OpenAI Needs YOU!!

Key Takeaways

okay so a few weeks ago samman basically put up a tweet asking people if open AI was to do a open- source project would it be more useful for it to be an 03 mini kind of model or a phone siiz model and you can see the votes that came out of this were pretty split with 03 mini basically winning out slightly now you can certainly see when you look at the comments here people especially companies have a very vested interest in here so you've got Aaron Levy who's the founder and CEO of box basically saying that don't trust the votes people don't know what they're talking about that people don't want local models that they want really big models that people want basically the best open-source model possible and you could imagine that his company probably has not a huge amount of problem running a 70b model or even a 400b model but looking at some of the other replies to this people seem to be landing the way that I would why don't you do both Jimmy apples obviously is someone on Twitter who's released a bunch of leaks about open Ai and then we've got other people that are well known on Twitter also chiming in to ask for both now this leads us up to earlier to today and this is what this video is all about is that I know that many of you want to basically use open models and perhaps run them on your computer or run them somewhere locally where you don't have to put your data into the cloud you don't have to share things you can basically make the model uncensored do all these kinds of things and I feel like now here's the chance to actually let open AI know this so Sam opin tweeted out that we're excited to release a powerful new open weight model with reasoning in the coming months and we want to talk to devs about how to make it maximally useful so after that we basically get them acknowledging that they haven't released an open weight llm since gpt2 now I know for most of you watching the video you probably don't even remember gpt2 gpt2 was way back in February 2019 and even then open aai basically declared that the model was so dangerous at 1.5 billion parameters for the biggest one that they were going to do a structured release so they released the smallest version which I think was a few hundred million parameters basically working up to something like a 700 million parameter 1 long before they released the full 1.5 billion parameter 1 and actually they were kind of forced into that because Richard soer and his team at Salesforce at the time released a really cool model called control which was probably deliberately 1.6 billion parameters just to show that okay maybe these models weren't as dangerous as open AI had tried to make out and in hindsight it certainly looks back at that time where they were clearly wrong and perhaps it was a marketing move and a move to help them raise funds as much as it was anything about any real danger so fast forward to today they're basically saying that okay they're going to release this model they've still got some decisions to make and they want to hold these developer events so while I'm somewhat suspicious about this and I really wonder okay and how is this actually going to go down how much are they going to listen to people when people say they want the best model at a certain size or something like that I do think that now is the opportunity for us as a community to be really vocal and talk about okay if open AI is saying they're going to do this and that they're really genuine about doing this let's let them know about what we want in the model for me personally I think the kind of model that I would love to see them do is an omni model last week I did a video about the quen 2.5 Omni model in there and that thing is fantastic that can basically do both text in and out audio in and out and then video in for processing either as text out or audio out that would be something that would be truly impressive for open AI to release now my guess is that they're probably not going to do that because they're probably going to decide that's too close to what they've already got with perhaps the 40 models or something like that but I really think that now is the time for us to basically put the pressure on them Tweet back of exactly what it is that we want go along to these sessions and if you want to sign up for the sessions you can basically come to their website sign up let them know what it is is that you want what models you've used in the past what are your areas of expertise Etc and I would say even if you know that they're not going to come to your hometown it's still worth putting in there about what it is that you want for example if you also like me want an omn model let's try and push them to do that like I said I think there's a small chance of that but I do think this is going to be a huge deal as to whether they listen to the community and actually give the community what they want and then they're going to come out looking really good or they just go off and make something that's like a dumb down reasoning model that perhaps is not that much better than anything that we've already got out there and then there's going to be a backlash so really is up to open AI now to either sort of put up or shut up in regards to open source it's very interesting that you we see with their new voice ASR stuff that it's seeming to be a lot better than the whisper models that released and yet it seems that they're clearly not going to make that stuff open source anytime soon so the other thing that you can always do is actually tweet at the people involved here you've got Kevin the product manager from openai you've got Greg Brockman one of the co-founders of open AI let them know what it is that you want to see in this open llm and I would say don't be afraid to specify that we would like a base model we would like something that's fine-tuned perhaps to do reasoning something that can clearly do function calling and some of the other bells and whistles that are out there because I'm very confident if the community response isn't overwhelming then they're probably not going to go ahead by making the best model that is out there Etc so looking at Sam multan's most recent post he's saying that they will not do anything silly like saying that you can't use the model if you have more than 700 million active users personally I couldn't care less about that clearly that's a jab at are basically trying to stop Tik Tok from using their models trying to stop Google from using their models there are not many companies in the world that have over 700 million active users and I would much prefer that they put out the best model that they could make and then limit it to who can actually use it that only startups or companies below 100 million users can actually use it then basically put out something that everyone can use but it's clearly not the best thing that they could have done anyway I've put the links for you to do the feedback in the description I do think this is something worth all of our time to basically let open AI know what it actually is that we would like to see from an open model or even a family of open models I would not be surprised if they go on to release multiple open models with perhaps a mobile version and a reasonably small version anyway as I said all the links are in the description please let me know what you think in the comments Etc and I will talk to you in the the next video bye for now