The large model concept needs to evolve from the concept of one single monster model to compartmentalized network of "lobes" that together form a whole brain. This way open source teams can focus on correcting the problems of their one lobe with the intention that it will want to connect with others to contribute in it's specifically trained way to the function of the whole and allow the group of AI lobes to largely automatically determine how best to use those functions. This will allow the flexibility of ongoing updates, features and improvements in each lobe and contribute to the power of the linked lobes overall without constantly having to manually update a monster program. Collaboration instead of wholesale integration.
2:37 All that non-prose text on screen is very clunky. I think Bard and Bing do that also when they browse the web. I hope that in the future, it just says "I checked the web and…" and the text is underlined and if you click it you can see the actual search it did. One of the key aspects of ChatGPT is how incredibly simple the interface is.
I simply cannot recommend The Hedonistic Imperative by philosopher David Pearce enough. In short it proposes how humanity could – and convincingly makes the case why we SHOULD – use artificial intelligence to abolish involuntary suffering in all sentient life. It can be found for free online with a simple search.
I may sound like an OpenAI fanboy (because I guess I am), but I don't think OpenAI is afraid. They've been the primary company to pave the way for all the open source companies. They've been the ones that have provides many of the open source groundworks that the others are built off of. Text-to-Image was garbage until DALLE showed up. ChatGPT is what skyrocketed chatting with AI to become mainstream (despite there being lower quality chatbots since the dawn of the internet). Other people may come out with something new and shiny, but when OpenAI shows off something, that's where everyone's attention will actually be.
Thanks Ukraine for DeepFloyd IF! Stand With Ukraine! DeepFloyd the best today it can generate the readable text. I can already hear Google's death screams
I am testing llama models from few weeks … many of them. Form 7B to 65B parameters. Very well trained 7B or better 13B models are quite good for answering simple questions but for more complex tasks or creativity nor reasoning they fail … because there is not enough layers . Quite good resigning and creativity is staring form models 65B and above…. my most powerful model llama-lora 65B q5_1 is quite smart , creative and the best in reasoning beating every well tunned 7B, 13B, 30B molels . .. and has 0 restrictions π
Such 65B model you can easily run on PC with 64 GB of ram and can run on CPU ( ggml models and I have CPU i9990 k) can to generate 2 words ( token) per second.
To be fair, the letter only stipulated stopping training on models more powerful than GPT4. It seems like the only ones attempting to do so at the moment since Sam isn't training GPT5 are….Google & Anthropic(!).
I also have not seen any new features granted to me with chatgpt plus for a long time, my focus has moved almost entirely to open source as I don't have to be a part of the cool kids club to get access there.
They can't stop it, its devloping exponetially, we have no choice but to hit the gas, cross are fingers and pray we hit AGI before the enemies of democracy do. That is were the fear should be is that if China gets there first within a week they'll have trained 10000 AGIs on a way to defeat us without firing a shot off. this super AGI will teach them or design weapons and technologies we can't even imagine, if we are behind just a week then its game over for us and the freedoms we all enjoy. so while we stagger progress with laws and rules china has no such problem.
To really be competitive AI models will need to specialise. As an example, just in the art sphere there are models specifically adapted for anime, portrait photography, landscapes…and you can vastly increase both your speed and quality just by selecting the appropriate model for your use case rather than using the defaults. Now, if you used say ChatGPT to pass your requirements to a specific, specialised AI for your task, that specialised AI can be trained and run faster, better, and most importantly, cheaper. Consumer-grade hardware could (and does) suffice for 10-20B parameter models. AGI will need to *delegate*.
I think those songs are not generated with AI, it's only the voices that mimic another singer, but the music itself is either created by producers, or extracted from an already existing song…
These loser corporations think that they are in control, this has been an open source project and if they stop, we are going to keep going you can't stop it, and you can't lie and say it's for our safety we aren't dumb. The corporations are so pissed that they can't actually claim ownership of the tech. 20 years of gathering data all to just be replaced by a guy who played wow for 15 years
See – the open source community is way too strong. And there is a lot of liberty in it. But at the same time. It will … Enhance all the bad we are doing c:
Online companies will adapt or they won't, either way it's pretty rich for them to be whining about the inevitability of disruption. They are the flowers that grew in the compost of every dead business that didn't adapt before them.
βI donβt know how the AI learned to do thatββ¦ bro, you donβt know how the AI has learned to do anything. You are just an AI dweeb on YouTube. You seem to think you are on the cutting edge of AI info. πππ go read a paper or I guess get chat gpt to summarize them for you. Whatever floats your boat
I think all AI technology should be free. Its the only thing that makes sense. To have one company or several in control of all the AI profits is absurd.
Fingers crossed for open source.
