tehsauce

10 ms·

1. 151.▲
   by tehsauce 4y ago

   The reason the similarly sized open source models aren’t as good is because gpt-3 has had a 2 year head start of additional development and improvement. The gpt-3 that is available today is Much better than what was initially produced in 20
2. 152.▲
   by tehsauce 4y ago

   Any videos of this in action?
3. 153.▲
   by tehsauce 4y ago

   Really nice article! I recently had a fun discovery that it’s possible to implement a reasonably efficient GPU sprite renderer using linear algebra with sparse arrays + jit in jax. The quick writeup - https://pwhiddy.github.io&#x
4. 154.▲
   by tehsauce 4y ago

   this is exactly what clip already does, and there will be massive improvements in this area in coming years, I promise.
5. 155.▲
   by tehsauce 4y ago

   Fish are not safe to eat in the entirety of Poland? Haha this is news to me.
6. 156.▲
   by tehsauce 4y ago

   All of those accidents combined killed less than 10% of the people that are killed by fossil fuel pollution every single year. Most nuclear accidents are entirely hypothetical.
7. 157.▲
   by tehsauce 4y ago

   Look at the numbers though. Fossil fuels kill many people every year, while nuclear accidents are almost entirely hypothetical. The point of the article is to show that fears like the ones expressed above just aren’t supported by the data.
8. 158.▲
   by tehsauce 4y ago

   If the US built more clean nuclear plants that would somehow help Iran develop nuclear weapons faster? Surely the slope cannot be that slippery.
9. 159.▲
   by tehsauce 4y ago

   I’m curious what caused those 0.02 deaths per terawatt for solar.
10. 160.▲
    by tehsauce 4y ago

    A dollar invested in the stock market has a much higher expected return than a dollar invested in the lottery.
11. 161.▲
    by tehsauce 4y ago

    Successful people probably are most happy when they are chasing after ambitious goals. It makes sense from an evolutionary standpoint.
12. 162.▲
    by tehsauce 4y ago

    It can do a lot more than answer questions, it can generate original content which google can not do at all.
13. 163.▲
    by tehsauce 4y ago

    Worth noting that in the limit the distribution P(human did/said A | context B) is a complete definition of human behavior. If you could model this perfectly, that would be a perfect model of a human ie identical.
14. 164.▲
    by tehsauce 4y ago

    If you can find a more efficient or effective approach than this “redundant” one you’ll be rich. There’s a lot of smart people working in this space.
15. 165.▲
    by tehsauce 4y ago

    Thanks for sharing your perspective. Best of luck
16. 166.▲
    by tehsauce 4y ago

    “Although ML code generation helps with getting the initial code written, it cannot do much beyond that - if that code is to be maintained and changed in the future (and if anyone uses the product, it is), the developer still needs to fully
17. 167.▲
    by tehsauce 4y ago

    I was quoting it as it was used in the footnote: “It's only because they live in a world created by the Enlightenment that kids from the suburbs can play at banning ideas.”
18. 168.▲
    by tehsauce 4y ago

    So true. Insane that he sees himself as the victim here.
19. 169.▲
    by tehsauce 4y ago

    Agreed. I’ve enjoyed a lot of his writing but this one is pretty embarrassing.
20. 170.▲
    by tehsauce 4y ago

    “Why? Why has this antiquated-sounding religious concept come back in a secular form? And why now?” Victims of racism and sexism are tired of living in a world built for someone else. My heart goes out to Paul Graham the victim who would no
21. 171.▲
    by tehsauce 4y ago

    The number of people working in academic research labs for major tech companies is probably on the order of 1% of all PhDs.
22. 172.▲
    by tehsauce 4y ago

    For a person that isn’t from a wealthy background already, it’s hard to justify the path of academia when a more lucrative path has much more potential to be life changing for them and their family.
23. 173.▲
    by tehsauce 4y ago

    [4] Goodfellow, Bengio & Courville 2016, p. 200, "The term back-propagation is often misunderstood as meaning the whole learning algorithm for multilayer neural networks. Backpropagation refers only to the method for computing the
24. 174.▲
    by tehsauce 4y ago

    The search strategy/optimizer is actually gradient descent, backpropagation is just an efficient way to compute gradients.
25. 175.▲
    by tehsauce 4y ago

    Curious, any specific examples of missing knowledge or misconceptions?
26. 176.▲
    by tehsauce 4y ago

    I think the reason low resource languages are prioritized is to compensate for the fact that AI research normally has a tendency to marginalize these languages.
27. 177.▲
    by tehsauce 4y ago

    I agree. Maybe there’s a reason for all the extra complexity in vulkan, perhaps because such a wide variety of devices are targeted?
28. 178.▲
    by tehsauce 4y ago

    Unless they are quietly sitting on some really big work (which isn't out of the question), openai seems to be getting substantially leapfrogged by google. In terms of how "well known" googles language models are, most of us a
29. 179.▲
    by tehsauce 4y ago

    Sadly openai has fallen behind google at their own game. Google was able to invest heavily and optimize it's hardware and infrastructure down to the silicon, which has turned out to be the critical competitive advantage in AI. Openai i
30. 180.▲
    by tehsauce 4y ago

    mercury is quite close to the sun which makes it hard to see sometimes as it’s only ever visible for a short time very close to the horizon at twilight