Preferences

hansworst
Joined 371 karma

  1. Following that logic, they’ll have to keep spending quite a bit to get to the user base of the current hyperscalers, some of which are already ahead of OpenAI in terms of LLM performance.
  2. Not entirely sure if you could use it, but wondering if you’ve heard about the origin private file system feature of modern browsers? https://developer.mozilla.org/en-US/docs/Web/API/File_System...
  3. The same goes for anything that provides value right? If you make some useful software, by that logic I should be allowed to copy it and use it in whatever way I see fit (including commercially), no matter what license you used?
  4. Is it just me or is Hono being astroturfed pretty heavily on HN lately
  5. Wouldn’t they still need to pay tariffs on all the parts they manufacture in china? Maybe I’m misunderstanding the tariffs but it sounds like Chinese companies would have to build completely separate supply chains to keep the US market
  6. Obviously there’s a balance to be struck here. We could legalise fentanyl and tell people to just not use it, but that probably wouldn’t have a very positive impact on society.

    At the very least we should acknowledge the negative externalities. Just leaving it up to the market to figure out (especially if we allow the current tech monopolies to exist) will result in serious societal impact.

  7. Pretty sure the romans had public executions too
  8. Hundreds? The bread and games thing is literally a quote from the Roman Empire, and I’m sure they didn’t invent it themselves either.
  9. Social mobility index doesn’t really look at how easy it is to become very rich (I.e. get into the 1%). This is also explained in the methodology section of the article you linked.
  10. But in gaming, real time means something different: https://en.wikipedia.org/wiki/Timekeeping_in_games#Real-time

    Context matters.

  11. You’re forgetting Canada. You’re also glossing over the fact that despite losing to the nazis, the occupied countries didn’t exactly just roll over when they got attacked. And during occupation there were resistance movements in those countries too. So there’s definitely a “we” here.

    And while of course most of those people are dead now, they were all part of cultures that still hold many of the same norms and values as they do today.

  12. Arguably Google is ahead. They have many non-llm uses (waymo/deepmind etc) and they have their own hardware, so not as reliant on Nvidia.
  13. Isn't that just the LLM equivalent of hardcoding though?
  14. Plenty of indie musicians nowadays that self-publish on platforms like Spotify and YouTube. That just wasn’t possible 30 years ago.
  15. Well, it’s going to be hard to talk their way out of that one.
  16. Overfitting on test data absolutely does mean that the model would perform better in benchmarks than it would in real life use cases.
  17. The article asks the following:

    > I can’t help but wonder if there’s a reaction-diffusion-model-esque effect at work here as well

    There are continuous approximations of the game of life that show this, for example this implantation:

    https://smooth-life.netlify.app

  18. Anonymous functions don't have names. This makes it much harder to do things like profiling (just try to find that one specific arrow function in your performance profile flame graph) and tracing. Tools like Sentry that automatically log stack traces when errors occur become much less useful if every function is anonymous.
  19. > they’re not art directable or dynamic

    This is not true I believe. There are plenty of papers out there revolving around dynamic/animated splat-based models, some using generative models for that aspect too.

    There are also some tools out there that let you touch up/rig splat models. Still not near what you can do with meshes but I think fundamentally it’s not impossible.

  20. Can’t you just instruct your llm of choice to transform your prompts like this for you? Basically feed it with a bunch of heuristics that will help it better understand the thing you tell it.

    Maybe the various chat interfaces already do this behind the scenes?

  21. > because they and their family get enough Bürgergeld, that actually working would lower their income

    This means it’s not UBI, and that’s kind of the whole point here. With UBI this welfare cliff wouldn’t exist; if you work, you still raise your income. That means, unlike the current German system, UBI still incentivises people to work to increase their income/wealth.

    Of course actively disincentivising people to work will cause them to not work. That’s just rational behaviour, you cannot blame anyone for that.

  22. This assumes people just stop doing anything of value if there no longer is a proverbial stick in the form of financial ruin if they stop working.

    Nobody is saying that the carrot (personal financial gain) needs to be removed from the equation. Just that everyone is guaranteed some basic level of financial support.

    Society already produces enough wealth to cover the expense of UBI. Remember it would replace any other welfare systems in place today.

    Personally I think I might take a bit more risk, and choose to do something that I personally believe is of actual value to society rather than please some corporation or VC.

  23. I think it’s true that people in Europe feel that welfare is part of the problem here. In the Netherlands for example, one of the main right wing talking point is that refugees are given free social housing which could have gone to locals that are often on waiting lists for years.

    In America on the other hand, land was forcibly taken from the natives by colonists centuries ago. Now, if you’re looking to move to the US, you can expect to work in poverty for a few generations as a second class citizen because that’s just how the “completely fair” capitalist system is set up. Forgetting for a moment that most capital is held by a single ethnicity, and they’re definitely not going to give it away for free.

  24. The way the brain does it is by giving users a largely untrained model that they themselves have to train over the next 20 years for it to be of any use.
  25. If that’s true, you could make a killing by creating soy bean based food products and actually marketing them properly.
  26. So we should just give up on trying to accurately price externalities until we’ve found a way to include all of them? (Hint: we never will, so that would mean never taking action)
  27. On the flip side: in some cases there are downsides to regular checkups, in particular false positive diagnoses leading to medical operations causing more harm on a population level than if the checkups hadn’t been done in the first place. Unfortunately our medical practices aren’t always good enough yet where doing early checkups at a large scale actually prevent harm.

    Combine this with the sometimes shoddy diagnostics processes in mental health, and suddenly it doesn’t seem too weird to me that we avoid poking too much until people complain themselves. I think that if we were to ask people to fill out every mental health questionnaire that exists, most people would likely test positive for some disorders. But if those symptoms are not causing unmanageable harm in people’s lives, and given the current state of treatments (not nearly always effective), I think we should think twice before subjecting everyone to a list of mental health questionnaires.

  28. > big loops or loops with unknown bounds at compile time

    That’s not allowed in webgl, so I doubt TSL can make it work somehow.

  29. I think it’s questionable whether you can actually use this bit count to represent the amount of information from the book. Those 1200 bits represent the way in which this particular book is different from everything else the model has ingested. Similarly, if you read an entire book yourself, your brain will just store the salient bits, not the entire text, unless you have a photographic memory.

    If we take math or computer science for example: some very important algorithms can be compressed to a few bits of information if you (or a model) have a thorough understanding of the surrounding theory to go with it. Would it not amount to IP infringement if a model regurgitates the relevant information from a patent application, even if it is represented by under a kilobyte of information?

This user hasn’t submitted anything.

Keyboard Shortcuts

Story Lists

j
Next story
k
Previous story
Shift+j
Last story
Shift+k
First story
o Enter
Go to story URL
c
Go to comments
u
Go to author

Navigation

Shift+t
Go to top stories
Shift+n
Go to new stories
Shift+b
Go to best stories
Shift+a
Go to Ask HN
Shift+s
Go to Show HN

Miscellaneous

?
Show this modal