Gregory's Server@mttaggart I do have to admit there have been some instances where I've passed code to microsoft's LLM, whatever they're calling it now to help me understand how certain bits of code work, and it can be useful as a force multiplier, but I can't stand that its being forced upon us all everywhere we go. Like, I have no idea where the model got its data. In google's case, It could have consumed a well thought-out bit of documentation, a blog post, or a collection of shitposts from Reddit.
|
18 comments
@mttaggart I cannot fathom the fact that there are companies that are like "yah, sure, spinning up our own private reactors and diverting major water supplies to power a datacenter are a perfectly rational venture to create a computer program that cannot tell you how many r's there are in the word strawberry."  @da_667 This is what I'm saying. And this isn't VC money; this is public corporation strategy intended to maximize shareholder value. How is that happening? The only conclusion that makes sense is that these execs are bought in on the (let's call it what it is) religious belief, the article of faith, that this technology must be pursued at all costs. And I guess, in MS's case, hopes that everyone's models will live in Azure data centers. @mttaggart we are the same mind on this, I think. There' is this willful suspension of disbelief that if enough tech heads say "AI is the future" it'll manifest into something useful. But all it is so far as a massive money sink. @da_667 So I'm writing this up properly, but MS Earnings Statements show their adventure has not been punished by the market. In the case of cloud, growth outstripped everything else, so I guess everyone really is drinking the Kool-Aid. @mttaggart correlation isn't causation, I think. There is a lot of sleight of hand going on here with Microsoft having tried to charge for AI services (e.g. github copilot) and just integrating AI into cloud services that nobody asked for (e.g. Microsoft 365 co-pilot). They aren't charging money for these services, just shoveling them out with whatever other stuff MS customers buy. @da_667 I'm not saying they're directly profiting from the models. I'm saying they're not being punished for it in any meaningful way. But it's worth noting that end users are not the customers for their cloud services, and we actually have reason to believe that those business may indeed want AI for the same religious reasons.  @mttaggart End users are *definitely* not the (primary) customer for Microsoft's cloud products. Some end users are also customers, but they get what the big-ticket customers ask for and pay for. Aunt Jane isn't going to have any meaningful impact on the direction of development of anything at Microsoft, even by getting their top-tier subscription. The CTO of a Fortune 500 company very well might. @da_667 @mttaggart I can’t get over how •••EVERY••• company is like, “ooh! Ohh! Me too me too let me get on the bandwagon too!!!” I am being force-fed technology that I don't want. I am paying to have my life and personal privacy streamlined into the coffers of ... I don't even know who... or where. It's a morass and I just want to unplug them all. It used to be such a great tool. Signed, not an engineer @mttaggart GenAI can't even get facts straight with regards to code where there's a clear "correct" answer and any number of incorrect answers. I recently discussed a coding problem with a coworker, who (I didn't know) fed some prompt relating to it into ChatGPT and came back with a "ChatGPT says $feature can be used for this". Knowing this not to be the case, I replied nope and a link to the documentation for $feature showing it can't. Response was "I know, but I thought maybe". ARGH! @mttaggart @da_667 I’m running it locally on my machine for coding purposes, because I don’t consider myself a coder but I need stuff on occasion. I bring it a concept of a plan and it spits out a plan, then I use it to assist with fine tuning that plan until the tests run correctly. I don’t think I can say I’ll use it less, but I have migrated to small, local, and focused usage. I’m not playing with it for fun.  @mttaggart @da_667 this almost exactly. Fisting it into everything like a toddler with superhuman strength doesn't mean it fits. It just means you've destroyed a lot of things to fist the "AI" into something that didn't need it, and no one wanted in the first place. There are finite, useful, things it can do. It's not worth the wholesale theft of artistic IP, resource costs, and the hit to every facet of data privacy everywhere.  @mttaggart @da_667 The one and only time I’ve asked an LLM about code, due to curiosity, it was so obviously rubbish I’ve never been back. Specifically I asked it to write me a function for packing and unpacking GSM 7-bit characters into 8-bit bytes and it’s answer was to concatenate the 7-bit characters directly together. |
@da_667 I would say code is where the utility is clearest, but the cost of making the tool valuable is utterly bananas compared to the value—and that's for quite the niche, to say nothing of everywhere else it's being mindlessly jammed.