I think it also helps them figure out which videos keep people on YouTube longer. If I scroll to a section of the page that has 6 videos, and I stare at them for 10 seconds, then scroll down, they'll know that one or two of those videos must have been somewhat interesting. But if I stare at 6 videos, then scroll away 2 seconds later, it knows that nothing in that batch was worthwhile.
The fewer videos they have in focus at a time, the more accurate their algorithms can be.
Advertisements have helped finance the web for decades. AI could be no different.
What type of advertisers would want to advertise next to an AI chat window? How often would ads show? Would the users still enjoy using the platform if you showed enough ads to offset the cost of running the service?
Lot of questions that all boil down to "it depends." None of the big players want to dilute their product with ads (yet). But I definitely think some will be willing.
My own discussions with advertisers have revealed a growing interest around the concept of conversational-based targeting (see website) and advertising. But, many are still skeptic and require additional CTR and ROI data, which is not possible since there's nothing like this on the market yet.
Ads shown would be dependent on the partner platform. For example, a platform like Cursor could deploy a simple agent dedicated to monitoring the conversation thread for ad invocation and display. This agent would be instructed to display only a limited number of them per conversation (e.g., 2 ads), based on a high-level summary + demographic information. The ad package returned would provide the text-ad itself, url link, and other necessary information. Finally, Cursor would showcase this ad within the chat tab itself, let's say after the LLM's response.
Also, after speaking with many users, it seems there's a willingness to make the tradeoff as long as ads are clearly separate from original LLM outputs, not overly targeted, infrequent, and accompanied by a clear reason for being shown. Also, only high-level contextual information + demographic data are shared. These requirements are definitely achievable.
Finally, pricing can only really be sorted if we have willing partners on both sides.
As of now, convincing advertisers and developers has proven to be difficult. It feels as if I'm speaking alien sometimes. I thank you for seeing the vision.
Good software can be art. And like all art, we have hit the stage in which code can also be cranked out en masse, thoughtlessly, for a quick buck. It was only inevitable.
The lack of domain re-verification seems important. The other things listed are the case for any social media platform, but they bear repeating.
I hope domain re-verification is fairly automatic once implemented. If I remove my Bluesky information from my DNS, it should be a safe assumption that the affiliated account will soon lose its username, maybe within a week or two. Same if I'm buying a domain; I wouldn't want lingering accounts for months or years after the fact. If it's a more manual process, that could be annoying, especially since you can also use subdomains -- someone could be "admin.example.com" and fly under the radar when selling example.com.
I find such thoughts exciting. In the future, children will be taught basic facts that, to us in the first half of the 21st century, are some of the most complicated questions of the universe.
I think knowledge about universe will be mostly the same but streamlined. In our modern day science we have a lot of concepts that exist only because of the path we took to get to where we are now.
Most of these interpretations will be cut out once a better ways to proper undestanding is found. I imagine electrons shells, wave function collapse, pseudo-vectors, relativistic mass, xyz ... will go away quickly to be replaced with more suitable concepts previously (and still) held back by necessity of humans to be able to do some math with pen and paper.
Not going away. It's based too directly in quantum mechanics principles, and the same tools are used in too may other problems with good results. If you talk to a hard core physicist, they may explain some minor corrections, but the simple model is 99% accurate and the corrections quite technical. Perhaps there is a better theory in the future, but it will be very weird, you really don't want to know it.
> wave function collapse
It's going away, but it may take 500 years. Nobody likes wave function colapse. There is people working to eliminate it, but we have no clue if it's hard, very hard or impossible. I think that a combination of the so-called-many-worlds-interpretation and something-something-decoherence will solve it in 50 years, or 100 years or 500 years. I'm optimistic, but it may take a while......
> pseudo-vectors
Solved? The problem is drawing normal vectors that are 1-forms and pseudo-vectors that are mostly 2-forms in the same space. Most pseudo-vectors are like a tiny surface area instead of a tiny arrow. But people love to draw all of them as arrows and that causes the problem. Also, in special relativity the electric field (vector) and magnetic field (pseudo-vector) are combined in a single weird entity that fixes the problem. There is still the problem with the weak force, but I think it's solved once you replace mass with the Higgs boson. So it's "solved" if you like to use a little more math and want to translate it to everyone else that likes arrows.
> relativistic mass
Solved. Most modern Special Relativity books try to avoid relativistic mass. The problem is that you need number to accelerate to one side and a different number to accelerate to the front/rear. So it's better to skip it and use other equations. The usual "relativistic mass" is good for accelerations to one side to get circular movements, so it's nice for some problems.
Has almost nothing to do with actual orbitals. "Filling electron shells", "octets" are just idiotic old rule of thumb ideas only accidentally aligning with reality.
> wave function collapse
I think it's going away pretty fast as we exprimentally find quantum behaviors in increasingly macroscopic objects. At some point it will become clear that nothing collapses into particles and it's just that through interaction wavefunctions narrow down when they exchange some energy and momentum. But other interactions can spread them apart back again. We are gonna create consistent description of the process in both directions.
