llm_trw

31 ms·

511.
by llm_trw 2y ago
It isn't. I want a place where people leave their politics, religion and genitals at the door. Op made it seem like bluesky did that. A look at the front page says otherwise.
512.
by llm_trw 2y ago
The first post when I opened it after reading yours: >The same people who’ve spent the last several years decrying “unqualified DEI hires” are now shoehorning through Cabinet nominations who can’t even pass a basic background test. This
513.
by llm_trw 2y ago
Memory is king. Anything that has more memory and adequate compute will win the coming AI wars. At the rate at which power consumption is growing now that the shortage of current gen cards has started to work itself out people are realizing
514.
by llm_trw 2y ago
This is exactly the type of novel work that llms are good at. It's tedious and has annoying internal logic, but that logic is quite flat and there are a million examples to generalise from. What they fail at is code with high cyclomati
515.
by llm_trw 2y ago
The onion has not yet bought the times.
516.
by llm_trw 2y ago
You should write them in a superior text editor like Emacs so they are clearer.
517.
by llm_trw 2y ago
If you're on emacs a directory tree with orgmode files works amazingly well when posted on gitlab or similar.
518.
by llm_trw 2y ago
> But uh... isn't it possible that one exists and the other does not? Going through the airport between 2001 and 2021 showed that having a Muslim sounding name was going to be a trigger for a random inspection. The Rotherham child r
519.
by llm_trw 2y ago
It's exactly the same meaning but with the sign reversed. Turns out identity politics is a terrible idea. Now the Guardian et al are finding out exactly why when the people they disagree with are doing it too.
520.
by llm_trw 2y ago
An editor is the least interesting thing to learn in computer engineering. If you enjoy wasting your time may I suggest something like stamp collecting that doesn't create billions of dollars of lost productivity for the rest of us?
521.
by llm_trw 2y ago
The typical bell curve meme: https://imgflip.com/i/9a9gm7 The fun begins when you can't debug by printf because the buffer is shared by multiple processes that change it while the printf command is running.
522.
by llm_trw 2y ago
Antique means stable. When you've been using the same thing for 20 years changing defaults to be trendy isn't a plus. You'll understand in 20 years when you're on your 4th IDE which will change the world^tm.
523.
by llm_trw 2y ago
>But becoming proficient in VS code takes maybe 25 minutes. Becoming proficient in kicking a ball takes 5 minutes. Becoming proficient in football takes slightly longer. Magit alone makes VS Code look like a child's finger painting.
524.
by llm_trw 2y ago
Since I graduated, I've had people try to sell me on the following editors, which were going to be the future: BBEdit, TextMate, Sublime Text, Atom, Komodo Edit, NetBeans, and Eclipse. Each was going to kill Emacs. Somehow, I think I&#
525.
by llm_trw 2y ago
Don't forget that there are a lot fewer kids today. One only need to look at how a single kid is parented vs four. I had a hilarious example at the park where the single child had constant supervision while the fourth kid of the same a
526.
by llm_trw 2y ago
To protect the guilty, and any future books I might write on the subject, I'll leave it to your imagination.
527.
by llm_trw 2y ago
>I’ve never thought of it being used maliciously, it’s for visibility. It would be a shitty manager that would use it that way and if they’re already shitty then this tool won’t change that. You are a member of the intelligence community
528.
by llm_trw 2y ago
I wasn't aware we were at war.
529.
by llm_trw 2y ago
I'm building the layer on top of the models. People call it agentic AI but that's a word without a definition. Needless to say better LLMs help with my work, the same way that a stronger horse makes plowing easier. The difference
530.
by llm_trw 2y ago
I'm sorry but you just don't understand what the paper is saying.
531.
by llm_trw 2y ago
Not sure why you're being downvoted that is exactly why I'm using that simple problem to benchmark LLMs. If an LLM can't figure out how to traverse a graph in its working memory then it has no hope of figuring out how to stru
532.
by llm_trw 2y ago
Lunch and breakfast are serious business.
533.
by llm_trw 2y ago
Alternatively the idea that people don't entertain at home is a sign of a completely broken housing market where you get a human storage unit rather than a home.
534.
by llm_trw 2y ago
Fascinating work. I never thought about what we could do if instead of trying to do work in real time we slowed robots down for non time critical tasks to latencies which current transformers can deal with on real hardware.
535.
by llm_trw 2y ago
Keep in mind that workers in the EU work for around 6 hours a day, a robot works for 24, so it's replacing 4 of them, not 1.
536.
by llm_trw 2y ago
Any benchmark which isn't dynamically generated is useless for that very reason.
537.
by llm_trw 2y ago
It is very much an open question just what an llm can solve when allowed to generate an indefinite number of intermediate tokens and allowed to sample an arbitrary amount of text to ground itself. There are currently no tools that let llms
538.
by llm_trw 2y ago
Section 2.2 from your paper, with less clarity and more obfuscation: >While the weights and the activations are quantized to low precision, the gradients and the optimizer states are stored in high precision to ensure training stability
539.
by llm_trw 2y ago
Finding a path between two vertices when given an itinerary of all the edges in a general graph, exactly what I said in the OP.
540.
by llm_trw 2y ago
I've not tested those models. Feel free to flick me through a couple of k in bitcoins if you'd like me to have a look for you.

More ›