I am an ATproto noob, so I can not really weigh in on your opposing positions, but I think substantive arguments should be upvoted. Taunting one-liners and deflections should be downvoted.
I realize this is a heated discussion with an apparent backstory, but when someone says they are not being paid, it should ideally be the truth. The denial does not feel intentionally misleading to me, in this instance, but still: not ideal.
It is perfectly fine (and in fact standard) to identify people without any payment processor involved.
Someone arguing for the ability to pay someone without surveillance does not argue for a world where nobody knows each other anymore. That would be awful, of course.
Concealed advertising is highly problematic and deserves discussion. It is not, however, what OpenAI introduced in the article above. Their current ad formats are, apparently, transparently labeled and clearly separated.
It is very likely that this will adjust in the future. Just look at how indistinguishable from organic content some prominent ad placements have become over time. But: There is merit in discussing the actual product, not some hypothetical version of it. If we do discuss hypotheticals, we should make that explicit.
This also helps the whole debate (and therefore, I believe, ad critics): arguments grounded in reality are usually far more effective than those that aren't. ... And there might come a point where OpenAI introduces some kind of native ad format. We would be well advised to preserve our credibility for that discussion.
edit: Paid content in the organic LLM output is a current problem already. It ist just not what this announcement is about.
Cloud risks are very true - for both GoPro and DJI.
About a year ago, I was looking for an action camera. I found a good DJI product, but then I learned that you have to link the camera to an app in order to use it. I am fundamentally opposed to that idea, so I looked up the equivalent GoPro camera. Turns out GoPro also requires you to activate the camera with a smartphone app.
I don't want to send my data to a company for something as trivial as an action cam. I was surprised to find that the US model had such an evil requirement, mirroring the Chinese model.
In the end, I bought the DJI action cam (cheaper, better) and activated it in a public location using a burner phone. You only need to use the app once to free the device and take it offline, after which you can delete the surveillance app. I have found no fingerprinting in the created media yet.
GoPro would have made an easy sale with me, if they had been more privacy/power aware.
It is pretty obvious, but as far as I can see, this article has not been disclosed as advertising or sponsored content. I'm in Europe, and I'm no expert on FTC advertising requirements, but this specific content arrangement doesn't feel super legal to me. My ad blocker did not pick it up either.
I have always assumed that WhatsApp just does a lot of things for Meta that aren't immediately obvious to the user.
For example, the app insists on downloading all the "status" pictures/videos that contacts post. I never open that part of the app, so I never see any of those media files. However, WhatsApp traffic stats inform me that it apparently consumed 1.4 GB of traffic to download 16,047 status updates. I have seen almost zero status updates and i have sent zero status updates.
This is, apparently, (16.7 / 1.4 ) × 100 ≈ 8.38 % of all traffic documented by my WhatsApp usage counter. .. and it is just wastefulness, probably to boost someones KPIs or something.
Compare that to Signal, where you can disable that entire feature in the settings.
Another example: Even though I never listen to voice messages, the app insists on downloading all of them. While other media can be configured to not download automatically, downloading all voice messages is mandatory.
Maybe in some group I am part of, WhatsApp is downloading a voice message (that I'll never listen to) right now.
I mostly used GPT-Venti for the complex part, but the documentation was done by either GPT-Grande or GPT-Tall.
On a more serious note, I can vividly imagine how difficult it is to agree on a set of words that could plausibly suggest a relational meaning while remaining non-diminutive in every individual model name. Adding to the complexity, it is going to be used globally, and the main competitor already has an arguably successful, fabulous naming scheme.
I'm not sure if this is some kind of tactic, but both of these claims seem trivially wrong.
Bluesky is available in both main app stores and on the web. There is no merit in denying this. There are often differences in content restrictions between what app stores enforce and what service providers enable, but one cannot really say, "Just don't use the app," if by "app" they mean every common way of accessing the service.
Also, I think the protocol does not matter much if the user experience of almost every user is entirely controlled by a single provider. In such a case, the protocol might as well be entirely absent.
It is difficult to tell whether there is some substance here, since you do not explain (or source) anything.
Yes, it is specific to Bluesky. It is not dependent on the official app, though: Both the app and the website are subject to this kind of censorship.
There seem to be loopholes [1] through some third-party apps, but that does not change much about the overall problem. Defaults matter a lot.
I have a hard time understanding why the Bluesky authors came up with an idea like that in the first place. Not to mention actually creating it. The internet (or social media) probably does not benefit from even more granular censorship tools for governments all over the world.
I'm no expert in atproto but it seems to be far worse than that. Apparently, Bluesky applies country-specific labellers automatically based on your IP address.
If you register a Bluesky account from Germany, your account is assigned the German moderation labeler with no option to opt out. As soon as I noticed, I created a new account using a US IP address. This fixed it.
