empath75

76 ms·

961.
by empath75 2y ago
Is this more general than LLMs? Is it possible to do something Chain-of-Thought-like in a transformer model that _isn't_ trained on language?
962.
by empath75 2y ago
There's a series of two excellent videos on youtube about this proof: Detailed description of the problem of constructable regular polygons and a gloss of the proof. https://www.youtube.com/watch?v=EX7U0DGBmbM A full e
963.
by empath75 2y ago
You do not understand how GPT models work. It does more than 'importance weighting' and an absolutely massive amount of knowledge about the world is encoded in those weights. --- I asked ChatGPT to help you better understand how
964.
by empath75 2y ago
Seems to me that the correct strategy for a successful print magazine is to not focus on the _content_, but to instead focus on the _form_. Which is to say that there _are_ people who do still enjoy reading periodicals and printed books, e
965.
AI can't cross this line and we don't know why [video](youtube.com)
13 points by empath75 2y ago|1 comments
966.
by empath75 2y ago
You should not be asking it questions that require it to already know detailed information about apis and libraries. It is not good at that, and it will never be good at that. If you need it to write code that uses a particular library or
967.
by empath75 2y ago
The reason why that question is a famous question is that _many humans get it wrong_.
968.
by empath75 2y ago
I think you just want Rust. You can write extremely haskell-like code with Rust. Here's the first example in rust: fn safe_head<T>(list: &[T]) -> Option<&T> { match list { [first, ..] =&
969.
by empath75 2y ago
Going to preface this by saying that it's completely uncontroversial that the Kerala school discovered a lot of foundational mathematical concepts independently and in many cases earlier than Europeans, but the evidence for any transmi
970.
by empath75 2y ago
_Sometimes_, and people do find features in neural networks by tweaking stuff and seeing how the neurons activate, but in general, no. Any given weight or layer or perceptron or whatever can be reused for multiple purposes and it's ex
971.
by empath75 2y ago
I have a question, which might not even be related to this -- one of the keys to the power of neural networks is exploiting the massive parallelism enabled by GPUs, but are we leaving some compute on the table by using just scalar weights?
972.
by empath75 2y ago
Even extremely complicated decision trees are interpretable to some extent because you can just walk through the tree and answer questions like: "If this had not been true, would the result have been different?". It may not be po
973.
by empath75 2y ago
What we really need to figure out, to fix pedestrian fatalities in the US, is to fix whatever perverse incentives there are that lead to the traffic engineering practices that make so much of America dangerous to walk in. Optimizing for th
974.
by empath75 2y ago
It's never good to be in a car accident but safety standards for cars have _dramatically_ reduced traffic fatalities for _passengers_ and I'm sure standards can improve pedestrian survivability, too. There's lots you can do
975.
by empath75 2y ago
python is basically "if pseudocode was a language".
976.
by empath75 2y ago
For data engineering -- R and Julia.
977.
by empath75 2y ago
Yeah, his resolution to the problem is him saying "nuh-uh" in a very long winded way. Sure if you just completely abandon any attempt to explain or understand things, a lot of difficult problems evaporate. But you're not &q
978.
by empath75 2y ago
Your internal state is influenced by your environment, and you have the power to control your environment. The idea that you can't control your internal emotional state because it's governed by chemicals is absurd. People do thi
979.
by empath75 2y ago
I had a drug and drinking problem in my 20s and the cause of it was just me dealing with unresolved childhood trauma from an abusive parent and years of bullying that left me unable to interact with other human beings unless I was drinking.
980.
by empath75 2y ago
This is a problem that will resolve itself with time as C developers retire and new projects start in Rust.
981.
by empath75 2y ago
There's tons of Rust jobs out there, and they tend to be on the high end of salary. Java devs are a dime a dozen.
982.
by empath75 2y ago
I've been writing rust code in production for 2 years and none of that happens, and this is with a team that was relatively new to rust.
983.
by empath75 2y ago
It's not really the focus of the article, and you can find out with 5 seconds of googling.
984.
by empath75 2y ago
Sounds about right to me, which is why the hysteria about AI wiping out developer jobs was always absurd. _Every_ time there has been a technology that improved developer productivity, developer jobs and pay have _increased_. There is not
985.
by empath75 2y ago
Sometimes it helps to read past the first paragraph before commenting on something.
986.
by empath75 2y ago
Very clever of you to post videos so that people won't actually check what they said.
987.
by empath75 2y ago
https://latecomermag.com/article/a-holistic-view-of-the-cell... There are isolated aspects of cells that are analogous to computation, but it would be extremely misleading to try and think of the cell as a computer or
988.
by empath75 2y ago
I think one thing we've learned in the past 5 years or so is how inadequate our vocabulary is to describe all the different aspects of intelligence and consciousness (and really just psychology in general). Everything is so handwave-y
989.
by empath75 2y ago
There is a low level of guard labor that needs to happen in a free society, that's true, but it shouldn't be _increasing_ as a percentage, and it's not a good sign for the health of society at large if it is. Cooperation _isn
990.
by empath75 2y ago
They're taking it slow at least in part because there were concerns about threads traffic overwhelming mastodon servers.

More ›