I tried garlic bread prompting with every AI. I think Meta was funniest. It responded with more garlicky images of garlic bread until GARLIC BREAD AAAAAAAAAHHHHHH at which point it gave me a scary vampire eating garlic bread. Given allergy information in vampires, the world model is incomplete!
"If I was paying by the token for my AI queries, and you offered me the option of a 100x cost increase that returned superior answers at identical speed, I would use the 100x costlier option for most purposes even if the gains were not so large."
This is of course a specific statement about where we are today. If the gains were not so large, you're probably willing to pay 100x inference costs for slightly better results, but you're probably not willing to pay (100^k)x for (slightly better)^k results for k much bigger than 1?
There's been some discussion at lesswrong as to what extent the labs made actual commitments (even informally) to present their models to the UK, or whether this was mostly media spin based on vague expressions of intention. It would be good to clarify this.
> Machine translation is good but not as good as human translation, not yet, once again: Anime attempts to use AI translation from Mantra, gets called out because it is so much worse than the fan translation, so they hired the fan translators instead. The problem with potentially ‘good enough’ automatic translation technology, like any inferior good, is that if available one is tempted to use it as a substitute. Whether or not a given executive understands this, translation of such media needs to be bespoke, or the media loses much of its value. The question is, how often do people want it enough to not care?
Something to note here: I've done machine translations on JP novels with March 2023 GPT-4 and it seemed pretty good compared to human translations back then. I'd say about 10% of it was off and would need some human corrections but 90% was about the same. Honestly content was fine, the major pain point was proper nouns. But I also tackled that by storing a dictionary to replace proper nouns with in my script.
I think the problem is a lot of fan translators aren't using the right tools.
If you look here, even 7 months ago, most people there are using Google translate or GPT-3.5 which are much worse. Honestly it's a bit irritating to see, why do so many of them still use google translate, which is awful compared to even 3.5? I would think the weeb community is at least a bit tech savvy, but they seem less tech savvy than your average high schooler cheating on homework.
I have noticed a lot of work's fan TL's paces pick up considerably in the last few months, so I would not be surprised if they are just editing ChatGPT/Cluade translated content.
Unpopular opinion time: all the people who got their fee-fees hurt by the Apple ad need to tone down the melodrama. I have never purchased an Apple product and never will, but the outage here is absurd. (Let alone the usual hypocrisy of expressing dislike of tech via the Internet.)
These Google AI overviews are really bad. Yesterday it told me confidently that a public figure had died in a motorcycle crash. (Narrator: they are still alive.)
I tried garlic bread prompting with every AI. I think Meta was funniest. It responded with more garlicky images of garlic bread until GARLIC BREAD AAAAAAAAAHHHHHH at which point it gave me a scary vampire eating garlic bread. Given allergy information in vampires, the world model is incomplete!
I've produced a muti-voiced ElevenLabs quality AI narration episode for this post:
https://askwhocastsai.substack.com/p/ai-63-introducing-alpha-fold-3-by
"If I was paying by the token for my AI queries, and you offered me the option of a 100x cost increase that returned superior answers at identical speed, I would use the 100x costlier option for most purposes even if the gains were not so large."
This is of course a specific statement about where we are today. If the gains were not so large, you're probably willing to pay 100x inference costs for slightly better results, but you're probably not willing to pay (100^k)x for (slightly better)^k results for k much bigger than 1?
There's been some discussion at lesswrong as to what extent the labs made actual commitments (even informally) to present their models to the UK, or whether this was mostly media spin based on vague expressions of intention. It would be good to clarify this.
> Machine translation is good but not as good as human translation, not yet, once again: Anime attempts to use AI translation from Mantra, gets called out because it is so much worse than the fan translation, so they hired the fan translators instead. The problem with potentially ‘good enough’ automatic translation technology, like any inferior good, is that if available one is tempted to use it as a substitute. Whether or not a given executive understands this, translation of such media needs to be bespoke, or the media loses much of its value. The question is, how often do people want it enough to not care?
Something to note here: I've done machine translations on JP novels with March 2023 GPT-4 and it seemed pretty good compared to human translations back then. I'd say about 10% of it was off and would need some human corrections but 90% was about the same. Honestly content was fine, the major pain point was proper nouns. But I also tackled that by storing a dictionary to replace proper nouns with in my script.
I think the problem is a lot of fan translators aren't using the right tools.
https://www.reddit.com/r/noveltranslations/comments/17cee84/how_does_the_quality_of_machine_translation_for/
If you look here, even 7 months ago, most people there are using Google translate or GPT-3.5 which are much worse. Honestly it's a bit irritating to see, why do so many of them still use google translate, which is awful compared to even 3.5? I would think the weeb community is at least a bit tech savvy, but they seem less tech savvy than your average high schooler cheating on homework.
I have noticed a lot of work's fan TL's paces pick up considerably in the last few months, so I would not be surprised if they are just editing ChatGPT/Cluade translated content.
Unpopular opinion time: all the people who got their fee-fees hurt by the Apple ad need to tone down the melodrama. I have never purchased an Apple product and never will, but the outage here is absurd. (Let alone the usual hypocrisy of expressing dislike of tech via the Internet.)
These Google AI overviews are really bad. Yesterday it told me confidently that a public figure had died in a motorcycle crash. (Narrator: they are still alive.)