Is jev another deepseek moment?Jev price literally replaces claude shit and openAI for like 90% of industrial applications.Jev won.
Whenever I go for open garbage it always sucks massive nigger dick. Just pay the sub broke ass nigga
ALRIGHT what's going on guys it's Jev.
>>109855112Deepseek moment wasn't really a moment, it was and is substantially behind the frontier. And the more I read up on Jev the less impressed I become (realizing it is a specialized classification tool rather than a leap ahead)
>>109855117They really don’t put effort into shilling anymore
>>109855137>Leaps in efficiency and cost aren't importantok fag.
>>109855112what it reveals is that most applications where LLMs can be deployed to actually automate something without being babysat by a human can be done just as well with a general classifier
use case for classification tasks?
>>109855177trading crypto.
>>109855177Sorting emails. 5 trillion dollar market cap.
>>109855112the fuck is jev
>>109855335like a LLM but it only outputs decision calls.
>>109855335jew - v
>>109855112I'll wait for the chinese to implement it, i don't use judeo-american AI
>>109855177u can buy airplane tickets saar. and reserve the hotel bedrooms. premium tier can suggest spelling and grammar corrections to emails
>>109855112>>109855335LLM made by Indians for Indians
>>109855137Deepseek forced frontier LLM companies to release reasoning models to the users for free
>>109855177Probably is efficient in making tons of decisions updating a codebase
>>109855335Jewish Electronic Vampire
>>109855335this video should help:https://www.youtube.com/watch?v=2zdlzxC_9iM
>>109855112>ToS prohibits benchmarkinghmmmm very curious why that would be the case
>>109855112Again
>>109855112Can someone explain jew in very simple terms like im 7yo? And if its good why these big companies aren't just embedding it in their models already
>>109855112You should ask your preferred LLM what jev really is.
>>109855177It sits at the top of the router and determines if an agent/task should be performed by an autoregressive LLM or classifier in order to optimize inference costs.
>>109858230They are https://deepmind.google/models/gemma/diffusiongemma/
>>109858230Jev is very fast and cheap (*according to the devs*), but it has very limited answering capabilities.Simple explanation would be that it can only answer yes/no questions or something like "is the sky blue, red or green? and then give a % on how confident it is on the answer.This can also be something like "Is this email I received an advert, yes/no?". Then from here we can step into automation, such automatically deleting emails that Jev thinks are ads or maybe ones that it thinks are ads with a certain confidence %. Obviously this is just a very simple example, but should get the idea across. All LLMs can do this too, but if Jev and other similar models are much faster and cheaper, it's better to use them for these tasks where more complex outputs from the model are not needed.Going even deeper with this idea, we can have an AI agent where Jev-like AI handles simple tasks and only calls for the bigger LLM when it's actually needed.You'll absolutely see more models like this very soon.
>>109855112Can you download and run it on your own hardware?If not, no, it's not another deepseek moment.
>>109858661Yes, it's ultra lightweight, that's the whole point. People even released CPU-only variants.
>>109858635It seems there are already experiments with using a big LLM as the orchestrator and Jev as its hands of sorts.
>>109858661I agree with this anon.I only want FOSS, offline-usable models. I will not pay for AI out my pocket, only others' pocket (e.g employers).>>109858946Is there a Hugging Face download of this Jev? I see Typesafe AI, but nothing specifically about "Jev".
>>109858985The real Jev is closed I think, but others have cooked up such decision models.The concept itself is not super revolutionary, we already had stuff like rerank models.
isnt the actual application in this case automation (say robotics) and not anything ITT is being talked about?just imagine if you have a model that is cheap, fast that you can just give robot arms and you give it a task and it does the thing you ask from itthats the concept here
>>109859119arent VLA models basically that?
>>109859119It's more for automating deskwork type tasks, although people have used it to make decisions for simple videogame inputs for example, that could be good for literal remote controlling.
starting programstarting main29524770 Females only. Flashing or accidentally showing pussy or nips0.9529526901 New0.0629529043 We all know that shaving is just unnatural and wrong.<br><br>Hairy underarms, bushes, hairy nipples, snail trails, leg hair, wild or neat... Anything goes so long as it's body hair!0.3629532349 Gay shit or femdom. Doesn't matter0.8929533701 Hotwives having fun0.729535054 natural is preferable, but I can appreciate a pair of great fakes.<br>no gross plastic bimbos or BBW.0.46seems accurate
>>109859184>>109859142the reason you dont see it being used like that yet is because it has no image input as for now so what VLA usually does gets kinda hacky with thisI've seen the videos of people hooking it up with video games and simulations and yeah it looks pretty promising. In real life you would just need good inputs it can use and I am sure that can be achieved.
but jev isnt a diffusion model, right? Some retard compared it to gemmy text diffusion (superior)
>>109859233and since the problem with VLA basically dumbs down to them being expensive and slow...Jev can change this easily
>>109855177you can classify if it's an use case or not
I can see this being used for automating moderation.
