jackblemming

7 ms·

91.
by jackblemming 3y ago
Because there’s only so many people in the world and of those people they only have so much money.
92.
by jackblemming 3y ago
HN is in a scary place if any critical post of Apple is being astroturfed by obvious bad actors and nothing is being done about it.
93.
by jackblemming 3y ago
I didn’t get into software to chase money. If I did, that’s a pretty dumb decision as there’s way more profitable careers. And working at Facebook seems pretty hellish.
94.
by jackblemming 3y ago
I don’t know why you would want to work for a company that treats you like this. I guess some people will put up with a lot for money.
95.
by jackblemming 3y ago
It doesn’t matter how many hundreds of rounds of interviews and checks you have, you will always run into Bayes Error. You will never be able to filter out assholes who can hold it together for an interview. Or lazy people who get hired and
96.
by jackblemming 3y ago
The state of art neural net architecture, whether that be transformers or the like, trained on self play to optimize non-differentiable but highly efficient architectures is the way.
97.
by jackblemming 3y ago
Impressive work. Nice job team. This is big.
98.
by jackblemming 3y ago
Telling people who depended on probably internal behavior that wasn’t an explicit part of your contract to fuck off when they whine about you changing something.
99.
by jackblemming 3y ago
I wouldn’t worry too much on a lot of these points. I won’t say the math behind AI is simple, but it’s mostly undergrad level. You can get up to speed on it if you really want to. The hard part is writing fast implementations, but many othe
100.
by jackblemming 3y ago
You bend the knee. The legal system exists to serve the rich.
101.
by jackblemming 3y ago
>As I told one team I led, “I’m your bitch. Tell me where your rocks are and I’ll move them out of the way so you can get your work done.” Great mindset and thoughts, but I hope this isn't what you actually said. I'd also like
102.
by jackblemming 3y ago
No, arbitrarily wide neural networks are approximators of Borel Measurable functions. Big difference between that and “any function”. RNNs are Turing Complete though.
103.
by jackblemming 3y ago
> the topics in the parent post should not be a major surprise to anyone who has read https://people.math.harvard.edu/~ctm/home/text/others/shanno ... ! > which clearly explains it (and said emergen
104.
by jackblemming 3y ago
They have no moat at all and the tech is advancing faster than enterprise can keep up, unless you’re OpenAI. Seems like a wild evaluation to me.
105.
by jackblemming 3y ago
Something tells me Karpathy rarely uses the “FSD” in his Tesla. He barely mentioned Tesla FSD in the blog despite being a key leader in the project. Perhaps he’d like to forget about it all together.. Maybe Elon really screwed the project b
106.
by jackblemming 3y ago
Seems like a lot of complaining and the only solution offered is regulation or “just do it better”? Ok.
107.
by jackblemming 3y ago
If the rich have their way we’re all going to be living like this within a few decades. Probably even worse.
108.
by jackblemming 3y ago
I chose a simple math equation that should almost assuredly be in a dataset of 1 trillion tokens exactly to check its basic pattern matching skills. The token "10" and "20" are even in its hardcoded vocabulary set. I thi
109.
by jackblemming 3y ago
You've actually shown your poor understanding of LLMs. I just asked Llama-2 7b the same question and it answered perfectly fine. It did not need to use an external python interpreter or a function call, or need to be prompted with chai
110.
by jackblemming 3y ago
What is 10 + 10? Answer: 0 + 10 = 10 + 10 = 10 + 10 = 10 + 10 = 10 + 10 = This seems like a waste of compute and time.
111.
by jackblemming 3y ago
Apple and Android mobile development both have some of the worst APIs and designs I’ve ever seen. I really can’t believe these companies have the most talented developers based on their objective output. I think at best, they have a small h
112.
by jackblemming 3y ago
There are a ton of people like this, trying to flag plant obvious “duh” ideas so they can jump in and say “see I knew it would require some kind of long term memory!”… Duh? The implementation and results are what matter. Nobody cares that y
113.
by jackblemming 3y ago
I haven't found this to be the case at all, and have found great success in taking the good parts of OOP and functional programming.
114.
by jackblemming 3y ago
>`send_emails_on` suggests strongly some scheduling capability Yes. >`current_time` says exactly what it is needed for. Not always.
115.
by jackblemming 3y ago
Don’t do something silly like take a parameter called current_time. Instead, name it what it’s actually used for or doing. This could be send_emails_on or the like. Parameters should tell the reader what not why.
116.
by jackblemming 3y ago
Agreed, the issue isn't AI, it’s poor taste. Nobody wants to see what looks like an uncanny valley picture of all large ethnic groups smiling weirdly in a circle at each other. This bizarre soulless corpo art trend started well before
117.
by jackblemming 3y ago
Stay far away from HuggingFace if you can. Battle-hardened if you only do the absolute simplest and boring stuff. Look at the number of open issues and skim through a few source code files and you’ll understand. Like LangChain, they were at
118.
by jackblemming 3y ago
This is completely right. People read about design patterns or DRY, usually apply them wrong since they're still learning... and then act like it's the source materials fault?
119.
by jackblemming 3y ago
People have tried using probabilistic SNNs because the gradient is well defined there. The issue is it’s computationally intractable to calculate so you still have to use an approximation and you’re thus not better off than the people using
120.
by jackblemming 3y ago
What stopped you from scaling this method to billions of neurons and training a LLM SNN?

More ›