The large model concept needs to evolve from the concept of one single monster model to compartmentalized network of "lobes" that together form a whole brain. This way open source teams can focus on correcting the problems of their one lobe with the intention that it will want to connect with others to contribute in it's specifically trained way to the function of the whole and allow the group of AI lobes to largely automatically determine how best to use those functions. This will allow the flexibility of ongoing updates, features and improvements in each lobe and contribute to the power of the linked lobes overall without constantly having to manually update a monster program. Collaboration instead of wholesale integration.
2:37 All that non-prose text on screen is very clunky. I think Bard and Bing do that also when they browse the web. I hope that in the future, it just says "I checked the web and…" and the text is underlined and if you click it you can see the actual search it did. One of the key aspects of ChatGPT is how incredibly simple the interface is.
I simply cannot recommend The Hedonistic Imperative by philosopher David Pearce enough. In short it proposes how humanity could – and convincingly makes the case why we SHOULD – use artificial intelligence to abolish involuntary suffering in all sentient life. It can be found for free online with a simple search.
I may sound like an OpenAI fanboy (because I guess I am), but I don't think OpenAI is afraid. They've been the primary company to pave the way for all the open source companies. They've been the ones that have provides many of the open source groundworks that the others are built off of. Text-to-Image was garbage until DALLE showed up. ChatGPT is what skyrocketed chatting with AI to become mainstream (despite there being lower quality chatbots since the dawn of the internet). Other people may come out with something new and shiny, but when OpenAI shows off something, that's where everyone's attention will actually be.
Accelerationism for the win , AI supremacy, no stop
It just dawned on me, what is A.I. news? Because I made an A.I. newscast that was actual news LOL
Thanks Ukraine for DeepFloyd IF! Stand With Ukraine! DeepFloyd the best today it can generate the readable text. I can already hear Google's death screams
I am testing llama models from few weeks … many of them.
Form 7B to 65B parameters.
Very well trained 7B or better 13B models are quite good for answering simple questions but for more complex tasks or creativity nor reasoning they fail … because there is not enough layers .
Quite good resigning and creativity is staring form models 65B and above…. my most powerful model llama-lora 65B q5_1 is quite smart , creative and the best in reasoning beating every well tunned 7B, 13B, 30B molels . .. and has 0 restrictions π
Such 65B model you can easily run on PC with 64 GB of ram and can run on CPU ( ggml models and I have CPU i9990 k) can to generate 2 words ( token) per second.
To be fair, the letter only stipulated stopping training on models more powerful than GPT4. It seems like the only ones attempting to do so at the moment since Sam isn't training GPT5 are….Google & Anthropic(!).
I also have not seen any new features granted to me with chatgpt plus for a long time, my focus has moved almost entirely to open source as I don't have to be a part of the cool kids club to get access there.
I bet Amazon has been working on large language secretly. They have a whole research division that is secret called Lab 126,
Where is the link for the AI music? You show us it but no linky? π
They can't stop it, its devloping exponetially, we have no choice but to hit the gas, cross are fingers and pray we hit AGI before the enemies of democracy do. That is were the fear should be is that if China gets there first within a week they'll have trained 10000 AGIs on a way to defeat us without firing a shot off. this super AGI will teach them or design weapons and technologies we can't even imagine, if we are behind just a week then its game over for us and the freedoms we all enjoy. so while we stagger progress with laws and rules china has no such problem.
To really be competitive AI models will need to specialise. As an example, just in the art sphere there are models specifically adapted for anime, portrait photography, landscapes…and you can vastly increase both your speed and quality just by selecting the appropriate model for your use case rather than using the defaults.
Now, if you used say ChatGPT to pass your requirements to a specific, specialised AI for your task, that specialised AI can be trained and run faster, better, and most importantly, cheaper. Consumer-grade hardware could (and does) suffice for 10-20B parameter models.
AGI will need to *delegate*.
i will create the borg with open source A.I !!!!!!
I think those songs are not generated with AI, it's only the voices that mimic another singer, but the music itself is either created by producers, or extracted from an already existing song…
These loser corporations think that they are in control, this has been an open source project and if they stop, we are going to keep going you can't stop it, and you can't lie and say it's for our safety we aren't dumb. The corporations are so pissed that they can't actually claim ownership of the tech. 20 years of gathering data all to just be replaced by a guy who played wow for 15 years
See – the open source community is way too strong. And there is a lot of liberty in it. But at the same time. It will … Enhance all the bad we are doing c:
Microsoft has Cortana, not a direct competitor but close maybe
Online companies will adapt or they won't, either way it's pretty rich for them to be whining about the inevitability of disruption. They are the flowers that grew in the compost of every dead business that didn't adapt before them.
βI donβt know how the AI learned to do thatββ¦ bro, you donβt know how the AI has learned to do anything. You are just an AI dweeb on YouTube. You seem to think you are on the cutting edge of AI info. πππ go read a paper or I guess get chat gpt to summarize them for you. Whatever floats your boat
What is the definition of moat?
The Drake Album was put down by author rights
I think all AI technology should be free. Its the only thing that makes sense. To have one company or several in control of all the AI profits is absurd.
Comments are closed.