While signing up for moderation sounds very attractive to me, Bluesky's whole layered moderation approach seems designed to maximize algorithmic censorship.
Compare that to, for example, a mastodon instance (or a forum like HN) where you participate because you align with the general moderation approach cultivated by and within the community.
Law enforcement at all levels traditionally has a strong lobbying presence. Their public affairs departments are well-funded and do not cease operations just because some initiatives are delayed due to temporary push back. Transparency legislation often does not apply to their efforts either.
Unfortunately, public institutions also lobby in favor of chat control, so it's not just private interests. Europol and friends will not run out of public money.
Depending on the target audience, public actors can even avoid disclosing their actions in existing transparency registries. Technically it is very elegant.
The activist is well known. They likely knew he would answer the door, yet they still broke it down. In the U.S., you'd probably shoot some dog in that situation, if one was available.
The entire scene is probably not meant as effective policing, but as punitive theater. This also explains why they disabled the cameras, as the theater was not intended for content reuse.
Given that, I'd assume they knew he wouldn't shoot them or do anything even remotely like that.
> ... what kind of laws should LLMs (attempt to) follow?
I think being able to align your text generators with a specific set of laws, speech codes, tonality, etc. is very useful. However, all of this should be a user choice.
Your spell checker does not stop checking your spelling just because you are describing a murder scene either.
Safety slop and boilerplating should be configurable as well.
I rely heavily on DeepL Write for my day job, but I dislike the constant logouts, nagging, and laggy UI. I coded a DeepL Write replacement that uses the same layout (two text windows side by side, with the left window for entry and the right window showing suggested edits as an actionable diff), but all suggestions are based on Harper + Gemma.
When I want the program to reformulate a sentence or phrase, it sends the sentence to an AI that provides word or phrase suggestions. I've connected this to Tinfoil.sh (not affiliated) via API key.
Now, I have a much more private DeepL Write replacement with a snappy, consistent user experience that costs much less. Unfortunately, the suggestions are not as high quality. It's very much an 80% solution. It was still fun!
Most of the rest concerns scraping. The biggest project is an extraction tool for the German transparency register that I need for work.
I doubt that. There are competing sync extensions in their extension store. If you do not want to use extensions, you can sync the vault folder with any syncing app for free.
The whole data structure is designed to make this easy.
I chose Syncthing for this purpose, and it is free and works flawlessly. You can even trivially disable their native sync, as it comes as an internal extension.
Mozilla could have avoided so much drama with Pocket, VPN, AI features, etc., if they just were as transparent and liberal with critical first-party services as Obsidian is.
> Tags and banners do not work. Completely understandable that someone as dismissive and seemingly isolated as you wouldn’t understand that.
One can reduce every tool to a toy and justify it with some hand-wavy security slop, but removing capabilities destroys use cases.
The ability to control your tools is good. You should be able to run anything on your devices. Therefore, those who propose the toyification of tools should carry the burden of justifying the change.
The same infantilization of users currently happens with Signal, where high-level decision makers are asked by strangers to share their deepest secrets. Since these strangers introduce themselves very nicely, users start blurting out their secrets. ... now everyone is pretending this is a Signal problem. It is not. The world is not a kindergarten and people have agency.
A good compromise is to set a safe mode as the default and include an option that lets users confirm they know what they are doing. Obsidian already does this. Given that, I do not understand why anyone would demand to make the entire tool worse.
I wonder: What level of user effort would make you comfortable with users exiting safe modes? Would you want users to be able to run software with full permissions at all?
Chrome gutted extension capabilities for safety and now it is so useless, politically unwanted extensions have "lite" versions and every big project and their dog ship their own chromium browser.
I use Obsidian because it does not treat me like a child. They can add more nags and banners for normies, but the capabilities should remain.
> Can we really victim-blame someone for falling for an attack
The victims may well be those who are potentially endangered by the leakage of information caused by the decision maker. Regardless of that hypothetical, the person responsible for the leak is not the victim.
If you deal with highly confidential information in your day-to-day work, you should be held accountable for keeping it confidential. This is nothing new in the corporate world, so I don't see why public officials should be held to different standards.
Remember: It was apparently a phishing attack. Someone literally asked her for her credentials. It is within the capabilities of an adult to refrain from handing out important information when asked in a no trust environment. If that's truly beyond their capabilities, they should consider another profession.
I'm not arguing for a witch-hunt or anything against this specific person. Learnings should be constructive and this could have happened to many other public officials. Just, maybe.. if you or I breach protocol, let's not call us the victims.
I realize this is a heated discussion with an apparent backstory, but when someone says they are not being paid, it should ideally be the truth. The denial does not feel intentionally misleading to me, in this instance, but still: not ideal.