Thanks for posting this, a great example of forcing the human element to fit the computer rather than the other way around.
I have my own experience with a similar issue as I work with regular expressions in multiple languages. Getting the cursor to behave while mixing RtL and LtR is a nightmare, and it's always a guess whether the regex engine will evaluate the text correctly. There's also the struggle between forms that look identical but are different unicode points e.g. between Persian and Arabic, as well as the issues with Persian zero-width non-joiners (ZWNJ) which are optional in the language but can look invisible (or nearly so) to the eye.
These are genuinely difficult technical challenges and I understand that there's no quick fix. But I wish there was more of a drive to solve them, rather than the default English-over-everything approach that's so common.
Yeah I had a similar experience, even as a teen I resisted the move from myspace to facebook because it felt so bland and lifeless. Hard to overstate the psychological benefit of having a customizable page that really feels like your own space, especially at that age.
I agree with the people saying that the product is a lot better once you're actively engaging with pages that align with your interests, so that the algorithm can feed you better content.
That being said, it's still sad that this is the default new/returning user experience. Imagine a world where a new user was met with real posts about a variety of interests, rather than a psychic barrage of insane AI posts.
I think even for someone who logs in daily and uses it a bit, it still shovels weird content and even if you repeatedly skip or don't engage with AI slop, you still get a lot of it.
I almost think we are seeing something similar to a CAPTCHA where the engagement is being used to tune which videos slip under the uncanny valley radar.
>an LLM that can generate textures to be fed into a human-coded 3d engine
I'm not certain but I think the LLM is also generating the physics itself. It's generating rules based on its training data, e.g. watch a cat walk enough and you can simulate how the cat moves in the generated "world".
reply