pythux

No, there are some meaningful differences in the Manifest v3 implementation in Firefox: https://blog.mozilla.org/addons/2022/05/18/manifest-v3-in-fi...

One such difference (and not the least) is:

> Mozilla will maintain support for blocking WebRequest in MV3. To maximize compatibility with other browsers, we will also ship support for declarativeNetRequest.

The operating system level does not have as much information as the browser so it would be much less effective. You could do domain-level blocking (mostly) but miss lots of granularity on URLs, types of requests, contexts of request (what's the URL of the parent frame?), etc.

Also it would be impossible to perform cosmetics injections (scriptlets, alterations of the DOM, etc.)

It might already provide a good base-line but definitely not enough for all cases (e.g. YouTube ads, etc.)

Income tax rate is far from 70% (not sure if this was second degree or not).

Up to €10,777 = 0% From €10,778 to €27,478 = 11% From €27,479 to €78,570 = 30% From €78,571 to €168,994 = 41% More than €168,994 = 45%

https://www.service-public.fr/particuliers/vosdroits/F1419?l...

It is very expensive to train these base models so a smaller size is more practical if you aren’t a big company with hundreds of powerful GPUs at hand. Table 15 from LLaMA paper[1] has some insightful figures: it took 135,168 GPU hours to train the 13B version and a bit more than 1M GPU hours for the 65B version. And we are talking about A100 80GB GPUs here (expensive and scarce). Not everyone can afford these kinds of trainings (especially if it takes a few attempts; e.g. if you’ve got a bug in the tokenizer)

[1] https://arxiv.org/pdf/2302.13971.pdf

Brave has its own Rust implementation of an adblocker embedded in the browser: https://github.com/brave/adblock-rust; so it does not embed uBlock Origin (but the filters are mostly compatible)

Disclaimer: I work at Brave but not on the browser.

To add to that, I believe the title of this submission is a reference ("Open-Lamam: A “real” open-source project to train LLM not just checkpoints") to this project you link, since they did not (to my knowledge), release the code for the training or detailed instructions to reproduce their experiment precisely, only checkpoints.

Hi, Brave engineer here,

We're always on the look-out to improve our search ranking. Would you be able to share some of the queries you made that did not return satisfying results? (or use in-page feedback to report them automatically to us). It would be very useful.

Thanks!

Only tangentially related but I wrote some code to manipulate these filters with the hope that I would help manage/centralize them more easily: https://remusao.github.io/posts/parsing-thunderbird-msg-filt...

Leaving it here in case that’s useful to anyone.

Hi, Brave engineer here,

Thanks for your feedback, we've fixed the !i image search bang.

One datapoint: I use Slack exclusively in the browser (both Firefox and Brave), and it works well. As far as I remember on Firefox there was some limitation on video calls but that was arbitrary (user-agent check or something like that).

Are you by any chance referring to Goggles (with two 'g's) which is indeed a beta feature, but has nothing to do with Google? https://search.brave.com/goggles/discover

(Disclaimer: I work at Brave)

Note that the lack of adblocker in nativefier has nothing to do with the lack of libraires to do the blocking but rather technical challenges with preload scripts, etc. (Not impossible but someone needs to implement it). See here for relevant issue: https://github.com/nativefier/nativefier/issues/177

You might like Goggles (also a Brave feature): https://search.brave.com/help/goggles

Disclaimer: I work at Brave.

> They charge by the GB and are not exactly super cheap so if the customer wants to store a big fat file of easily compressible zeros then whatever, they got their money.

And what if they can charge you the uncompressed size and only actually store much smaller compressed files behind the scene. That seems like having your cake and eat it too.

That does not match my own experience with a very similar t14s equipped with same CPU. I get much better battery life than 3 hours on Linux even with moderate workload (compiling, browsing, etc.). I often use it on the go and it’s been pretty great.

It cannot do CNAME uncloaking because Chrome's extension API do not provide dns resolving like on Firefox. Another example if HTML filtering where on Firefox an extension can rewrite the HTML from requests on the fly (e.g. to remove script tags, etc.). That's also not possible on Chrome and uBlock Origin will only make use of this on Firefox. There are probably more limitations, these are just two from the top of my head.

Thank you, very interesting ideas. I like the option to boost based on a variable depending on the page itself. How would we define “adcount”, number of network requests blocked by some adblocker when loading the page?

Thanks for the great feedback! As you mentioned, before being able to apply multiple Goggles at once[1], we would need to define precedence rules. It's technically possible, but then we would also need to figure out some UX around it, etc. In the meanwhile, and although this might not be the most convenient, it is still possible to combine them and host them separately.

