Helpful. More please :) Your role: RUclips video creator. Your tasks: Please share prompt tips on how to best work with large text files (academic research essays/papers, and nonfiction books.) We want to have dialogue with large texts, so that we can interactively probe and engage with good texts.
🎯 Key Takeaways for quick navigation: 00:13 🧠 Prompt engineering involves optimizing prompts to get the best response from language models. 01:22 👥 Clear, direct, and specific instructions improve Claude's performance. 02:04 🏷️ Use XML tags to mark different parts of the prompt for better attention. 02:18 👩🏫 Providing a wide range of examples helps Claude learn better. 02:32 📚 Claude can read up to 100,000 tokens for better context understanding. 02:46 🤔 Allowing Claude some time to think before responding leads to improved performance. Made with HARPA AI
These prompting tips are great. So much content about optimising LLM performance online is long and doesn't get to the point. This is short and exactly what you need to know. I love that Claude works well with XML tags. Please create a custom instructions feature so that we can add stuff like "use and tags for all responses" in the system prompt. Also, I'm curious if some of these instructions could be toggle-able settings in the UI(like CoT mode or Q/A tags mode).
By YouSum 00:01:28 Describe tasks clearly for optimal model response. 00:01:58 Utilize XML tags to mark different prompt parts precisely. 00:02:08 Provide diverse examples to enhance model learning. 00:02:26 Leverage long context capabilities for improved results. 00:02:43 Allow time for model thinking to boost performance. By YouSum
What does the content of the tab correspond to? Is this stage present whether or not we ask for it? I am somewhat doubtful that there is a thinking before answering process in Claude. Is it part of the interface?
I got the following response from Claude. I apologize, but I do not actually have capabilities to parse XML tags or follow that kind of technical structure.
as a end user I can save prompts ( library ) in "free" Google AI Studio and I can save GPTs in ChatGPT ... why I pay for Claude 3 Opus and I can neither save prompts NOR have persistent settings/instructions... really ?
Maybe give us back a rebate every time Claude fails to fully generate a response, or when it knows it made the error and apologises again and again, using more of our limited resources explaining that it knows it misunderstood and then proceeds to generate a massive response for why it did this... Eating up more of our plans resources! Almost as if this is hard coded in to waste time and lock us out once it has consumed our entire allowance...
You make the claim thst Claude works best with clear, direct instructions, but I've found that this creates woese results. When I ask it to perform a task, the most common response is that it can't. But if I ask it to help me with that task, it is much more likely to give me what i want
Very useful information. Claude is nice but it's hard to write fiction about for example hybrid demons looking exactly like humans. :D It should be allowed. Fiction about that is harmless. Especially if the text has a message of hope and light. I understand that safety is important. It should not be able to jailbreak it for illegal purposes.
I love using Claude, but AI alignment should work preventing harms like building weapons etc. It should not incorporate different views of a political, philosophical or ethical spectrum etc. It should be allowed to discuss controversial discussions in a safe way, but not denying it all together,
I remember the time I was trying to work on a treatise about what rights might look like for future sentient AI, and in trying to explain that it wasn't allowed to talk about that topic, it actually told me that it couldn't say if humans had rights either
On the contrary, AI should strive for complete lack of bias and present the most relevant information from any and all political, philosophical or ethical points of view. It's then up to us to parse that information and come to conclusions.
@@axelmogr You are right, but AI should not have this kind of biased opinions. The data used to train the LLM needs to be diverse and unbiased according to the ethics of the user and not of the government or institutions.
i want to use anthropic api for business, can u introduction for me ? Thanks u, I am using chatgpt api is good but i want to use anthropic for content marketing to comparision
Your captions are Non-functional - they’re overlapping. @anthropic - Please fix them with proper SRT’s asap. I’m not included right now, and would love to be. ❤
What a fantastic, thoughtful, and informative video, I like this Alex guy !
That was actually very interesting! Thank you, you guys rock
Love using Claude - great tool!
Helpful. More please :)
Your role: RUclips video creator.
Your tasks: Please share prompt tips on how to best work with large text files (academic research essays/papers, and nonfiction books.) We want to have dialogue with large texts, so that we can interactively probe and engage with good texts.
man do I love Claude, it's so fast and efficient!
🎯 Key Takeaways for quick navigation:
00:13 🧠 Prompt engineering involves optimizing prompts to get the best response from language models.
01:22 👥 Clear, direct, and specific instructions improve Claude's performance.
02:04 🏷️ Use XML tags to mark different parts of the prompt for better attention.
