youssefabdelm

Does anyone know any good underrated educational resources for more "specialized" icon design?

I have a big bone to pick with these types of sites. Everything has developed this homogenized feel. Even with the slight variations in every set of SVG icons on that website, they all look the same kind of, and the idiosyncratic "fingerprint" so to speak of a specific kind of set has been erased in place of this constantly never-ending minimal style.

Don't get me wrong, these things are useful if you wanna get up and running and don't care about design being optimal, but if you want your website to have its own "character", more energy needs to be invested.

It's a bit less "anodyne" than GPT. GPT tends to give the most "mainstream" answer in many cases and is less "malleable" so to speak. I remember the differences between RLHF'd GPT and the original davinci GPT-3 before mode collapse. If you spent a while on a good prompt, it really paid off.

Thankfully, Claude seems to maintain this "creativity" somehow.

It's excellent at recommending books, creative writing, etc.

For coding, it's not as good as GPT-4, but still helps me more than GPT in certain coding tasks.

I guess people could use it as a form of post-training "compression"

For me what's compelling here is "auto-dark mode". As a user I don't want to keep making CSS changes to every site I visit to that's light mode. I wish the web in general had these 2 styles and automatically inferred from my device that "This person is a vampire, give him dark mode"

It needn't be multiscale entropy precisely, but anything similar to it.

The way I intuit it, though I could be wrong, is that as you have a "ground level" or "close" measurement of white noise, it appears complex and varied. As you go up and up in the scale, you get the "eagle eye's view" of white noise, you realize "oh shit, it's all kinda the same!"

So it wouldn't necessarily prove anything, but if the entropy dropped as you scaled, it would show that the pattern isn't really that complex, and there's a lot of hidden redundancy. I mean it is called a "pattern" after all to be fair.

Lol it repeats alright... blah blah blah "it doesn't technically repeat"... To the brain it does, it's like white noise. If I showed you an evolving animation of this shit for 3 hours it would be torture. Someone should measure something like the multiscale entropy of this pattern.

Edit: I will say the coolest thing about this is the cross-disciplinary connection hints at a metapattern: https://youtu.be/48sCx-wBs34?t=1007

I initially thought it meant $10k per massage

I wonder what the performance is on English specifically.

Edit: Just checked the paper, it seems to be worse[1][2] but feel free to correct me.

I feel like they should've just taken the Whipser architecture, scaled it, and scaled the dataset as they did.

[1] Page: https://i.imgur.com/bq15Tno.png

[2] Paper: https://scontent.fcai19-5.fna.fbcdn.net/v/t39.8562-6/3488279...

I was joking

Is there a way to get off the waitlist?

They all suck. Nowhere near Jukebox. Depressing that seemingly zero progress has been made since that.

Wish I had a shit ton of money just to do this

I love Jupyter notebooks I just wish they looked as good as observable notebooks[1], not just in the overall layout but the charts/graphs you could make in general (plotly, matplotlib, etc don't even come close to d3.js, Observable Plot, etc)... I don't know why there seems to be a hole in the Python ecosystem for good designers or something

This seems to be a step in the right direction with reactivity though. But it's not instant like Observable notebooks. But still good

[1] https://observablehq.com

Completely agree. Perhaps they were planning to fine-tune it for something though.

Fair point

title is missing: "trained in 1 week, and like most open source LLMs so far... it sucks compared to the closed source alternatives"

Great effort of course bla bla bla...

Open source really needs some benchmarking, and up their game quality-wise.

And yes I know they're expensive as shit to train... let's not keep wasting our money and actually work together, pool our resources, to make a GOOD model.

But oh no, everyone wants to put their stamp on it. "Replit did this! Look at us!"

Meh, results feel hodge podge like a bunch of models were stitched together

Worked for a few seconds, and was very impressive while it lasted... then "Sorry... something went wrong"

Can someone explain why these types of projects are compelling (running an LLM in the user's browser?)

I guess privacy and ease-of-use?

So you don't need to download llama.cpp, and run it in the terminal or something?

With image input capability?

Person: Woaah good girrrllll

Parrot: AAAHHH!!!

Rip headphone users.

They're loaded, and we know scaling works, they'd massively benefit... both in marketing and profit.

Although it is open source to be fair.

You could've said the same to OpenAI when they were scaling GPT from 1 billion to 175 billion parameters. We're all grateful they didn't follow that line of thought.

But Stability does have access to a pretty big cluster, so it's not paying cloud compute (I assume), so cost will be less, and data of course is not infinite...never stated that.

But considering 3.7 million videos are uploaded to youtube everyday, 2 million scientific articles published every year, yada yada...that argument falls apart.

At the very least implement spiral development... 1 trillion... 3 trillion... (oh it seems to be getting WAY better! There seems to be a STEP CHANGE!)... 5 trillion... (holy shit this really works, lets keep going)

That's great news, but one would think that since they're behind Stable Diffusion, that they'd use the insights behind it and scale data even more than that to result in better quality at a smaller scale model that can run on most people's machines.

Like... try 10 trillion or 100 trillion tokens (although that may be absurd, I never did the calculation), and a long context on a 7B parameter model then see if that gets you better results than a 30 or 65B parameter on 1.5 trillion tokens.

A lot of these open source projects just seem to be trying to follow and (poorly) reproduce OpenAI's breakthroughs instead of trying to surpass them.

I get where you're coming from and I wouldn't do that either but... we do use it for code, sometimes in languages we don't know, and 8 times out of 10 it works. It does know something (many things) we (as individuals, not humanity) don't know.

Nothing is being turned on its head. It's still a lottery for job hunters.

From their website (https://www.jumpstart-uk.com/):

    For Startup Employers. Join us as a Startup Partner and we’ll build out your team with proactive people.

    Pitch your role(s) to a cohort of top-quality candidates, and show them why your startup is the one to join.

    ******Meet 30+ people who have been hand-selected from 2k+ applicants in just******

    Hire people who have received 20+ hours of essential training (communication, problem solving, marketing, Excel and so on) before their first day.

    Find your next generalist or commercial hire,

Touch is pretty important though...

Agree, but it's obvious why... this is how they get clicks

So far I think what these models lack is memory of people and other things. Especially if not as popular. And probably a ton more.

E.g. try asking it "Who is Tyler Volk?"

Then try asking GPT-4 "Who is Tyler Volk?"

Then check who he is online.

TL;DR?

(Yes, GPT-4 is prob better... but by how much and on what?) A table would've been easier

i like the name it's like "shhh bb its ok"