3 ms·

> The best I can hope for is that some hacker collective manages to make an open source version of it, that can be more trustworthy than the current one. It's

> The best I can hope for is that some hacker collective manages to make an open source version of it, that can be more trustworthy than the current one. It's always interesting to me when someone makes an assert like this for a Big Data technology. If this were tractable, we'd have an open-source Google alternative right now that someone would have built for the sheer joy of being the folks that took on Google. But open source doesn't work that way because code is download-once, use-forever, but data is continuously changing and costs perpetual money to update and maintain. "Open source data" looks like Wikipedia, and the world won't sustain more than a few of those; Wikipedia has about 100,000 active editors. So instead of some hacker-alternative-to-Google techno-utopia idea, we've got plenty of open-source crawlers and a handful of services paying the bills via rent-seeking their database and, often, advertising. No reason to think a ChatGPT-heavy future will be different.

munificent 4y ago
> because code is download-once, use-forever, but data is continuously changing and costs perpetual money to update and maintain. Not just the data, but also hosting the service and keeping it available. Open source works because the marginal cost of code is essentially zero. But the marginal cost of serving users is definitely not zero.
breckenedge 4y ago
Unlike Google, you can download Wikipedia and use it offline. I hope to see the same thing happening for a useful LLM. Of course that’s not feasible right now, but hopefully those costs will come down and we do actually get to run a self-hosted version of this.
- shadowgovt 4y ago
  The tricky thing about data is that the world constantly changes. A downloaded Wikipedia has a lot of value, but it does grow stale. And it has the advantage of being a repository of relatively static facts in a way that, say, a search engine is not. Search engines (and I suspect a ChatGPT-style engine, if one wants to talk about it about current events, things currently available, or other topics of the day) have to be continuously refreshed to be relevant. So many things that those engines are used for frequently (including the keyword "ChatGPT" itself) had no definition months ago, let alone an inaccurate definition. Most data isn't static like code; it must be continuously re-invested in to stay relevant.
  - JohnFen 4y ago
    > Search engines (and I suspect a ChatGPT-style engine, if one wants to talk about it about current events, things currently available, or other topics of the day) have to be continuously refreshed to be relevant. Maybe. It depends on what you're searching for. I'd say that 80% of the searches I engage in don't need a particularly fresh database to satisfy.

munificent 4y ago

> because code is download-once, use-forever, but data is continuously changing and costs perpetual money to update and maintain. Not just the data, but also hosting the service and keeping it available. Open source works because the marginal cost of code is essentially zero. But the marginal cost of serving users is definitely not zero.

breckenedge 4y ago

Unlike Google, you can download Wikipedia and use it offline. I hope to see the same thing happening for a useful LLM. Of course that’s not feasible right now, but hopefully those costs will come down and we do actually get to run a self-hosted version of this.

shadowgovt 4y ago
The tricky thing about data is that the world constantly changes. A downloaded Wikipedia has a lot of value, but it does grow stale. And it has the advantage of being a repository of relatively static facts in a way that, say, a search engine is not. Search engines (and I suspect a ChatGPT-style engine, if one wants to talk about it about current events, things currently available, or other topics of the day) have to be continuously refreshed to be relevant. So many things that those engines are used for frequently (including the keyword "ChatGPT" itself) had no definition months ago, let alone an inaccurate definition. Most data isn't static like code; it must be continuously re-invested in to stay relevant.
- JohnFen 4y ago
  > Search engines (and I suspect a ChatGPT-style engine, if one wants to talk about it about current events, things currently available, or other topics of the day) have to be continuously refreshed to be relevant. Maybe. It depends on what you're searching for. I'd say that 80% of the searches I engage in don't need a particularly fresh database to satisfy.