Microsoft is the winner here. They will probably use Sam/Greg's technical know how to reproduce GPT4 internally, and also direct future research based on current OpenAI approaches which they are certainly aware of. This also shields Microsoft from being dependent on an external entity that they cannot control.
Anyways, Satya played very smart with the hands he was dealt, got what he needed.
Not just Sam and Greg. From all rumors, they'll get a cream of researchers from OpenAI, plus access to most of already developed tech. Not sure about training data etc. that they may have to recreate. And from my understanding, in LLM tech know how is more valuable than actual data. If you know what to get, few 100 million should get them that.
The crypto bros switched to AI hype and are now hyping OpenAI/GPT4 hoping to pump MSFT/NVDA. In every HN conversation where someone mentions competing products, there are people talking it down and saying GPT4 is miles ahead, and in a tone to undermine the competition. I see a pattern and it is definitely not sincere.
I for one thought Karpathy would side with the core researchers and not the corpos. To me, this whole ordeal is a clash between profit motives of Sam vs Non Profit and Safety motives of OpenAI's original charter. I mean didn't HN hate when OpenAI changed their open nature and become completely closed and profit oriented? This could be the healing of the cancer that OpenAI brought to this field to make it closed as a whole.
One is Sutskever, who believes AI is very dangerous and must be slowed down and closed source (edit: clarified so that it doesn't sound like closed down). He believes this is in line with OpenAI's original charter.
Another is the HN open source crowd who believes AI should be developed quickly and be open to everyone. They believe this is in line with OpenAI's original charter.
Then there is Altman, who agrees that AI should be developed rapidly, but wants it to stay closed so he can directly profit by selling it. He probably believes this is in line with OpenAI's original charter, or at least the most realistic way to achieve it, effective altruism "earn to give" style.
Karpathy may be more amenable to the second perspective, which he may think Altman is closer to achieving.
Now regardless, the new CEO Shear is also very much in the current development of AI is dangerous (not just hypothetically in the future as AGI becomes more plausible), comparable to a nuclear weapon, and wants to slow it down. This will definitely pit researchers into camps and have plenty looking at the door.
Karpathy is a very agreeable guy and a fantastic educator, and he's very respected by everyone including leader-owners like Altman and Musk, but he doesn't seem like he has very strong opinions one way or another about the hot button issues.
Karpathy is a hybrid. He’s smart, but he clearly enjoys both the money and the attention. This is the guy who defended Elon’s heavily exaggerated self driving claims when the impact was actual human lives.
> This could be the healing of the cancer that OpenAI brought to this field to make it closed as a whole.
I don’t know. The damage might be permanent. Everyone is probably going to be way more careful with what information they release and how they release it. Altman corrupted the entire community with his aggressive corporate push. The happy-go-lucky “look what we created” attitude of the community might be probably gone for good. Now every suit is going to be asking “can we make massive amount of money with this” or “can I spin up a hype train with this”.
Yes, I felt the same. In every piece, there was very little news but a lot of fluff to lead the public with opinions. Probably VCs saw their money burning and wanted Sam back at the helm to protect their asset.
That is just current narrative, we don't know the details right? There was immense pressure from investors/Microsoft and board had to have that meeting. But the board probably already made their mind and did not balk under pressure.
Yes, the media and the general conversation seems very one sided; and I do not see any basis to take a side. Sam being an overachiever, might actually be acting with maleficence and deceit. Why is no-one even considering that?
Everyone is just reiterating that board is inept and trying to undermine them. This does not sit right with me.
Tridactyl exists in the Arch community repos [1] if you would rather use that; you're effectively shifting your trust to an Arch trusted user away from some random person (me) and Mozilla.
Can the FP32 and Tensor core modules compute at the same time, or are they an abstraction over the underlying silicon, and we can only utilize one at a time?
Here I give my analysis of what the best solution could be. Would like to know your thoughts too.
0. Lower rate from 30% to 15% -Apple still has monopoly, nothing changes.
1. Side loading apps: Not good! malicious apps can run amok, eg. $BIG_CORP$ will say - you will get 20$ credits if you sideload our app, and then surveil everything that is possible on the device. Here, we expect an average user to give all the permission that the app requests for.
2. allow secondary app stores: -not good as it depends on the quality of enforcement in secondary app store. For Apple, it is in their interest to maintain app quality in their appstore to maintain overall good user experience in their device, but motivations are not same for secondary app store. May allow malicious apps which deteriorate the user experience/privacy similar to 1. And there will be a state where you will have to install 10s of app store just to install specific programs which is also not ideal.
