It has happened to me a few times. You spend time learning something or sometimes start working on a project that uses LLMs that you find super original, and suddenly BOOM! An OpenAI update makes your project obsolete. A good example of this is what happened to plugins after the update that brought the code interpreter. I think this is the feeling of inferiority that we are going to start experiencing towards AIs. 😅
Also keep in mind that you're paying for the context window, whereas you don't have to for the vector search (at least not inference costs). @@samwitteveenai
Hey Sam, love your videos where you go around some models ! I think that one domain you don't address is being multilingual. It might be interesting to include a non English prompt in your prompt set.
This is totally true and something I want to address. I especially want to show how to train models for other languages. I have been thinking a lot about how to do it so people can understand even if they don't know. One challenge with testing just multi lingual prompts is how to judge the responses if I don't understand the language. If you have ideas please share them, I am very open to suggestions.
Hey Sam, thank you for all your hard work on these videos. Just wanted to understand why you said the Claude2 model is cheaper to use than ChatGPT. Were you referring to GPT4? Because GPT3.5 Turbo is $4per million tokens vs. $32 per million tokens on Claude 2 output. Am I missing something?
ChatGPT (3.5 Turbo) is cheaper than Claude Instant and Chat GPT 4 is cheaper than Claude 2.0. Claude 2 is neat but GPT 4.0 is still the best overall imo - BardAI is actually stronger than GPT 4... for multimodal i.e. reading images with google lens. GPT4 pretty much wins eevrything else against any AI model right now. Claude vs GPT 3.5 is tougher. GPT 3.5 is a better coder imo despite Claude 2 having coding upgrades. Claude 2 is surprisingly good at creative writing, at least for me.
Awesome video! Could you pls create a video playing around with Claude 2 API and Langchain. The OpenAI’s Function capability is very powerful, I am not sure if there are ways of replicating that with Claude.
Hi Sam. I'm working on a problem where I need to decern the difference between a quantity order that also mentions the size. E.g. 1 3/4" which could be 1+3/4" no quantity specified, vs a quantity of 1 and a size of 3/4". Are you aware of any models that could reasonable job of figuring that out or would I need to use a function with regex? Thanks.
Hi Sam, Claude 2 is running on cloud, right? Which LLM has performance on par with Claude 2 or ChatGPT 3.5 that can be run on premise due to all those customer confidentiality, privacy and security considerations?
If you have the money I think Anthropic is more open to the idea of on prem models than OpenAI. If you mean an opensource model that is as good, we may need to wait a few more weeks for LLaMA2
unfortunatly not available outside us and uk .... was looking for the api calls, but unable to find anything. Is there any documantation / langchain integration?
Could someone explain to me the difference between ggml and gptq templates? I see in the repository of TheBloke models: "TheBloke/Starcoderplus-Guanaco-GPT4-15B-V1.0-GPTQ" and "TheBloke/Starcoderplus-Guanaco-GPT4-15B-V1.0-GGML" what would be the difference between them (the second being triple the size). Grateful.
I have used Claude older version a lot. Basically, asking questions about social issues, arts, politics, humanities. Claude is far better than ChatGPT or Bard when it comes to un-biased answers.
For my needs, Claude 2 is SOOO MUCH better than ChatGPT. I'm embarrassed to say that because I can't stop telling people about ChatGPT. Now, I am a big fan of Google and I own shares. Bard SUCKS!! I hate to say that. I hope Gemini will blow the pants off of all these other models because Google needs it.
Given it a quick try uploading a PDF and it did really well. Makes me worry that I've wasted time on embeddings and VectorDB.
It has happened to me a few times. You spend time learning something or sometimes start working on a project that uses LLMs that you find super original, and suddenly BOOM! An OpenAI update makes your project obsolete. A good example of this is what happened to plugins after the update that brought the code interpreter. I think this is the feeling of inferiority that we are going to start experiencing towards AIs. 😅
I can relate. It is pretty amazing having a 100k context window. I do think embeddings and vector DBs will be useful for a while though.
Also keep in mind that you're paying for the context window, whereas you don't have to for the vector search (at least not inference costs). @@samwitteveenai
I am pretty sure people will now want to compose more complex queries. The context window will never big enough in my opinion
Also the model gets worse and worse if you increase the context tokens. After 4k token the performance just drops of
Great video Sam. Love your content!
Thank you - great video. It looks like Claude 2 is a better option than free version of chatGPT3,5.
Hey Sam, love your videos where you go around some models !
I think that one domain you don't address is being multilingual. It might be interesting to include a non English prompt in your prompt set.
