rgbrgb

15 ms·

181.
by rgbrgb 2y ago
Speed is useful for batch tasks or doing a bunch of serial tasks quickly. E.g. "take these 1000 pitch decks and give me 5 bullets on each", "run this prompt 100 times and then pick the best response", "detect which
182.
by rgbrgb 2y ago
> Last year I got my hands on a rare, black PlayStation 1. This is called a Net Yaroze and is a special console that can play homebrew games as well as ordinary PSX titles. It was part of a special Sony project to get hobbyists and stude
183.
by rgbrgb 2y ago
Cool, congrats on releasing. Have you seen graphile worker? Wondering how this compares or if you're building for different use-cases.
184.
by rgbrgb 2y ago
I’ve heard that these guys use it to run a fairly successful fund https://numer.ai/ Not an expert but my understanding is that you can vend encrypted versions of otherwise proprietary datasets semi-publicly for random data
185.
by rgbrgb 2y ago
One of my fav resources for binomial experiment evaluation + a lot of explanation: https://thumbtack.github.io/abba/demo/abba.html
186.
by rgbrgb 2y ago
We were all remote and cut together separate videos when we did it in W15. 1 data point, but I know it's not a dealbreaker.
187.
by rgbrgb 2y ago
cool! i want this for safari please. is that an easy port?
188.
Happy DOM: headless web browser in JavaScript(github.com)
3 points by rgbrgb 2y ago|0 comments
189.
by rgbrgb 2y ago
Seems like there’s some confusion about what this is. I read it as an open source alternative to something like https://www.imgix.com (a product we used at my last company). Granted there are a lot of those (e.g. imageproxy) but
190.
by rgbrgb 2y ago
what is wasm-console / what is a wasmshell?
191.
by rgbrgb 2y ago
And don’t forget, people change too, sometimes becoming misaligned over time. Maybe that’s a little like cancer. Org resilience is hard to model.
192.
by rgbrgb 2y ago
Very interesting marketing development but I do not at all understand why that would be a feature. Would love someone to explain. "No crypto" labels in the last cycle made sense to me. Similar to "no ads", it points to t
193.
by rgbrgb 2y ago
It’s not about ads, it’s about distribution. If people find your product via google then SEO matters (at least for the page they land on).
194.
by rgbrgb 2y ago
This looks nice, first time I’m seeing it. Just curious, what feels rough?
195.
by rgbrgb 2y ago
> 50% accuracy on the public test set for ARC-AGI by having GPT-4o Isn't the public test set public on github and therefore GPT-4o trained on it?
196.
by rgbrgb 2y ago
It’s encouraging that we can get domains like aviation to 5 or 6 9s of reliability with human scale systemization. To me, that implies we can get there with AI systems at least in certain domains.
197.
by rgbrgb 2y ago
great context. the whole narrative you present kind of begs the question of whether swift could actually be a good systems language. as a SwiftUI app dev user I feel like this (and the OP's post) lines up with my experience but I'
198.
by rgbrgb 2y ago
I use metamask and safari is not supported.
199.
by rgbrgb 2y ago
yep, incredibly convenient. big reason i use safari as the default on my MacBook too. honestly safari is great these days for everything except web3.
200.
by rgbrgb 2y ago
haven't done anything with a model so small. what kind of stuff would you do with it?
201.
by rgbrgb 2y ago
looks nice! another open source one of these I found intriguing: https://github.com/jasonjmcghee/rem i personally worry about these slowing down my machine but I'm torn because for privacy reasons I'd only ev
202.
by rgbrgb 2y ago
i found a nice thread-level walkthrough of this paper by the first coauthor here: https://twitter.com/jacob_pfau/status/1783951795238441449
203.
Lark: Gsuite Competitor by ByteDance(larksuite.com)
1 points by rgbrgb 2y ago|0 comments
204.
by rgbrgb 2y ago
It's a 480B mixture of experts model with 17B active at any time. Looks like llama3 70B beats it on every eval they show. I remain super impressed by llama3 and just deployed it for a (simple) production use-case. Excited all these lar
205.
Snowflake Arctic Instruct Open LLM(huggingface.co)
1 points by rgbrgb 2y ago|1 comments
206.
by rgbrgb 2y ago
I’ve seen several people fine tune mistral 7B on MacBooks.
207.
by rgbrgb 2y ago
Good idea, that could make for a pretty interesting eval. It's similar to a timed test... we don't really care how long it takes or how much scratch paper you needed as long as you deliver the correct answer within the time limit.
208.
by rgbrgb 2y ago
> just for better writing in my use cases, better writing makes a better answer
209.
by rgbrgb 3y ago
Here's my naive intuition: in general bigger models can store more knowledge but take longer to do inference. MoE provides a way to blend the advantages of having a bigger model (more storage) with the advantages of having smaller mode
210.
by rgbrgb 3y ago
Please no. 0 chance I want my phone to be controlled by some enterprise device management crap. Employers should provide a dedicated 2fa device (maybe a phone) if the employee wants but I can't think of the security case for employers

More ›