Its for purchasing AI summaries. By default each user gets 1000 article summaries to begin with, which should suffice for several months for most users.
Initially I didn't want to put any in-app purchases but decided to add it to prevent abuse.
These datasets would definitely have a lot of Text => Sketch pairs as well. I wonder if its possible to extrapolate from Text => Sketch and Text => Image pairs to improve Sketch => Image capabilities. The models must be doing some notion of it already.
That is an interesting observation. I wonder what percentage of the training text data for LLMs contains proper dashes, since a large part of it is user-generated content.
People are coming to ChatGPT with an intent to type. Its much harder to introduce text-based interface to otherwise click-based interfaces because of the friction. Thats why chatbots often fail to gain adoption, because they are a byproduct of the product team being lazy and not thinking through the UX.
Thats why I go for the nuclear option (the one that requires higest attention, like account cancellation) when presented with a list of options by a bot. More likely to get a person that way.
Q4 comes out to be ~26GB but Apple doesn't let you load it on a 32GB Mac machine because they put a limit on the max usable unified memory at ~21GB (`device. recommendedMaxWorkingSetSize`) [1]. So for Q4 Mixtral MoE you'd need a 64GB Mac machine unfortunately.
Initially I didn't want to put any in-app purchases but decided to add it to prevent abuse.