Although ideally a follow up work would be something where there won’t be any potential legal trouble with releasing the complete model so people can play it.
A similar approach but with a game where the exact input is obvious and unambiguous from the graphics alone so that you can use unannotated data might work. You’d just have to create a model to create the action annotations. I’m not sure what the point would be, but it sounds like it’d be interesting.
A really goofy way I can think of to get a bunch of data would be to get videos from youtube and try to detect keyboard sounds to determine what keys they're pressing.
It is a product. It is something people will buy. Lots of people just don't care about privacy that much, and the privacy risk factors with this are potentially lower than for other things people use with roughly the same level of "realness".
In a sense GPT-4 is self-training already, in that it's bringing in money for OpenAI which is being spent on training further iterations. (this is a joke)
To summarize the analysis: you would need 100-200 square meters of farmbotting to get 2000 daily calories from farmbots alone. If you're just trying to get your daily recommended servings of veggies, you instead target volume of veggies, because that's how the recommendations work, and it's a much smaller quantity. It ends up being 3-7 square meters per person for that.
A small farmbot covers 4.5 m^2 and costs $3,000, an XL covers 18 m^2 and costs $4,500.
I think 3-7 sq m makes sense as a practical range, maybe round up to 10 to have some wiggle room.
Or 5%, right? What's the minimum amount that's "killer", on a societal level and on an industry-specific level? Human transcription might be a good case study.
No camera is interesting. On one hand, it's aesthetically strange - you would expect a computer on your face, putting screens right in front of your eyes, would be able to see what you can see. On the other hand, it improves the social experience if people don't feel like you're putting them in front of a potentially recording camera all the time.
On first blush I like the low-res, stylized display. Depends on the actual experience of it though. Seems like it might make long-form reading less pleasant.
For a similar goal, I included "Lean towards casual and conversational speech, try to avoid sounding corporate or like copywriting.", which partially worked. I'm going to try yours, it sounds fun.
It's good if you really don't want your LLM to mention specific things, which I can see some groups wanting. Having it mention some things even when they're not related could be good for integrated ads in a chatbot, which sounds evil in that it would be really annoying. Your friend's account gets hacked, a chatbot LLM is finetuned on their message history, it's able to carry on a conversation while slipping in a mention of Joe's Hot Dogs every now and then.
It probably also could help with consistency when trying to do LangChain-type stuff.
When someone says "It's a private company, they can do what they want" in defense of a company doing something, it's probably because in that case they like what the company is doing, and they're probably not being "taken for a ride" by them.
But you can go to non-mainstream subs where there's a different voting population and that doesn't happen (well, it does, but the "popular opinion" can be different within the sub).
I had the thought recently that theorem provers could be a neat source of synthetic data. Make an LLM generate a proof, run it to evaluate it and label it as valid/invalid, fine-tune the LLM on the results. In theory it should then more consistently create valid proofs.
I think that's a little harsh, at least grabby aliens theory is based on certain assumptions and logic that can be falsified, argued against, etc. It's better than Russell's teapot.
I like the map being a toggle, because I could navigate by looking at the map while I move around (helps with that confusion). The 3d view is needed to see items, pots, and enemies, and that's fine, but ideally those would show on the map too.
I don't see why something like this would work even in theory. They're scamming to make money, it doesn't sound like recruiting them to help with open-source projects aligns with that at all, there's no hook.
When I was looking for a mattress recently, I wanted something affordable, but I shied away from really cheap options, because I was afraid they'd be very low quality. Similar reasoning could apply to many things.