The idea of having a separate Goggles aggregator was also mentioned yesterday; imagine a site when you can click a few check-boxes of Goggles you like, then get a new link with the combination of all of them (you could then submit this link in Brave Search to use the aggregated Goggle).

In any case, applying multiple Goggles at once is something we have discussed internally, we're not yet planning on adding this but feedback like yours is super valuable for us to decide on the next steps.

[1] https://github.com/brave/goggles-quickstart/blob/main/faq.md...

Hey, I work at Brave,

This is great feedback, thank you! We already have a few ideas on how to make Goggles more powerful in the future, and the things you listed sound very interesting. Would you mind elaborating a bit on how you would see yourself using these features in a Goggle? (e.g. ads count, linking to high-quality content, etc.)

Nice project,

Goggles is also open-source, the main difference of the approach is that Goggles collaborates with Brave Search index while your approach uses any search index as source of URLs.

That difference is fundamental, a client can use a host search engine to do query expansions to build a large result set and then apply the filters and boosts that the user defines. However, that recall set is going to be in the order of hundreds URLs (more will take either too long or the client will be blocked); and I assume it would be challenging to apply thousands of filters at once. The smaller the result set, the smaller is the effect or benefit of the user-defined rules.

Goggles, because it collaborates with the host search engine — as of today only Brave search — can apply the filters and boosts to a recall set of tens of thousands of URLs. So the net effect of such rules is much larger.

Goggles is a bit more complicated, but it's for a reason.

Disclaimer: I work at Brave.

Hi, Brave engineer here,

It is true that there are currently less examples of negative boosts, but it is supported by Goggles thanks to the 'downrank' option.

For example:

$downrank,site=example.com

If you want to be stricter, you can also discard sites completely of course:

$discard,site=example.com

I hope that helps,

From what I can see the Lenses feature only allows to specify up to 10 domains? (Screenshot in the article linked above). If that’s the case I don’t see how Lenses and Goggles are comparable.

Goggles allows to specify thousands of rules, not only on domain but also URL patterns and in the future matching on elements of the page as well (e.g. titles, etc.)

It is also not clear that Lenses predates the Goggles whitepaper. We’ve been playing with the idea of Goggles for a long time.

Disclaimer: I work at Brave.

Thanks for clarifying, makes sense. I think the comment might predate the switch to boost=1 and was not updated accordingly.

Hey, Brave engineer here,

In this case it is needed because there is a generic '$discard' rule in the Goggle (which means: discard any result that does not match any other instruction from the Goggle; you can see this as a 'default action' applied to results if they are not caught by any other instruction).

Using a 'boost=1' allows you to keep some sites, that you don't necessarily want to boost more than their "natural ranking".

We have a bit more info about that in the "Getting Started" guide here: https://github.com/brave/goggles-quickstart/blob/main/gettin...

I hope that helps!

Definitely a matter of cost/benefit ratio. DDOS using headless browsers are pretty common.

Brave - Brave Search | Remote (Anywhere, but significant overlap with European timezones preferred) Brave Search is looking for talent to make search.brave.com better. Your work will impact the future of search to be used by millions of people. Brave Search is an independent search engine with its own ranking, crawling and indexes, aiming to offer a true alternative to Microsoft and Google.

* Search Engineer https://brave.com/careers/?gh_jid=3274278

* Software Engineer (Backend) https://brave.com/careers/?gh_jid=3274276

* Software Engineer (Fullstack/Frontend) https://brave.com/careers/?gh_jid=3274277

For more roles at Brave, check out https://brave.com/careers/.

Aren't both companies potentially loosing money when their products don't work? The fact that it's crucial to businesses seems to be the client perspective, not the company perspective. It could also seen as critical for some businesses to advertise on Facebook. This could call for a different culture internally but I'm not convinced this is necessarily the case.

I don't disagree with this, they could/should have transitioned already. But for one, cultures are hard/slow to change. And second, as an example, Facebook had the motto "move fast and break things" until 2014, and by that time they also were beyond the startup phase(), so this kind of culture is not only for early days.

() They were founded in 2004, that's 10 years in. By that time in 2014 they had 800M+ monthly active users and $12 Billion revenue; and they had this culture internally until this point.

I have no idea if this is remotely close to reality but, what if, their culture of breaking things and bad uptime is what allowed them to move fast and build a great product in the first place?

> The built-in AdBlocker in Brave is based on AdBlock Plus, so it is not as powerful as uBlock Origin.

This is not accurate. Brave's adblocker has compatibility with most uBlock Origin rules as well as ABP syntax (uBO is also based on AdBlock Plus syntax and extends its feature with custom additions).