to do that, they develop internal model that minimizes error across wide range of contexts - and emergent feature of this process is that the model DOES perform more than pure compression of the training data.

    For example, GPT-3 is able to calculate addition and subtraction problems that didn’t appear in the training dataset. This would suggest that the model learned how to perform addition and subtraction, likely because it was easier or more efficient than storing all of the examples from the training data separately.

    This is a simple to measure example, but it’s enough to suggests that LLMs are able to extrapolate from the training data and perform more than just stitch relevant parts of the dataset together.