Found this on my feed
@limotux
This is new! But what do you think about āoffensiveā responses? I do not say yes it is OK, but I see it as āmore humanā (unfortunately). This is new! But what do you think about āoffensiveā responses? I do not say yes it is OK, but I see it as āmore humanā (unfortunately).
ColossalAI only have a basic filter to filter out offensive responses like NSFW or potentially discriminative responses/queries, but itās imperfect and sometimes either block non-offensive questions or responses or leave some of them out; but at least you are sure that it wonāt likely have controversial practices of letting Kenyan people to filter out offensive content with extremely low wage and without any mental health insurance
They said that because of limited resources they only offer their service for limited time, but their chat UI will be open source soon.
I see! Now you clarified it. It is all about imperfect human work that leads to this not a perfect human work that made AI really act as humans do!
I really appreciate clarifying it for me. Thank you @natsumeshokogami
Private gpt looks interesting. And itās cool that it can run on 1 gpu.
Does it mean that auto-gpt is chat-gpt with internet access?
I just saw a presentation about chat gpt4. They have really dumbed it down. They say itās for safety reasons. I donāt understand why it is more secure because it is less advanced.
On a side-note:
Microsoft says it has developed a single system for OpenAI, the company behind ChatGPT, that is equivalent to a supercomputer: 285,000 CPU cores, 10,000 GPUs and immense network bandwidth. The cost of training large language models in the older GPT-3 language model alone is estimated to be a few million dollars. Current models are even more complex and thus likely to be even more expensive.
Even possible open-source AI models, for which there is apparently growing internal respect at Google in particular, will not be able to break up this concentration on a few players without further ado. After all, the use of complex models always faces a certain entry hurdle in terms of the availability of hardware and data volumes, albeit perhaps to varying degrees.
In short, as magical as AI tools may seem, the associated concentration of power and dependence on a few tech companies that are even capable of developing or operating these resource-intensive systems is problematic.
yes itās chatgpt with internet access, but you can disable itās access, but thatās kinda pointless imoā¦
what I saw about gpt4 looked like itās a lot better with problem solving, maybe more limited in some aspects idk, but it produces better code from what Iāve seen.
Anyone try out this one yet? Alpaca-Lora:
Scalable Personal AI: You can finetune a personalized AI on your laptop in an evening.
ok, it looks good. But you have to use the paid version of the api key. Do you think it is worth paying for the api key?
Does this mean that it is not worthwhile to create an open source Ai project, unless there are huge amounts of sponsor money?
Not at all, though one might come to think so at first glanceā¦
The article talks about political and economical concentration of power in the hands of a few. AI and ChatGPT is not quite an āiPhone-momentā that the media are trying to make of it by hailing it.
Google, Meta and M$ are the big players, but at least someone with Google is beginning to take OpenSource AI projects serious.
chatGPT surely is convenient and I am still using the free GPT3 model. The paid model GPT4 is a lot better (I have access to the paid version at work):
Source? I am genuinely curious.
Needs testing, like I said itās not expensive, ~50 cent an hour. Iām waiting for my gpt4 api access until I will play with it again, some people already have acces, they wrote on github that the gpt4 version is a lot better in problem solving and programming.
But idk if you need this thing to read a book, it can. If you want to compile information about the top 50 twitter accounts it will do it. Some people had luck making simple websites, it even installed the web server for them.
Is it gpt4 api access or just that chat interface?
just the chat interface but that is already extremely useful
Anyone tried Anthropic? I have requested access, not sure if I will get access anytime soon.
I donāt have any official sources, other than I have a friend who works at one of these telecom companies who told me.
I think there are sources all over the internet. Itās no secret, everyone does it.
Under armor bought Endomondo for 500 million dollars. For well under 500 million dollars, Under armor can make an app themselves and market it. But the reason they chose to pay 500 mil for it was because of all the data Endomondo had mined on its users.
Data and private life information are gold. This is why facebook, google, microsoft, amazon etc have become so powerful.
My search skills arenāt the best, I found that ISPs collect and sell data but not how much revenue they make from data export.
ok i will pay for an api and test it.
Iāll be home in two hours and start installing it.
When I ask chat gpt if it can recommend a certain brand of Chinese parking heater. Then it says that it cannot recommend a specific brand as it is an Ai.
I think itās a shame that they dumbed it down like that.