pull down to refresh
reply
Saves vendor lock in tears too if you separate them. For me personally, spending a day or 2 generalizing a bespoke claude workflow framework into a multi-vendor, multi-bot, highly concurrent one must have been the most golden tooling I have deployed to date.
Combined with RBAC, and barring getting hacked, I have full control over what Anthropic/OpenAI/US govt knows, what the Chinese govt knows, what is kept local, and what can be done through a 3rd party TEE-based inference provider. Exerting such control over the day-to-day work queue is nice.
reply
Well... you should look at that as an access control thing. "AI" as we know it right now on its own anyway doesn't learn - it is static - so the intrinsic learning only happens if someone captures the traces and trains the next version (or another model) on it. So we're talking a peripheral database to which access can be granted. You should not ever even consider once to outsource that to one of the corporate solutions, anything that is coercible by any government, or in fact any central repository that is such a high value target that it will per definition be hacked, at all.
Instead, keep your thoughts private and off the cloud. That way, it is your database, your own evolution and no one else's business, as it is supposed to be.