This is totally true and something I want to address. I especially want to show how to train models for other languages. I have been thinking a lot about how to do it so people can understand even if they don't know. One challenge with testing just multi lingual prompts is how to judge the responses if I don't understand the language. If you have ideas please share them, I am very open to suggestions.
Hey Sam, thank you for all your hard work on these videos. Just wanted to understand why you said the Claude2 model is cheaper to use than ChatGPT. Were you referring to GPT4? Because GPT3.5 Turbo is $4per million tokens vs. $32 per million tokens on Claude 2 output. Am I missing something?
Yea i have the same question. GPT3.5 is actually 2$ per mil unless you use 16k version
ChatGPT (3.5 Turbo) is cheaper than Claude Instant and Chat GPT 4 is cheaper than Claude 2.0. Claude 2 is neat but GPT 4.0 is still the best overall imo - BardAI is actually stronger than GPT 4... for multimodal i.e. reading images with google lens. GPT4 pretty much wins eevrything else against any AI model right now. Claude vs GPT 3.5 is tougher. GPT 3.5 is a better coder imo despite Claude 2 having coding upgrades. Claude 2 is surprisingly good at creative writing, at least for me.
Thanks, Sam. Do you happen to know whether it’s integrated with langchain? Would you consider doing a video on this? Thank you, as always.
I am trying to get API access to show exactly this, but it seems hard to get at the moment. I will hopefully get a solution soon.
O Brasil está esperando!
usa vpn
@@theshrubberer I will for sure, I use Opera Developer.
Awesome video! Could you pls create a video playing around with Claude 2 API and Langchain. The OpenAI’s Function capability is very powerful, I am not sure if there are ways of replicating that with Claude.
Making a LangChain + Claude-2 vid currently
Thanks for the video. Does it have code interpreter abilities?
Similar but perhaps not built for code in the same way Code Interpreter is fine tuned etc.
Hi Sam. I'm working on a problem where I need to decern the difference between a quantity order that also mentions the size. E.g. 1 3/4" which could be 1+3/4" no quantity specified, vs a quantity of 1 and a size of 3/4". Are you aware of any models that could reasonable job of figuring that out or would I need to use a function with regex? Thanks.
Hi Sam, Claude 2 is running on cloud, right? Which LLM has performance on par with Claude 2 or ChatGPT 3.5 that can be run on premise due to all those customer confidentiality, privacy and security considerations?
If you have the money I think Anthropic is more open to the idea of on prem models than OpenAI. If you mean an opensource model that is as good, we may need to wait a few more weeks for LLaMA2
Where do I find information about running Claude on prem? I couldn’t find it in their website
I think it is best to reach out to their sales people.
unfortunatly not available outside us and uk .... was looking for the api calls, but unable to find anything. Is there any documantation / langchain integration?
API access is hard to get, but for testing a decent VPN is your friend here. also can check out POE.com
How well does it do if a PDF contains graphs/charts? Do Anthropic work on bi-modal (text+image to text) models?
I don't think it is supporting any images just yet. You can click on the PDF after it is uploaded to see what the LLM is getting etc
I attached a CSV file and it said it cannot directly read the contents and wanted me to describe the contents ...not sure what is going on
my guess is this could be because your csv has no header?
Could someone explain to me the difference between ggml and gptq templates? I see in the repository of TheBloke models: "TheBloke/Starcoderplus-Guanaco-GPT4-15B-V1.0-GPTQ" and "TheBloke/Starcoderplus-Guanaco-GPT4-15B-V1.0-GGML" what would be the difference between them (the second being triple the size). Grateful.
if i'm not mistaken the GPTQ allows for 8k context window
They are different types of 4bit quantization
@@samwitteveenai thanks
@@MrAmack2u thanks
@@MrAmack2u thanks
how many tokens in output ?
I attach pdf files but it says it cannot read or summarize
try again it should be working ok
Does this model accept images as inputs?
Not yet
Thanks
I have used Claude older version a lot.
Basically, asking questions about social issues, arts, politics, humanities.
Claude is far better than ChatGPT or Bard when it comes to un-biased answers.
I have been using Claude to help me write essays and I found it to top notch 👍
Just another closed-source censored model
Only available in the US and UK 👎 - 1 Billion in funds but still not available worldwide... Coma on 🙄
Agree, but VPN is your friend here. no more to be said :D
For my needs, Claude 2 is SOOO MUCH better than ChatGPT. I'm embarrassed to say that because I can't stop telling people about ChatGPT. Now, I am a big fan of Google and I own shares. Bard SUCKS!! I hate to say that. I hope Gemini will blow the pants off of all these other models because Google needs it.