As a dev how do I avoid exposing IP through coding assistants?
1•dd-sharma•1h ago
I’m new to the community.
I use coding assistants for the most part to - quickly get my refactoring done, root cause/ debugging, research of large logs and other tasks. So I’m sharing a large chunk of our code. I’m getting a bit concerned about the exposure of IP.
Comments
notrustincloud•39m ago
Have you ever heard the phrase, "shutting the barn door after the horses are gone" ?
This concern should be dealt with before leaking IP into the hands of the model owners.
It is a reasonable expectation that those owners will apply the same attitude to IP rights of the users of thd models that they have demonstrated during the training of those same models... that is to say, the willing disregard of anyone's IP in pursuit of their own enrichment.
It would seem that responsible protection of internal IP requires using hardware you own or at least fully control from the bare metal layer up, running models trained or distilled from public datasets which have been rigorously protective of privage IP rights.
or we can just all give up on the idea of knowledge and intellectual concepts as having any inherent limitability... our ideas are not 'the moat'.
The manner in which we engage with our customers will be the real moat. If anyone can recreate our software by asking the model to do the work, what is left for any business to differentiate from those copies becomes the ability to maintain a community of users/customers/participants.
notrustincloud•39m ago
This concern should be dealt with before leaking IP into the hands of the model owners.
It is a reasonable expectation that those owners will apply the same attitude to IP rights of the users of thd models that they have demonstrated during the training of those same models... that is to say, the willing disregard of anyone's IP in pursuit of their own enrichment.
It would seem that responsible protection of internal IP requires using hardware you own or at least fully control from the bare metal layer up, running models trained or distilled from public datasets which have been rigorously protective of privage IP rights.
or we can just all give up on the idea of knowledge and intellectual concepts as having any inherent limitability... our ideas are not 'the moat'.
The manner in which we engage with our customers will be the real moat. If anyone can recreate our software by asking the model to do the work, what is left for any business to differentiate from those copies becomes the ability to maintain a community of users/customers/participants.