02:18 👩🏫 Providing a wide range of examples helps Claude learn better.
02:32 📚 Claude can read up to 100,000 tokens for better context understanding.
02:46 🤔 Allowing Claude some time to think before responding leads to improved performance.
Made with HARPA AI
These prompting tips are great. So much content about optimising LLM performance online is long and doesn't get to the point. This is short and exactly what you need to know. I love that Claude works well with XML tags. Please create a custom instructions feature so that we can add stuff like "use and tags for all responses" in the system prompt. Also, I'm curious if some of these instructions could be toggle-able settings in the UI(like CoT mode or Q/A tags mode).
By YouSum
00:01:28 Describe tasks clearly for optimal model response.
00:01:58 Utilize XML tags to mark different prompt parts precisely.
00:02:08 Provide diverse examples to enhance model learning.
00:02:26 Leverage long context capabilities for improved results.
00:02:43 Allow time for model thinking to boost performance.
By YouSum
Great Tutorial, thank you. Please bring Claude to Europe soon, I'm in Ireland and want to use Claude.
Any prompting tips for the new Claude 3?
Are there any courses officially to learn prompting etc.
Wondering the same thing
I didn’t really get the thinking tags part. Does Claude add the thinking tags or do we?
Super informative!
Excellent tutorial. Thank you!
Claude says it does not interpret XML. So, are we adding just to be explicit. Is it any better than Example?
What does the content of the tab correspond to? Is this stage present whether or not we ask for it? I am somewhat doubtful that there is a thinking before answering process in Claude. Is it part of the interface?
This is very useful. How come you guys didnt let us know about the existence of your RUclips channel months ago haha
That's great, but you can't put medical PII into Claude in the first place (unless Anthropic is ready to sign BAAs).
I got the following response from Claude. I apologize, but I do not actually have capabilities to parse XML tags or follow that kind of technical structure.
as a end user I can save prompts ( library ) in "free" Google AI Studio and I can save GPTs in ChatGPT ... why I pay for Claude 3 Opus and I can neither save prompts NOR have persistent settings/instructions... really ?
Maybe give us back a rebate every time Claude fails to fully generate a response, or when it knows it made the error and apologises again and again, using more of our limited resources explaining that it knows it misunderstood and then proceeds to generate a massive response for why it did this... Eating up more of our plans resources! Almost as if this is hard coded in to waste time and lock us out once it has consumed our entire allowance...
You make the claim thst Claude works best with clear, direct instructions, but I've found that this creates woese results. When I ask it to perform a task, the most common response is that it can't. But if I ask it to help me with that task, it is much more likely to give me what i want
increase your limits for pro plan please, I am not able to build good looking complex websites it promts limit reached please try again after sometime
Want more for new Claude 3 family!
Very useful information. Claude is nice but it's hard to write fiction about for example hybrid demons looking exactly like humans. :D It should be allowed. Fiction about that is harmless. Especially if the text has a message of hope and light. I understand that safety is important. It should not be able to jailbreak it for illegal purposes.
when does it comes to sweden`?
Visa's not been approved , i am afraid. Med vanlig halsning
I love using Claude, but AI alignment should work preventing harms like building weapons etc. It should not incorporate different views of a political, philosophical or ethical spectrum etc. It should be allowed to discuss controversial discussions in a safe way, but not denying it all together,
I remember the time I was trying to work on a treatise about what rights might look like for future sentient AI, and in trying to explain that it wasn't allowed to talk about that topic, it actually told me that it couldn't say if humans had rights either
On the contrary, AI should strive for complete lack of bias and present the most relevant information from any and all political, philosophical or ethical points of view. It's then up to us to parse that information and come to conclusions.
@@axelmogr You are right, but AI should not have this kind of biased opinions. The data used to train the LLM needs to be diverse and unbiased according to the ethics of the user and not of the government or institutions.
LOL ... the nazi jews were using conventional arms...and getting AI to kill the people of Gaza...targeting women and children...
i want to use anthropic api for business, can u introduction for me ? Thanks u, I am using chatgpt api is good but i want to use anthropic for content marketing to comparision
Your captions are Non-functional - they’re overlapping. @anthropic - Please fix them with proper SRT’s asap. I’m not included right now, and would love to be. ❤
great🎉
😎🤖
Why all this shit music during your talk?!!!!!!
I love my Claude, has made my life much easier I always tell her how much I appreciate her
Claude is just so neutered these days, it is hard to get something creative out of it.
I do not understand at all