Deeply disappointing distortion?

Have I been pulled too far into the Reality Distortion Field?

Either I am missing something or the AI hype for MacBooks is missing an asterisk.
I am very happy for the overall build quality of my MacBook Pro but I think local AI quality, and definitely speed versus battery life is overhyped.

I am beginning to think I would have been better off with a MacBook Air/Neo plus Mac Mini/Studio combo, rather than the MacBook Pro.

As noted earlier, I feel like running Mac software in a remote desktop without a Mac keyboard is painful (latency and keybindings on my Chromebook was horrible), so an Apple Silicon Mac laptop with Xcode and iOS Simulator, etc is fantastic.

But I haven’t gotten much traction out of local AI models and anything agentic yet bar some “trivial” single page web apps.

So far I have used various models in LM Studio and am experimenting via Cline in Visual Studio Code. I saw some odd behaviour in both that stem from the classical unrepeatability of AI. Something that works once fails later and it fails to learn.

It’s definitely a significant step down from the cloud models. However while I haven’t gotten much out of them, the alternative makes me uncomfortable with the prospect of exposing my ideas and errors to the providers. Their unclear application of intellectual property principles and arbitrary local command execution and permissions definitely gives me pause before launching into cloud AIs beyond the chat bot. I would definitely want a better sandbox!

The speed and battery life tradeoff was enormous. I guess I shouldn’t be surprised, it should be akin to gaming (which I haven’t done). The other night when I did experiment with Cline I found it burning up my lap and I charged the MacBook Pro twice, and with the supervision (which, I feel it needs) I lost interest before it completed the task.

The reasoning and planning was good, happy with that, it was the execution into my code base and arbitrary tool calls that failed.

I know there is (must be) a way but I haven’t investigated yet for leaving the laptop tethered and having it send its supervision prompts to my phone for approval, and then have it send the test builds over Wifi deployment, but ultimately the next step is to dust off the gaming rig, possibly enter the minefield of upgrading that GPU, and use that as the near-local thinking server.

I see the value in the AI, and so I should likely just pay instead but… I’m not there with the privacy yet.
🤐