rkwasny

9 ms·

91.
by rkwasny 4y ago
No pricing and no waitlist ...
92.
by rkwasny 4y ago
Of course it is possible for LLM model to learn this! The problem was with token embeddings, not with math.
93.
It Turns Out Money Does Buy Happiness(bloomberg.com)
72 points by rkwasny 4y ago|49 comments
94.
by rkwasny 4y ago
Nice! I've been using Maps.me but recently it got acquired and new owners had some problems with understanding why people use it
95.
by rkwasny 4y ago
What's remarkable here is: 562 Billion parameters! Even with int8 it's 0.5TB of memory to run this
96.
by rkwasny 4y ago
Looks like the problem of worst ever California drought just solved itself
97.
How ChatGPT was built from the people who made it(technologyreview.com)
6 points by rkwasny 4y ago|0 comments
98.
by rkwasny 4y ago
Pricing is good because OpenAI does not need to make any money but needs data for feedback, if everyone switches to open source ( Llama etc. ) they won't get the data they need. Google is testing their system internally with XX thousan
99.
by rkwasny 4y ago
What is a User Reservation System? a system that reserves the username when you are creating an account?!
100.
by rkwasny 4y ago
Good idea, however remember to calculate power requirements in the DC and get at least 100% more space than you currently need! I've seen cases where you want to provision the new rack but unfortunately DC is full and you have to end u
101.
by rkwasny 4y ago
Can Karabiner use fn as a modifier key? fn+sth would be optimal additional shortcut layer
102.
by rkwasny 4y ago
Why not use an external dictionary? http://fileformats.archiveteam.org/wiki/Zstandard_dictionary
103.
by rkwasny 4y ago
Turns out it does not matter if you have transformer/MLP/lstm or whatever, as long as there are enough parameters and training epochs over large dataset things "just work"
104.
by rkwasny 4y ago
Most important thing is the environment that allows you instant feedback when developing, which means just one thing: Ability to run the whole product no your Laptop/in docker and have INSTANT RELOAD, you change something and test it i
105.
by rkwasny 4y ago
They basically have far too many small regions and are growing like crazy, multi-region deployments will be a must unfortunately. Maybe you can spin up some part of the infrastructure that are not latency sensitive in the nearby region?
106.
by rkwasny 4y ago
Oh they will, there will be no iPhones this holiday season :-)
107.
by rkwasny 4y ago
I created one over the weekend :) https://github.com/mag-/folx Stay tuned for launch next week! :)
108.
by rkwasny 4y ago
Damn, I wanted to buy one, too late :( It's an amazing piece of hardware, you can mount it on a car and drive around looking for interesting places with RF emissions
109.
by rkwasny 4y ago
He is going to rewrite twitter from scratch in 12 weeks :)
110.
by rkwasny 4y ago
Yes, DC provides racking servers, plugging in cables etc. I just write instructions in a ticket. I go to the data centre from time to time, mostly taking junior engineers for a sightseeing trip to show them how "the cloud" actuall
111.
by rkwasny 4y ago
Ubuntu MAAS to provision servers - when new server arrives guys at the DC just need to plug it in and power it once, everything else is just clicking around. Pyinfra to manage and install software (I'm fed up with custom DSLs, pyinfra
112.
by rkwasny 4y ago
Anyone knows where to buy a ready-made low profile split keyboard like this?
113.
by rkwasny 4y ago
Well he didn't burn 44B, twitter market value is 44B right now because someone just bought it for 44B,
114.
by rkwasny 4y ago
Pyinfra is great, simple and no DSL! just pure python
115.
by rkwasny 4y ago
Huh, magic, obviously useful only if database size > RAM
116.
by rkwasny 4y ago
That's taken out of context! He mean that some companies that raised billions of dollars are a scam.
117.
by rkwasny 4y ago
Can be replaced with 300 servers with ScyllaDB :-)
118.
by rkwasny 4y ago
Yup, good one for a light read and thinking about the world
119.
by rkwasny 4y ago
Self driving cars will work, very soon, people underestimate recent breakthroughs in AI. ( Leaving a note on HN just to say "I told you so" ;-)
120.
by rkwasny 4y ago
I think the problem is not in the robot hardware but in the software, same as with self driving cars. What would help the development would be a "standard" hardware platform one can just buy and add own code.

More ›