3. Allow secondary payment methods: -Average user will have to give up their payment info to everyone who asks for it. Most of them will not be trustworthy nor we can expect all of them to maintain good security standards for saving payment info.
The biggest culprit of all this drama is Apple does not allow secondary payment inside apps AND also, if you have secondary payment outside of app, they do not allow that price to be lower. There is no competition, thus Apple can get away with whatever it chooses to. Thus the monopoly.
4. SOLUTION:
a) charge a flat fee for reviewing/serving apps. If necessary, linearly increase it based on daily active users if they need more resource to support that app.
b) allow whitelisted secondary payment providers. Only whitelisting few payment methods which are trustable eg applepay, googlepay, paypal, stripe, etc will maintain security of payment data.
c) allow secondary payment price to be lower than Apple.
With this solution, there will be competition between payment providers which will drive the price down.
Side loading apps would not and does not have to be bad. iOS is protected by technical restrictions as well as ToS restrictions. A sideloaded app would not be able to grab your gps and contact list without requesting permission because those are technical restrictions. It would however be able to show adult content and 3rd party payment processing because those are ToS protections.
When I worked in ad tech I, I was tangentially a part of a project where we were working with a sister company to integrate in their app to enable data collection.
My work wished to use fine-grained location and there was concern that our integration and usage of the gps API's in an app that didn't otherwise have a good reason to use them would cause it to be rejected (apparently this had happened before). I don't know for certain whether this ended up being the case, but I would certainly believe it. If there was a 2nd App Store that didn't enforce standards in the same way as the Apple run-store, I absolutely believe ad-tech companies would go to lengths to push clients to using it so they could vacuum up more data.
Android has had sideloading for like a decade and no major app is installed via sideloading. Its entirely used for beginner developers, open source repos and piracy. The inconvenience of being out of store is bigger than the benefits of being in store. I think ultimately epic does not want the solution to be sideloading since they are having the same issues on android. They want to be on the app store and to have no fees.
If I pay you outright for the device, you should have zero input about what software I run on it. Give me a little checkbox that says "I will not go to the Genius bar" and let me have at it.
Here I give my analysis of what the best solution could be. Would like to know your thoughts too.
0. Lower rate from 30% to 15%
-Apple still has monopoly, nothing changes.
1. Side loading apps:
Not good! malicious apps can run amok, eg. $BIG_CORP$ will say - you will get 20$ credits if you sideload our app, and then surveil everything that is possible on the device. Here, we expect an average user to give all the permission that the app requests for.
2. allow secondary app stores:
-not good as it depends on the quality of enforcement in secondary app store. For Apple, it is in their interest to maintain app quality in their appstore to maintain overall good user experience in their device, but motivations are not same for secondary app store. May allow malicious apps which deteriorate the user experience/privacy similar to 1. And there will be a state where you will have to install 10s of app store just to install specific programs which is also not ideal.
3. Allow secondary payment methods:
-Average user will have to give up their payment info to everyone who asks for it. Most of them will not be trustworthy nor we can expect all of them to maintain good security standards for saving payment info.
The biggest culprit of all this drama is Apple does not allow secondary payment inside apps AND also, if you have secondary payment outside of app, they do not allow that price to be lower. There is no competition, thus Apple can get away with whatever it chooses to. Thus the monopoly.
4. SOLUTION:
a) charge a flat fee for reviewing/serving apps. If necessary, linearly increase it based on daily active users if they need more resource to support that app.
b) allow secondary payment providers (whitelisted only).
c) allow secondary payment price to be lower than Apple.
With this solution, there will be competition between payment providers which will drive the price down. Only whitelisting few payment methods which are trustable eg applepay, googlepay, paypal, stripe, etc will maintain security of payment data.
Allow side loading of apps. But broadcast that this device has sideloaded apps via an API. So banking apps can be disabled themselves if there are sideloaded apps. Block iOS updates. Also voids warranty.
Secondary payment method can be implemented along with apple's own payment system just like how it is for `Sign in with Apple` . If you have `Login with Facebook`, then there should be `Sign in with Apple`. The alternative payment option can be turned on and off in settings. If it is turned on, then no more Support from Apple
Anyways, Satya played very smart with the hands he was dealt, got what he needed.