noosphr

21 ms·

571.
by noosphr 10mo ago
Probably, but this is one of the few cases where instead of being told how amazing some AI tool is we are shown just what it can do.
572.
by noosphr 10mo ago
If this is what makes the AI bubble pop I'll laugh so hard.
573.
by noosphr 10mo ago
If I have zero wives yesterday and one today, by next week I will need a new house for all my new wives. Like I said in the original post: >Even the people who understand the scale don't understand the purpose. >The Chinese grid
574.
by noosphr 10mo ago
Nuclear build out, wires and transformers. China has been building 5% extra nuclear capacity every year for the last 30 years. On target for making up 24% of their energy mix in 2060.
575.
by noosphr 10mo ago
Even the people who understand the scale don't understand the purpose. The Chinese grid isn't renewable or non-renewable. It's built to keep the lights on for anything short of a thousand year catastrophe. Their 2060 plan has
576.
by noosphr 10mo ago
People should take this as a lesson on how much we are being subsidized right now. Claude code runs into use limitations for everyone at every tier. The API is too expensive to use and it's _still_ subsidized. I keep repeating myself b
577.
by noosphr 10mo ago
Xfce is the definition of comfy computing. They have a visual language that's not changed for decades and just works. I prefer tiling window managers with no decorations, but whenever I have an app that doesn't play nice with xmon
578.
by noosphr 10mo ago
Collision is perhaps the wrong word. But llms definitely have trouble disambiguating different symbols of a language that map to similar tokens. Way back in the gpt3.5 days I could never get the model to do a parse of even the simplest gram
579.
by noosphr 10mo ago
Indeed, filling the adversitsed context more than 1/4 full is a bad idea in general. 50k tokens is a fair bit, but works out to between 1 and 10k lines of code. Perfect for a demo or work on a single self contained file. Disastrous for
580.
by noosphr 10mo ago
https://arxiv.org/abs/2209.04881
581.
by noosphr 10mo ago
I've been building Ai apps since gpt 3 so 5 years now. The pro AI people don't understand what quadratic attention means and the anti-ai people don't understand how much information can be contained in a tb of weights. At the
582.
by noosphr 10mo ago
Thieves don't care about locks, so doors are pointless.
583.
by noosphr 10mo ago
Use a license that doesn't allow it then. Not everything needs to be mit or gnu.
584.
by noosphr 10mo ago
I've found that short symbols cause collisions with other tokens in the llms vocabulary. It is generally much better to have long descriptive names for everything in a language than short ones. An example that shocked me was using an x
585.
by noosphr 10mo ago
Each section is quite standard in presentation which isn't a bad thing. I love that each citation has back references to _all_ the places that it is cited from. I wish more books did this.
586.
by noosphr 10mo ago
So just diagnostic manuals then?
587.
by noosphr 10mo ago
Are we at the level of needing a license to read a medical textbook too?
588.
by noosphr 10mo ago
You need a lot more money than $200 if your code base is more than 100,000 lines. If only more people understood what quadratic attention means in the real world.
589.
by noosphr 10mo ago
The only time I've felt this much behind was in high school when everyone was talking about how much sex they were having. AI code is the Canadian girlfriend of programming.
590.
by noosphr 10mo ago
All embeddings of hyperbolic space into eucleadean space suck. You can't preserve distances and areas between them. Trees live in a hyperbolic space so every visualization of trees on a screen will suck in some way. This simple math fa
591.
by noosphr 10mo ago
Stuff like this is fascinating. Truly the road not taken. Tin foil hat on: i think that a huge part of the major buyout of ram from AI companies is to keep people from realising that we are essentially at the home computer revolution stage
592.
by noosphr 10mo ago
g
593.
by noosphr 10mo ago
Yes, submarines swim just like how people sail the breaststroke.
594.
by noosphr 10mo ago
LLMs think in the same way submarines swim.
595.
by noosphr 10mo ago
Sound like you should use a BSD instead of Linux if you don't like choice.
596.
by noosphr 10mo ago
At 34 hours of downtime that's two nines of uptime At this point my garage is tied for reliability with us-east-1 largely because it got flooded 8 month ago.
597.
by noosphr 10mo ago
Feast and famine. In three years we will be swimming in more ram than we know what to do with.
598.
by noosphr 10mo ago
If you can I strongly suggest going to a Bortle 1 site and staying there for a month, preferably in winter. The sheer amount of _stuff_ in the sky is mind boggling, the silence is deafening. That we spend all of human existence until little
599.
by noosphr 10mo ago
Ha, amazing, I build the same 24 hour clock with night time for daily planning. You can use an js SVG animation to have it run in real time: https://tomchen.github.io/animated-svg-clock/clock.svg I have no idea why we
600.
A bitwise reproducible deep learning framework(github.com)
39 points by noosphr 10mo ago|1 comments

More ›