> pseudo-vectors
they are still used but they are gonna be replaced by bivectors as they are more natural
> relativistic mass
true that it's partially sovled, but we need a generation or two of people not mentioning at all in educational context or metioning it negatively for it finally go away ... today it's still treated as "useful educational metaphor" which it is not
> xyz
Basically breaking down math calculations to coordinate wise caluclations
People will stop doing that because most symbolic maths in education is going to be done with computers and rarely anyone will be doing any element-wise transforamtions on anything.
It's a good rule of thumb for hand waving chemistry. It's not good enough to predict protein folding, but it's good enough to understand how amino acids connect. I don't expect it to disappear.
> wave function collapse
I disagree because I expect a different solution to the problem.
> pseudo-vectors -> bivectors
I agree. We only have to convince the other 7999999998 persons :) .
> relativistic mass
Another good rule of thumb, but I'm not sure for whom. This days nobody has to make a DIY synchrotron at home. It can probably go away, but it will resurface from time to time like a clever trick in a YouTube video.
> xyz
I like covariant equations, so I agree. Anyway, at work we sometimes use some non-covariant approximations but we add a search to optimize the base to get the best one were we can apply to nasty coordinate tricks.
Anyway, I needed like 10 years to understand the difference between a matrix and a linear transformation. 20yo probably only can use coordinates until they grow up.
> how does bluesky solve the problem of building your castle in another man's kingdom?
Bluesky (the platform) doesn't, and they acknowledge that. It's centrally owned, and is prone to all of the risks that any other centralized platform offers.
> if I do something controversial or using regulatory arbitrage, I'm interested in how AT is useful for managing that risk.
AT is completely decentralized, like email.
If your account is @motohagiography.example.com, other AT instances will make a DNS query to example.com to see if that has an entry that the AT protocol recognizes. If so, it will make a connection to that instance, and gather your content for display.
However, if a particular instance sees their a volume of unwanted accounts from example.com, they could blacklist that domain from interacting with their instance, so, even with this setup, you are at the mercy of the "big players" respecting you — just like if you try to send email to users using Gmail and Google decides you're suspect.
And, if you violate the laws of where you're located, law enforcement will handle that the same as they would if you violating the laws over HTTP or over email.
Nope. Almost everybody has more than one device (laptop, phone, and maybe a tablet) with more than one IP (both home wifi and phone data). Everyone has multiple email addresses.
You could get by with requiring a unique phone number, but that still risks excluding users, and can get expensive if you intend on catering to an international audience. Even in that case, some people may have a landline and a cell phone, or they may use a friend/spouse/relative’s phone to circumvent your limits.
> You could get by with requiring a unique phone number
In the US, anyway, you can also get burner phones for about $10 at local stores. I do this routinely if someone is requiring a phone number to register for something that I really want to register for.
It's no more or less disconnected than any other store purchase. You can assume there are cameras in any store. You can indeed pay cash. The only records generated are the usual sales records, and if you're concerned about minimizing those, then you use the same mitigations that you'd use with any other purchase.
Personally, I'm not concerned with that level of anonymity, though. I just don't want to give my actual phone number to random companies.
Fun! I played this without sound, and got all of them correct.
I think part of the problem is that I knew that some videos were fake, so I was looking to see if their lips matched other movements. If somebody is talking fast, but their body language/movements are far slower than their talking, then it’s a pretty obvious tell.
If I had just seen one of these videos out in the wild, I can’t say if I’d immediately notice they’re fake, since that wouldn’t be the first thing on my mind. I think it’s probably impossible to get an accurate test given this limitation, but this test would be good for more casual people to try (i.e., people outside of HN).
Given what you wrote, it's hard to tell one way or another what they think about you personally. Was the code stored on your personal device, or a company-issued one? If it's company-issued, it's probably nothing to worry about, since, if they were to terminate you, they could immediately restrict your access to the codebase.
I view it vastly more likely that this isn't anything personal, it's just a new corporate decision to limit who has access to the code. If someone's job is a bit more complicated, but they can still do their work, while the company is far more protected, that is a good trade-off for lots of folks.
Also, your company "looking to reduce expenses" doesn't mean anything. Every company is. You will hear that, in some form or another, in almost any organization. If they have to increase spend for cybersecurity, they will.
I see your points, and I genuinely hope you're correct—if this is merely a new policy aimed at limiting access to the code, then I can understand the broader motivations behind it. That said, given my concerns about cost and efficiency, the question becomes whether it's worth the effort to try and get leadership to reconsider. From a practical perspective, the restriction makes my job notably more difficult. The Inefficiencies introduced directly translate into lost time, hindering my ability to troubleshoot, test and debug efficiently. Over time, this could affect my productivity, or at least the appearance of it, which in turn could be detrimental when my output is closely scrutinized. The indirect, long-term impact on the product is another rabbit hole entirely.
TL;DR
If due to policy changes and my concerns are valid, do I pursue raising my concerns to leadership?
Is it your first job ? If it is, don't worry, it's way worse everywhere else. Sometimes you have committees eating many man-hours, every day, to green light releases with non-technical people having the last word, asking no question, and always, always approving.
When I do a release as a dev, I don't do it myself: someone in another country presses the buttons I ask them to press, type the linux commands I ask them to type, and accept my answer when I say it looks good. Because I am, and all my colleagues are, considered a security risk, and it's better we dictate everything to someone who has no idea what we're releasing, for security reason. We call that segregation in duty, instead of "complete waste of time".
