> Nvidia parakeet and canary are better and faster
Is that based on your own experience using those and also Whisper, comparing them side-by-side? Or is that based just on those benchmark results?
24,763 karma · joined March 2, 2012
https://bsky.app/profile/victor.earth
hn@victor.earth
aspe:keyoxide.org:Q6B7ZBQITV7IE2RG4EMVKWT4VI
Is that based on your own experience using those and also Whisper, comparing them side-by-side? Or is that based just on those benchmark results?
I helped my wife with something the other day, noticed the ads everywhere, while I was sure I had installed uBlock for her in the past. Went to the Chrome's addons page, and Google apparently is automatically disabling uBlock and calling it unsupported, yet you can enable it until next time you restart Chrome. But seems Chrome is actively trying to get rid of adblockers lately.
As someone who is currently relying on Whisper for some things, what models are those exactly? I still haven't found anything that is accurate as Whisper (large), are those models just faster or also as accurate/more accurate?
I dunno, it mostly seems to not work when companies change their field names/IDs, or just 3rd party authentication, then you need to manually add domains. Otherwise my password manager (1Password) works everywhere where I have an account, except my previous bank which was stuck in the 90s and disallowed pasting the passwords. If you find that your password manager doesn't work with most websites (since it's "extremely common") you might want to look into a different one, even Firefox+Linux combo works extremely well with 1Password. Not affiliated, just a happy years+ user.
> Passkeys seem like the best solution here where you physically can not fall for a phishing attack.
Yeah, I've looked into Passkeys but without any migration strategy or import/export support (WIP last time I looked into it), it's not really an alternative just yet, at least for me personally. I have to be 100% sure I can move things when the time ultimately comes for that.
What's "serious" exactly? Codex is open source, is software, can be run with open/downloadable models/weights.
In my testing using Gemini, Claude Code, Codex, Qwen Code and AMP side-by-side for every prompt for the last two weeks, Codex seems the best of all of them so far.
Have the TOTP in the same/another password manager (after considering the tradeoffs) and that can also not be entered unless the domain is right :)
I'd probably go looking for a new password manager if it fails to do one of the basic features they exist for, copy-pasting passwords defeats a lot of the purpose :)
> That being said, if you’re making login pages
I think we're doomed on this front already. My previous bank still (in 2025!) only allows 6 numbers as the online portal login password, no letters or special characters allowed, and you cannot paste in the field so no password manager works with their login fields, the future is great :)
Speaking of which, anyone have a link to the study itself? Since the linked marketing post doesn't even provide a link to it, just a bunch of links to their own platform and social media accounts.
Probably better for the submission to link directly to the study, without the extra call-to-action fluff at the end?
Can't really tell you what not to do, but if you're not already using a password manager so you can easily avoid phishing scams, I really recommend you to look into starting doing so.
In the case of this attack, if you had a password manager and ended up on a domain that looks like the real one, but isn't, you'd notice something is amiss when your password manager cannot find any existing passwords for the current website, and then you'd take a really close look at the domain to confirm before moving forward.
My worst nightmare is to wake up, see an email like that and hastily try to recover it while still 90% asleep, compromising my account in the process.
However, I think I can still sleep safe considering I'm using a password manager that only shows up when I'm on the right domain. A 2FA phishing email sending me to some unknown domain wouldn't show my password manager on the site, and would hence give me a moment to consider what's happening. I'm wondering if the author here wasn't using any sort of password manager, or something slipped through anyways?
Regardless, fucking sucks to end up there, at least it ends up being a learned lesson for more than just one person, hopefully. I sure get more careful every time it happens in the ecosystem.
Yeah, I don't think me myself had good reasons beyond "They seem like the good guys who won't sell out", but I was also way younger and more naive at that point (it was like 15 years ago after all).
I think I mostly just drank the cool-aid of what you mentioned as "if you weren't on GitHub, your open source software would not find contributors". There was a lot of "We love Open Source and Open Source loves us" from GitHub that I guess was confusing to a young formative mind who wanted it to become like the projects they wanted to host. This hope was especially fueled when they started open sourcing parts of GitHub, like the Gollum stuff for rendering the wikis.
Personally, I remember the initial selling point of GitHub being that it was more "social" than any other forges at the time, since we were all wrapped up in the Web 2.0 hype and what not. I think they pushed that on their landing page back in the day too.
It was basically Twitter but redone specifically for developers, and focus on code rather than random thoughts.
It can be, but also not. If you isolate them into libraries with clear interfaces, you can kind of avoid that. I think clojure.core.async is an excellent showcase in something you couldn't do in other languages, where asynchronous channels were possible to add to the core language without changing anything in the core compiler itself, and because of the small interface, you can still use it without ending up with nightmares :)
I'm not sure "frowned upon" is the right expression, but I'm not a native speaker.
