@@1littlecoder I found the Sally question confused a number of LLMs including Mistral. However I got it to answer the question correctly by rephrasing it to: "Sally (a girl) has 3 brothers. Each brother has 2 sisters; And Sally is one of the 2 sisters. How many sisters does Sally have?"
@@1littlecoder CORRECTION! I meant to say, I got other LLMs not DeepSeek that were answering incorrectly to answer correctly by rephrasing the question.☝🏾 DeepSeek still answered incorrectly...
@@1littlecoder Mistral - mistral:7b-instruct-q5_K_M to be exactly. "Sally (a girl) has 3 brothers. Each brother has 2 sisters; And Sally is one of the 2 sisters. How many sisters does Sally have?" Sally has 1 sister. Here's the reasoning behind this: 1. We know that Sally is one of the 2 sisters. 2. Since she has 3 brothers, we can infer that her other sister must be married to one of those brothers (as it says each brother has 2 sisters). 3. Therefore, Sally's only remaining sibling is her sister who is married to her brother. 4. So, Sally has 1 sister.
Very nice. Regarding licenses, I assert that models are the output of algorithms; therefore, they lack human authorship and are not granted copyright protections under most national laws, including US law. The only intellectual property rights they may have are those of trade secrets, which must be kept confidential to maintain such protections (e.g., Claude, GPT-3+). Some models attempt to "license wall" access, requiring agreement to the license terms for access. While the enforceability of such a license for content lacking intellectual property rights is questionable for the initial user who agrees to the terms, it is clearly unenforceable for subsequent users who acquire it through re-uploads. Licenses like the MIT license are only enforceable to the extent that they rely on underlying copyright law to protect the original content.
China is currently gaining an advantage in the field of AI; it may have the potential to surpass the US, although the probability is relatively low. Meanwhile, India has yet to make significant strides in AI!
as always, you are appreciated 👊🏾
Thank you very much!
@@1littlecoder I found the Sally question confused a number of LLMs including Mistral. However I got it to answer the question correctly by rephrasing it to: "Sally (a girl) has 3 brothers. Each brother has 2 sisters; And Sally is one of the 2 sisters. How many sisters does Sally have?"
@@TheWallReports Mixtral or Mistral ?
@@1littlecoder CORRECTION! I meant to say, I got other LLMs not DeepSeek that were answering incorrectly to answer correctly by rephrasing the question.☝🏾
DeepSeek still answered incorrectly...
@@1littlecoder Mistral - mistral:7b-instruct-q5_K_M to be exactly.
"Sally (a girl) has 3 brothers. Each brother has 2 sisters; And Sally is one of the 2 sisters. How many sisters does Sally have?"
Sally has 1 sister.
Here's the reasoning behind this:
1. We know that Sally is one of the 2 sisters.
2. Since she has 3 brothers, we can infer that her other sister must be married to one of those brothers (as it says each brother has 2
sisters).
3. Therefore, Sally's only remaining sibling is her sister who is married to her brother.
4. So, Sally has 1 sister.
So fast, hope you hit the 100k soon!
Thanks, that will take a while I guess. I hope it happens:)
Thanks for the video. We need a comparism with mistral.
Great videos. Funny how it started to give you a reply on the question about Taiwan and then eradicated it.
i love your channel keep at it you are doing great!
Very nice.
Regarding licenses, I assert that models are the output of algorithms; therefore, they lack human authorship and are not granted copyright protections under most national laws, including US law. The only intellectual property rights they may have are those of trade secrets, which must be kept confidential to maintain such protections (e.g., Claude, GPT-3+).
Some models attempt to "license wall" access, requiring agreement to the license terms for access. While the enforceability of such a license for content lacking intellectual property rights is questionable for the initial user who agrees to the terms, it is clearly unenforceable for subsequent users who acquire it through re-uploads. Licenses like the MIT license are only enforceable to the extent that they rely on underlying copyright law to protect the original content.
China is currently gaining an advantage in the field of AI; it may have the potential to surpass the US, although the probability is relatively low. Meanwhile, India has yet to make significant strides in AI!
Great video 1little! How to train it to response in base a pdf ?
Deep seek is using straight forward translation for chat in other languages.
Could you elaborate what you mean ?
That rap section just gave me life 😅😂
Fine tune the same way as llama-2-70b?
Lol, it took OpenAI years to build GPT-4 and it's already surpassed by DeepSeek. Imagine their faces a few week ago when it came out.
The documentation itself is a valuable tool - even with the model being rather underwhelming compared to the state of the art.
Does this model gives summarisation if i give set of comments
Don't even think about asking it about Tank Man or the Uighurs!
Bro can you share the link of thumbnail girl photo or the prompt to make such images
here you go - imgur.com/a/EKJWbVe
@@1littlecoder thanx bro but it says "we couldn't find this page"
Uh, I support a license from a place where copyrights and intellectual property are taken very very seriously.
Does the model have a bias to the CCP or a bias against the west?
Phind is better