rkwasny

Yeah, now look at the incorporation dates of these companies...

Whoah it gives some seriously wrong answers to coding questions, to the point that they are dangerous!

I think we have to wait for the explanation is chat-bison PaLM 1 or 2

meilisearch 1.1.1

Yup, that's exactly what I'm saying there is something wrong with data model. I also noticed it's only using a single thread.

Out of curiosity what dataset size you guys are testing on? Indexing 2M documents on a server in 2023 should be instant, or at least <60 seconds.

Other learning is, not every software written in Rust is fast :-)

I tried Meilisearch, unfortunately despite being build on LMDB it does not work at scale.

For any collection above 1.5M documents, insert times are crazy, like 60 seconds to index additional 6000 documents. There is something very wrong with the core data model and single writer LMDB bottlenecks IMO.

Just buy gpt-4 + github copilot/cursor for 2x$20

What? we have a telescope in L2 point(the second Lagrangian point)! it is soo much better there compared to the moon

Looks like a screenshot from some sci-fi movie:

https://satellitemap.space/?constellation=starlink

Inspired by fish shell scripts seen on HN

I am running my own server @entropy.be for last 20 years. Maybe I'm lucky but it just works? The secret is only to have trusted users that do not send spam.

OK I get the occasional spam through spamassasin, but I can live with archiving 5 spam emails/day.

Unless you know the weights.

No.

What's the exact prompt?

No pricing and no waitlist ...

Of course it is possible for LLM model to learn this!

The problem was with token embeddings, not with math.

Nice! I've been using Maps.me but recently it got acquired and new owners had some problems with understanding why people use it

What's remarkable here is:

562 Billion parameters!

Even with int8 it's 0.5TB of memory to run this

Looks like the problem of worst ever California drought just solved itself

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 thousand users, OpenAI with XXX million users ...

What is a User Reservation System? a system that reserves the username when you are creating an account?!

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 up running infrastructure spanning across different buildings or even different DCs.

With Zen4 I would go with single CPU option, it's hard to saturate 256 threads.

Main cost in the cloud is always the database server, because it needs to be on 100% of the time, cannot be on spot instance, and needs lots of memory and CPU just in case of a spike of traffic.

Can Karabiner use fn as a modifier key? fn+sth would be optimal additional shortcut layer

Why not use an external dictionary?

http://fileformats.archiveteam.org/wiki/Zstandard_dictionary

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"

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 in <5 seconds.

Tests/CI etc can be run when feature is complete and you are making a PR.

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?

Oh they will, there will be no iPhones this holiday season :-)

I created one over the weekend :) https://github.com/mag-/folx

Stay tuned for launch next week! :)

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

He is going to rewrite twitter from scratch in 12 weeks :)

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" actually looks :-)

If you need colocation in London I can put you in touch.