matusp

18 ms·

301.
by matusp 3y ago
this is beautiful
302.
by matusp 3y ago
I think it's more readable. It matches with the type name, it is consistent when you define different structures (list, set, dict vs [], set, {}), you can pass it as a function, etc. [] and {} is unnecessary syntax sugar, not to mentio
303.
by matusp 3y ago
I wouldn't say it is an unexpected behavior. I remember reading papers about this memorization behavior few years ago (e.g., [1] is from 2019 and I believe it is not the first paper about this). It should be expected from OpenAI to kno
304.
by matusp 3y ago
I had similar dilemmas, I wanted something that is easily portable and that will work in 20 years. Ultimately I wrote a very simply system myself and I write most of my blogs in HTML with a little bit of custom templating that I implemented
305.
NLP Models Think about Gender Stereotypes(opensamizdat.com)
2 points by matusp 3y ago|0 comments
306.
by matusp 3y ago
The type of projects where you want to be able to run them in 20 years.
307.
by matusp 3y ago
That's an argument from authority fallacy. It doesn't matter how many citations you have, you either have the arguments for your position or you do not have them. In this particular context, ML as a field looked completely differe
308.
by matusp 3y ago
I agree that docker is great. But docker solves only one of the problems mentioned above (env) and even that solution does not work for some teams that run their experiments in GPU clusters where docker is not supported.
309.
by matusp 3y ago
Another issue is that making things reproducible costs you time and that is exactly what most researchers do not have. For example, many ML papers have code that is just a barely working Jupyter notebook. To make it reproducible you would h
310.
by matusp 3y ago
Claude by Anthropic has 46% winrate with GPT4 according to Chatbot Arena. That is pretty close.
311.
by matusp 3y ago
This seems like a valid exception from this rule. The interesting tidbit here is what the title suggests, even though it is not the title of the paper. It should be possible to point out to facts like these hidden in papers, reports, etc wi
312.
by matusp 3y ago
You need to rerun eval after each LLM update. GPTs have a new version every couple of months and their capabilities can change quite drastically pretty much randomly. Maybe they will make it more robust in time, but I think this is the feat
313.
by matusp 3y ago
I like Carmack but is it really necessary to have a post about everytime he mentions AI (which is every week for the past year)? So far, his track record in AI research is non-existent.
314.
by matusp 3y ago
Also kinda awesome -- Theft.
315.
by matusp 3y ago
He most certainly didn't. I am pretty sure it was used in ML in early 2000s, especially in vision.
316.
by matusp 3y ago
It's as if creating a new state from a thin air without any consideration for the ethnical minorities that will stay there was a bad idea. The problem here is that US politics in the region was never meant to create peace, it was meant
317.
by matusp 3y ago
I really do not see what google did wrong here. There should be no flag-planting in the open source community. The fact that one person develops an idea does not mean that nobody else has a right to develop their own version. In this case t
318.
by matusp 3y ago
What about a population with that size AND age distribution?
319.
by matusp 3y ago
I am quite skeptical about this kind of studies. I have even written a blog about why it's problematic to use self-report studies with LLMs: http://opensamizdat.com/posts/self_report/
320.
by matusp 3y ago
There are tons of kids still doing games. The AAA scene is fully professionalized, but if you look at what's going on at itch.io, there is no lack of young talent. The older generation forgot that when they were starting there was no m
321.
by matusp 3y ago
Isn't that also a part of the same sclerosis?
322.
by matusp 3y ago
It is interesting to see how much of the replication is done by the Chinese and how little is done by the Western countries. Is this the difference between the making-stuff-happen attitude and the sclerotic attitude?
323.
by matusp 3y ago
A rigorous peer review by graduate students.
324.
On Using Self-Report Studies to Analyze Language Models(opensamizdat.com)
1 points by matusp 3y ago|0 comments
325.
by matusp 3y ago
Is it really an invitation? The paper shows that the current models are worse for some marginalized languages that are used as OOD datasets. I am not really that surprised that the modelals don't speak those and I don't know anybo
326.
by matusp 3y ago
I used to control the passwords manually, but now I just rely on the Firefox password manager. I have it on my notebooks and in my smartphone, it syncs all the passwords, generates them and that's basically all I need. It completely ab
327.
by matusp 3y ago
I see no mention of reCAPTCHA tracking users. They are "sending data about device and application data", which seems like a completely natural thing considering what reCAPTCHA is used for.
328.
by matusp 3y ago
This is the problem with the catch-all EU regulation. They needlessly create regulations that are so vague and broad that even completely bening activities are prohibited and you need your own law department to understand what exactly is re
329.
by matusp 3y ago
Peer review is not a certificate of truthfulness, peer review just says that a grad student reading the paper did not found anything too suspicious there and it looks like a typical paper in that field.
330.
by matusp 3y ago
I am actually not sure how wide it will end up. Dnipro is actually reservoirs all the way up to Kyiv. It will be Ukrainians in Dniper HES that will control the flow from now on.

More ›