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

help-circle

  • I doubt anyone you are talking to is opposed to all human rights, that sounds very much like a straw man statement. Reasonable people can disagree about whether any particular right should be protected by law.

    The reason is simple: any legally-protected right you have stands in direct opposition to some other right that I could have:

    • Your right to free speech is necessarily limited by my right to, among other things, freedom from slander/libel, right to a fair trial, right to free and fair elections, right to not be defrauded, etc.
    • Your right to bodily autonomy can conflict with my right to health and safety when there is a global pandemic spreading and you refuse vaccination.
    • Your property rights are curtailed by rules against environmental harm, discrimination, insider trading, etc.

    No right is ever meant to be or can be absolute, and not all good government policy is based on rights. Turning a policy argument into one about human rights is not generally going to win the other person over, it’s akin to calling someone a racist because of their position on affirmative action. There’s no rational discussion that can be had after that point.


  • I believe the answer is, unfortunately, no.

    Long answer: In the past, an ML researcher trying to do this would have used either manual labels (for example a dictionary of parts of speech for each word) or multiple sub-models trained to solve each sub-problem before combining into a full prediction model, and even then performance is not great.

    However, once the models grew to billions of parameters it turned out that none of this external linguistic knowledge is necessary and the model can learn it all on its own. But it takes billions to trillions of examples to learn all these weights, which means a double hit to the training time: each step is slower due to more parameters, and more steps are needed to train on the full dataset.

    None of these models are trainable without a cluster of GPUs, which massively parallelizes the training process.

    That doesn’t mean you can’t try, but my results training a small toy model from scratch for 20-30 hours on a consumer GPU have been underwhelming. You get some nearly-grammatical sentences but also a lot of garbage, repetition, and incoherence.


  • I don’t know if it helps, but this is not really a lie, and you shouldn’t feel bad about saying it. You have your own reason for not being able to do something you committed to. Someone else might have a different reason that is equally personal that they don’t want to share. “I forgot and I’m sorry” is a socially acceptable way to take responsibility without sharing specifics and potentially making someone else feel confusion or pity.

    You can still work on the “why wasn’t I able to do the thing I felt I needed to do” without worrying about “why wasn’t I honest about my reason”.

    Just my two cents though.


  • Incomes don’t follow a bell curve, so the choice of mean income is a red flag to me. Imagine you had 9 citizens making 100k and one billionaire, the mean is now 100,090k.

    Relatedly, being in the bottom 10% doesn’t necessarily mean the same thing in these different countries, in some of them that might not be below the poverty line so it’s comparing apples and oranges.