Where do you think the AI got this awareness from in the first place? Everything it has learned is from what humans have done before. So unless you are saying everything has been done before and there is nothing new to invent or create, then this clearly cannot be true.
Not at all. As companies refine their models to perform better at specific tasks, they need more data to train on that specific task. I know first hand that the frontier labs are reaching out to all kinds of companies asking to license their software to run so they can train on it. The stated reason is not to reproduce that tool, but to get better at operating it, though history tells us that reproduction will naturally follow even if unintended.
We have talked to many customers who specifically wonder if our software could learn to perform better for them by learning from their interactions - they are suggesting we train on their work! This tuned model would only be allowed within their enterprise, of course. It's understood that the models know how to do this generally, but every situation is different, and results could be better if taught how these specific situations work.
Breach of contract at an enterprise level is a corporate death sentence. Individuals might get screwed over but billion dollar corporations do not.
As an example, in the industry I work in, software license enforcement is quite easy to break if you want to. But you never would, because if your employer found out they would fire you on the spot. They don't want to violate their license agreements even if it meant free use because of the legal liability and the future loss of access to that product. Enterprises make a point of tracking their own license usage specifically to make sure they don't accidentally end up in violation. The systems are mostly about tracking and auditing for compliance, not about technical blockers to make it impossible.
The issue is that the contract can not be enforced. For your company to be penalized for violating the license agreement, first it needs to be detected by the license provider. If you are sending your code to OpenAI/Anthropic to get inference, how can one detect that it is not being copied by them and being used for training?
How much of your data do you give to Azure or AWS? If you use Microsoft enterprise products or Google Workspace, all of your most sensitive data, such as corporate level strategy and trade secrets, are being shared with a third party. Buyers accept this because the sellers have a good reputation and they have strong contracts in place. Enterprises form strong contracts with their AI suppliers, though I don't think they are trusted quite as much.
Patently false. The mag 7 all use AI extensively. I'm part of another 300k person company who uses it extensively. We have been slow to allow it as we come up with enterprise contracts and policies and figure out if generating code jeopardizes our copyrights, but the company as a whole is encouraging use of AI.
There was no cyber attack here. They released a cert with a trivial key. They sold a lock with no pins. When people realized what was being delivered, they blew the whistle. They were killed by a self own of epic proportions.
Yes it did help, at least back in the day when they weren't doing tons of RL. If you just train on random text, your prior probability is to return the average of everything. Ask for software, get average software, or worse yet, randomly get either gold or shit. By adding constraints to your prompt like "you are an experienced engineer" it will prefer to return answers like an experienced engineer would give and less like a random noob would. It doesn't suddenly make it smart if it never was, but it helps to reduce the variance and bias towards the good outcomes.
They certainly are! Nvidia creates and maintains an awful lot of software. Selling GPUs is what gives them revenue but they produce a lot of software to generate additional pull through demand for their hardware.
reply