@rommsey@elonmusk@cb_doge The note is still accurate. Open source is often used to refer to open weight models anyway Whether Elon's comment is right depends on interpretation and should not be taken as right or wrong on its own - that's over simplifying and creates unnecessary tension
@dragonstxne When I ask about expressiveness, I mean how is he original black and white clip more expressive if it has no facial expressions? I guess Flandre's smile is not as distinct here (kind of) compared to the original but is there more to it?
@dragonstxne I mean, it doesn't look too bad to me - what do you mean by "expressiveness"? I'm not artistic so it would help if you could explain a bit
@TechDaddyKB I'm under the impression that complaining gives them the satisfaction of being right, and happiness is derived that way somehow. I don't think thats good though as the reaction can lead to people just... not doing things because it brings hate.
@pabis_eu@blueemi99 Yeah, but just because Chinese labs publish their techniques doesn't mean that the labs have better capabilities to train good models in general. Their research has value, but they are not quite there it seems
@cloneofsimo I'm pretty sure if they tried, the labs would try to get their models to use editors and animation software for humans instead of going the diffusion or world model route
@P0chu_L1ver@ancientrunekrp@GoldenJoysticks@house_house_@PotNoodle It is generative - the AI generates new code for the game. It's feels less egregious than directly copying an art style, but AI also learns (and arguably steals) human code. I'm fine with the use of AI here but calling it not generative is inaccurate.
@zylumimaxx The premise is infeasible right now. Current models need large amounts of data for their outputs to make some sense at all. If the premise holds then it probably could be ethical, but I don't see how it could be true in the foreseeable future.
I am sick of people saying Big Walk's bug fix on code isn't generative AI. It is, it generates code, just not art directly. I'm fine with Big Walk using it for a bug fix but it's dishonest to say it's not generative AI. Coders are still technically getting ripped off.
@jfwong Wait, I thought LLMs were considered generative AI, not just diffusion models for art? I think it's fine to use AI for software engineering but calling it not gen AI makes it easier for people to dismiss the argument
@forloopcodes This song already existed as Liar Macaron, I'm pretty sure this is not a demonstration of AI capabilities but satire for AI copying other work, and it's also somewhat ironic because Liar Macaron itself was accused of having AI generated music
@peach2k2 Is this a joke I'm not getting because didn't some one make this song already and it already was controversial for supposedly using AI even before this?
@yoheinakajima Ok I thought this was constraint decoding by changing logits/sampling (confusing name), but predicting using token positions is what people used to do for NLP before ChatGPT. Here is an example: https://t.co/9x7o6NKgqp
It's not as general as modern LLMs but it's not that new
@noahreeveshq If you asked them about projects that they handed in for school work they might remember some projects. If they never did any project through school then they might not be a suitable fit for the company, although it might be harsh to say they work for the money only
@C4V3R47@ylecun@Noahpinion AGI can mean that machines can do everything a normal human can do. A human can interact with the physical world with their body, hence AGI should be able to control a physical body. Whether "everything a normal human does" includes body control depends on who you ask.
@dwlz@itsdistribution If you agree with this, then unit tests aren't useless? They are for us devs to build confidence in the code LLMs write, especially if our requirements aren't that specific at first, right?
@khoiiiind I believe it's because writing specifications in the past is too expensive and hard to integrate. It is easier now thanks to Lean, but it will still be tricky for popular languages like Python, C++ and Java since there aren't complete formal models for them yet