• 0 Posts
  • 7 Comments
Joined 3 years ago
cake
Cake day: August 27th, 2023

help-circle

  • This is awesome! I desperately want to replace the stock firmware, an open source alternative would be fantastic. Have you poked around the stock firmware to see how the telemetry is delivered via the modem, out of curiosity? For me that’s the number one thing to stop while building out an open alternative in the meantime.


  • They are bleeding resources so bad they wouldn’t wait to put those efficient models in play. Every better model requires more and more resources.

    You are on the money about the models not being good enough, let alone a harness. This whole worry is a nothing burger. Nobody outside of these labs has terabytes of RAM to run a model anywhere near a level of concern, let alone somehow distribute it across IoT devices or something. Anyone who has run a model locally knows how woeful a small model’s abilities are.

    This is just more Rationalist cult hand wringing. They believe they are building a digital god, so of course they are dreaming all of this nonsense up.




  • Personally, I have given up, and have decided to give to the c-suite exactly what they demand, to the letter. That ends up as a bunch of AI written code that looks like it does what it does, with tests written to the best of AI’s ability, as fast as possible. I used to care deeply, and we were making strides climbing out of the tech debt that had been accumulating for years, but then AI came along and all hell broke loose.

    Our productivity goals are impossible. There’s too much to actually review and figure out exactly what it does, so I just rubber stamp PRs as long as the tests look sane and are passing, and a once-over of the code doesn’t look like anything obviously malicious. It’s a gamble on several levels, but I’ve decided to care about code quality as much as the CEO does. The CTO just bailed out and they aren’t bothering to replace them, which is telling. If they don’t give a shit, why should I? They are already trying to push everyone out and replace them with AI anyway, so why not taste the results of the policy made by slop, for slop, sooner rather than later? Either way we’re out of a job.

    After all, if this were a decade ago, they could have just hired a bunch of juniors and ended up with a bunch of spaghetti, basically roughly the same state as this AI code, and they still wouldn’t blink an eye. So what does it matter?

    In the end, I see it as a wash. Either the models get better and can clean the mess up, or just figure things out and outright replace large swaths of nfg code all at once, so why do I care… Or the whole thing blows up, the costs are astronomical compared to just having humans do it, and we have jobs fixing the mess that AI left behind.

    Even if/when the AI bubble bursts though, Pandora’s box has been opened. It’s been demonstrated that machines can write some semblance of code that compiles and executes along the lines of what a product manager dictates to the machine, as long as they can describe it well enough. Doing so seems to have astronomical costs hidden at this time, so for a while it may be prohibitive to bring things back to today’s scale. But it’s only a matter of time before someone revives it, like Steve Jobs revived the Newton ala the iPhone.