The way I've internalized it, is basically "Avoid macros unless there is no other way", which basically means use functions/anything else whenever you can, but if you absolutely have to use a macro for something (like you wanna read the arguments before they're parsed), then go for it.
Do you think they would have bought it otherwise? Same for NPM, they got bought for huge sums of money because they were "critical" already.
I dunno, it'll certainly revolutionize my world once it's ready. A editor connected REPL, changes to running games on the fly while keeping existing state, using a well-designed language like Clojure but getting the performance (or similar) of C++ and native binaries.
It's pretty much a win-win-win for me, especially if I can replicate the speed of development I get with normal Clojure but for game dev.
I think there are multiple things going on. First, models are either trained with tool calling in mind or not, the ones that don't, won't work well as agents. Secondly, each companies models are trained with the agent software in mind, and the agent software is built with their specific models in mind. Thirdly, each model responds differently to different system/user prompts, and the difference can be really stark.
I'm currently working on a tool that lets me execute the same prompts with the same environment over multiple agents. Currently I'm running Codex, Claude Code, Gemini, Qwen Code and AMP for every single change, just to see the differences in responses, and even reusing the same system prompt across all of them gives wildly different results. Not to mention how quickly the quality drops off the cliff as soon as you switch out any non-standard model for any of those CLIs. Mix-and-match models between those five tools, and it becomes clear as day that the model<>software is more interlocked than it seems.
The only project I've had success with switching out the model of, has been using GPT-OSS-120b locally with Codex, but that still required me to manually hack in support for changing the temperature, and changing the prompts Codex use a bit, to get OK results.
No, none of the moderators were paid, but I do think the ~2/3 admins were paid. But yeah, I did it purely out of the want for the forum to remain high-quality, as did most of the other moderators AFAIK.
> Recent social media (& maybe "recent" no longer applies) doesn't have this kind of community run tooling
Agree, although reddit with its "every subreddit is basically its own forum but not really" (admins still delete stuff you wouldn't + vice-versa) kind of did an extreme version of community run tooling, with the obvious end result that moderation is super unequal across reddit, and opaque.
Bluesky is worth mentioning as well, with their self-proclaimed "stackable" moderation, which is kind of some fresh air in the space. https://bsky.social/about/blog/03-12-2024-stackable-moderati...
As someone who spent an embarrassingly long time on what lots of people claim to be the most toxic forum in the world (not sure about that, it's the biggest in the Nordics though, that's for sure), and even moderated some categories on that forum that many people wouldn't touch with a ten-foot pole, it really isn't that hard to moderate even when the topics are sensitive and most users are assholes.
I'd argue that moderation is difficult today on lots of platforms because it's happening too much "on the fly" so you end up with moderators working with the rules differently and applying them differently, depending on mood/topic/whatever.
If you instead make a hard list of explicit rules, with examples, and also establish internal precedents that moderators can follow, a lot of the hard work around moderation basically disappears, regardless of how divisive the topic is. But it's hard and time-consuming work, and requires careful deliberation and transparent ruling.
If you reframe the issue from "Nepal wants to punish the users" to "Nepal wants to punish the companies", implementing an easy DNS block makes a lot more sense. As long as most users are unable to access the platforms, the companies will get hurt by it, I think the idea is at least.
Yeah, of course, similarly if US decides that they need people on the ground so they could execute them in a CIA blacksite in case they commit crimes.
But obviously that's way too much, same for Nepal, not sure why you're immediately jumping to kidnapping, rather than "So a person can be put in front of a court".
> If I want to run a Mastodon instance (which is blocked by this), do I need to hire an employee/representative for every country in the world?
If you want to operate a service at scale, which you gain profits from, in another country than where you live, it's fairly common to have some sort of representative in that country, one way or another. Usually it's ignored when the scale is small, but once you reach the size of Facebook, I think it's expected that you get some representative in the countries where you operate, yeah.
> I'd rather just keep the maximum leverage most countries have as being to block the site if they don't like it.
Exactly what we saw happen right here :) Ignore the laws, get blocked, then the companies can decide if they wanna start operating again by following local laws, or exit the country.
Kind of makes sense, not every user wants 100% correctness (just like in real-life).
And if I want correctness (which I do), I can make the models prioritize that, since my satisfaction is directly linked to the correctness of the responses :)
The move seems to not be about blocking citizens access or trying to prevent communication at all, but rather to punish those specific companies because they weren't following the law, since there are companies who weren't blocked.