|
|
|
|
|
|
by sircastor
901 days ago
|
|
|
> I mean it's like turning down free work... Why? When you're in a competitive industry, you want to commodify the spaces where your competitors have an advantage, and keep proprietary the spaces where you have the advantage. A good example is Google's early embrace of XMPP. They made their chat system use the open standard and then when they had a strong base, they started to build proprietary things on top of the standard. Then they were diverged enough and it was their way or the highway. It was a page out of Microsoft's book. I think Google really could have owned modern chat if they'd stayed-the-course with one of the chat systems they announced. Open source is a great philosophy, but whenever you adopt it you're (potentially) funding and doing development work for your competitor. In theory those competitors are going to also contribute, but it's often assumed that competitors are fair players. That you'll both be trying to make the best thing. Your competitor might simply take your work, and undercut you on price. You're left holding the bill for all the work you put in, and they make a profit (Smaller than you would have, but their cost was negliable) |
|
|
So I guess I should be clear, I'm thinking source available. I know Open Source is a loaded word.
But doesn't this kinda support my thinking? They accelerated by using the open tools and then slowed once they had a more mature product? I mean we've seen huge advancements in LLaMA and StableDiffusion because of their open source-ness. One could argue that GPT is also getting major benefits from this, but I think also partially because it's the most common LLM to perform analysis on (which is a bit weird that researchers do this on a closed source/data model...)
I definitely get the argument around your competitor being able to snag your work, but if we're talking about source available then that becomes a legal matter. And truth be told, we could say the same about a lot of hardware. I mean anything hardware you can visibly inspect.
I definitely don't know how that all plays out in literal numbers and I do wonder if there is some counterfactual data out there. I'm just highly suspicious that it isn't as big of a cost as many assume it is. Even software you can do a really good job at cloning if you have access to it (even if completely black box). There's definitely costs, my question is just how much this compares to benefits. And is source available cost being confused with first mover cost? Because that has the same implication as you note in your conclusion.