1. When this did occur in ChatGPT, you could just ask the API version of the same model and it will tell you without erroring out.
2. There is a notable enough David Mayer and you can see that by Wikipedia that isn't named in the article. However, since the API version of ChatGPT works to answer who is this person, it probably was a moderation issue with the general public versions instead of a conspiracy.
3. OpenAI fixed this, and ChatGPT no longer exhibits this behavior.
1. Those in favor of identity based control over using electronic devices, attaching a real world identity to your computing use. (Including up to signed software only allowed to run on devices).
2. Those in favor of the restriction of access of the internet by a government entity (Including up to regulated speech).
This is similar to the break up Google arguments. Instead you just have to out-code it. (For the counterarguments, this applies to code-based companies, instead of infrastructure based)...
If a social media is harmful, then parents can put those restrictions already on devices. But letting the government do this gives much more freedoms away than initially thought. Not just this, but if there happens to ever be a social media that is 10 times better than all existing ones, where even more real speech discussion and conversation happens, or whatever form it is, it will still be regulated and controlled using the same methods.
I usually don't comment often, but I would usually expect someone here to bring up these concerns... which is unusual to not find them. And I mean the specific two points are the argument. Those must be flipped to allow this law to be just, which is to say, most likely not.
And a third point, even in a perfect implementation, (imagine a perfect world), you end up with the first two points still being an issue. The technological ability to restrict devices by signed software, and speech to a central authority, allows the attack surface for anyone to have the ability to control the entirety of the devices on that system, in ways that would be hard to discern. Extend this technology to other devices, like VR, and see the domino...
On top of that, some networks like Spectrum already report all the MAC addresses that are connected to it remotely to the Spectrum database, instead of just on your network panel locally (because there isn't a Spectrum network panel anymore, only the app). This means that a nation state (USA) can see real time minute by minute who is on that network, and recent devices on that network because Spectrum designed this in their firmware.
You can check yourself from the app:
Services > Devices on Network > Manage
And it will show all of the MAC addresses connected, and recently connected. Even remotely if you are not logged into your network.
You also can see the *plaintext* password to your router from this app.
Services > Your WiFi Network
Which means a nation state also can remotely login to your network without you knowing, and otherwise is bad for security if passwords for millions of homes are plaintext.
---
Moral of the story is that even if Apple eventually fixes this, the other side of the tracking that nation states could do could be done at the ISP firmware level. To solve this kind of attack, either allowing open firmware or new legislation is the only to stop this. (Which when has privacy legislation ever happened... is another question for another day).
Yes this helps the MAC concern, but this means we need wide scale device manufactures to enable this by default, because users won't. Similar level of consensus.
Also why is some devices don't support this randomization, or even if they do, the first connection is not supported. When you first activate an iPhone or use a Windows computer, it still does not expose all the settings to randomize the MAC address until you setup the device, so the first connection exposes the actual address to the network. Yet again we need deeper levels of change to fix this.
At least now we know GPT-5 has finished development and is now in training from this (I would hope that Iyla got to add all that he hoped to before leaving).
Ilya, thanks for all you have contributed within OpenAI!
He wouldn't have left if he could advance hoomanity further there, the guy has like a 800ms delay for each word and that does not make for a very good liar, perhaps a dutiful one.
The word delay depends on who he is talking to. On his Dwarkesh interview from a year or so back he speed up noticeably, presumably because Dwarkesh is a fast thinker/talker.
The crazy part is GPT-4o is faster than GPT-3.5 Turbo now, so we can see a future where GPT-5 is the flagship and GPT-4o is the fast cheap alternative. If GPT-4o is this smart and expressive now with voice, imagine what GPT-5 level reasoning could do!
