Wonderfall

9 ms·

1.
by Wonderfall 2y ago
I do actually read a lot about LLM interpretability, and this is my own conclusion (should've phrased it like "they don't seem to have"). I do actually consider this an open question, so I'm a bit confused as to why
2.
by Wonderfall 2y ago
> From your article and comments, it sounds like the take is something like "humans don't actually reason autoregressively" which could be true, I don't know enough to know, but sort of like saying physics models aren
3.
by Wonderfall 2y ago
I admit my phrasing was poor there, and I got too excited. I will clarify since I don't really disagree with you or what others said (claiming it's the best way is an overstatement). Well, one could say that neural networks pionee
4.
by Wonderfall 2y ago
This is mainly a misunderstanding due to the way I phrased it. This is what I think. I know for a fact that is the case for other AI researchers having watched many conferences - "all of them" is not what I meant (I wrote "ma
5.
by Wonderfall 2y ago
I come from a medical science background, where I studied the brain from a "traditional" neuroscience perspective (biology, pathology, anatomy, psychology and whatnot). That the best way is actually to try to recreate it is honest
6.
by Wonderfall 2y ago
I completely agree. I never said that "intelligence cannot be auto-regressive", I just questioned whether this can be achieved or not this way. And I don't actually have answers, I just wrote down some thoughts so it would sp
7.
by Wonderfall 2y ago
LeCun is for sure a source of inspiration, and I think he has a fair critique that still holds true despite what people think when they see reasoning models in action. But I don't think like him that autoregressive models are a doomed
8.
by Wonderfall 2y ago
Yeah, this is mentioned in the article. (The LLaDa paper is even what triggered its writing!)
9.
by Wonderfall 2y ago
I guess semantics matter. Language is primarily hierarchical, but its presentation is what's linear. And LLMs mainly learn and work from this presentation; the question is, and one of the main points, whether emerging patterns is enoug
10.
by Wonderfall 2y ago
> There's not much signal here, just basic facts about LLMs and then leaps to very bold statements. The article wasn't supposed to be informative for people who already know how LLMs work. Like the title said, just wanted to wr
11.
by Wonderfall 2y ago
Author here and I welcome the feedback, but I don't really understand your point. My post is clearly not dismissive of efforts to make LLMs reason using CoT prompting techniques and post-training, and I think such efforts are even ment
12.
Some thoughts on autoregressive models(wonderfall.dev)
79 points by Wonderfall 2y ago|58 comments
13.
by Wonderfall 5y ago
Unsure why you think the storage access framework doesn't let you grant access to a directory. You can literally try it on a relatively modern app such as Lemuroid: the system file picker lets you grant access to your ROMs folder. Whol
14.
by Wonderfall 5y ago
Stock OS ships security updates on the latest major version. It means that you can only get a given patch level on the same version for a given device. CalyxOS wasn't rebased on Android 12 until fairly recently. As of January 2022 (pri
15.
by Wonderfall 5y ago
Once again, these management features don't expect an app to bypass the trust boundaries by allowing the user to arbitrarily add repositories. There is a to-be-exposed toggle that allows user profiles to install apps from trusted sourc
16.
by Wonderfall 5y ago
Sounds like a you problem: https://developer.android.com/reference/com/google/android/t... https://developer.android.com/reference/android/os/UserManag... You're welc
17.
by Wonderfall 5y ago
Any regular app can be considered an installer. Such APIs like the one controlled by requireUserAction which allows seamless app updates since Android 12 are declared in the app in question, and can even allow apps to update seamlessly. The
18.
by Wonderfall 5y ago
This paper is not exhaustive and there is further documentation on the APIs in question on the Android official website. You can easily guess the problem involved with the security model when the OS expects an app repository to represent a
19.
by Wonderfall 5y ago
You misunderstood what they said. Indeed, Android can have multiple app repositories and this is an integral part of its security model design. However, for the security model to be respected, each app repository should represent a single s
20.
by Wonderfall 5y ago
Well, surely there are biases in some ways, including some I may not even be conscious of. That being said, I try to follow a fact-based approach: yes, the conclusion is biased in the sense that it was written in a personal context (again,
21.
by Wonderfall 5y ago
> False dichotomy. I still have the sandbox with F-Droid. This is chaff. I didn't imply that you wouldn't benefit from the app sandbox by using F-Droid. I meant that the practical approach to privacy should come from relying on
22.
by Wonderfall 5y ago
Thanks for the heads-up! It's also a new account since I've never posted on HN before, so maybe that's why. (Also I would like to correct myself on my previous comment: I meant "GPG" as the reference implementation
23.
by Wonderfall 5y ago
Sorry, of course meant GPG, OpenPGP being the standard on which GPG is based on.
24.
by Wonderfall 5y ago
Funny you should mention NewPipe since they discourage users from relying on the official F-Droid repo for the very reason of slow updates, which you don't want for an app relying on parsing web pages that are likely to change. The app
25.
by Wonderfall 5y ago
> because there are no restrictions on things that just work with root, donation links for open source open geo data contribution platforms[1], roll your own payment scheme without giving anyone a cut, put ads in it if you like... Respec
26.
by Wonderfall 5y ago
I fail to see how it's fallacious since you're confirming my point. > I'm not starting with an app, adding my trust of the developer, and adding my trust of F-Droid You are. You seem to believe they actually read the whole
27.
by Wonderfall 5y ago
There is literally a link in the article that takes you to the Debian website: https://wiki.debian.org/Teams/Apt/Spec/AptSign I might rephrase that since they didn't technically moved away from OpenGPG (
28.
by Wonderfall 5y ago
That's not a sustainable approach as these old APIs will be obsoleted by a future version of Android. The correct approach is to follow modern practices that will guarantee security & privacy. In the case of Termux, they're we
29.
by Wonderfall 5y ago
Author here. I'm neither a GrapheneOS project member or an Accrescent developer. I'm trying to push these solutions because I think they actually do things right. It's really frustrating that the first reaction for some would
30.
by Wonderfall 5y ago
Author here. You're wrong, as I spent several paragraphs explaining how Play App Signing works and how it's being enforced for new app bundles. The Huawei app repository has a similar approach to Play Store, and to my knowledge Am

More ›