anthropics/courses

Anthropic's educational courses

Anthropic courses

Welcome to Anthropic's educational courses. This repository currently contains four courses. We suggest completing the courses in the following order:

1. Anthropic API fundamentals course - teaches the essentials of working with the Claude SDK: getting an API key, working with model parameters, writing multimodal prompts, streaming responses, etc.
2. Prompt engineering interactive tutorial - a comprehensive step-by-step guide to key prompting techniques

• Google Vertex version

3. Real world prompting course - learn how to incorporate prompting techniques into complex, real world prompts
4. Tool use course - teaches everything you need to know to implement tool use successfully in your workflows with Claude.

courses's Issues

In anthropic_api_fundamentals/05.Streaming.ipynb error: unexpecter keyword argument 'event_handler'

Path to course in question:
https://github.com/anthropics/courses/blob/master/anthropic_api_fundamentals/05_Streaming.ipynb
In the Streaming helpers part of the 05.Streaming segment of the api_fundamentals courses when running the last piece of code I am getting the following error:

I tried downgrading to an earlier version of anthropic (0.33.0) but the problem persists.

Can you please provide a tutorial for the Artifacts in API?

How can we enable Artifacts via the Claude API?

Bad link in medical prompt chapter

The link is ToolUse instead of tool_use when referring to the tool use course.

ToolUse : How to use tools to scalably chunk output beyond output token limit?

I am writing a module to utilize AI techniques to digest large quantities of information and in a layered stepwise fashion distill it cumulatively into an intermediate structured form that can be used as an input to create a written human rights application.

I started with OpenAI GPT4o but hit issues with their max 4,000 token completion token cap.
I then moved to the OpenAI equivalent of the "tools" feature in an effort to have the AI generate the output in "chunks" that, collected together, represent the final output, which may be very long and exceed a model's token completion cap.

I then moved to adapt the script to Anthropic (which is almost complete).

It appears the API wants previous content to be sent back to it in order to complete the "conversation" but this may not scale for me. To summarize two months of evidence, and provide appropriate system content, I am already using 150,000 tokens out of the 200,000 token max context length. I will try to make this work by sending back the most recent tool_use clause that matches the tool_result response and see if the content flows properly into my tool (which saves chunked content to be reassembled later).

Initially, I have the AI in a first pass take detailed content from particular files containing emails and text messages from a particular date range, the output is a summarized structured format. I use the majority of the context window size (system prompt contains legal and medical background documentation, user prompt contains email/text message data) and the output is relatively small.

In pass 2 I take all those pass1 output blocks and cumulatively combine them, meaning that both input and output sizes are expected to grow to be quite large. More accurately, I do token counting and fit as many input blocks as I can while factoring in the expected output token size into the calculation. While Claude's 200,000 context window is certainly better than gpt4o's 128,000 window, I still need a scalable solution to summrize vast quantities of data and not hit any output token limits.

I should have asked : Does Claude-3.5 Sonnet have any restrictions on output/completion max token count?

Is the tools feature the most appropriate feature to use in this case? If so, how can I avoid sending in past context so huge that it quickly consumed the context window?

https://community.openai.com/t/inception-based-design-for-the-ai-assisted-creation-of-a-written-human-rights-complaint/863669

PS. I'm getting the following error because I'm omitting to send back previous context when responding with tool call results:

Error code: 400 - {'type': 'error', 'error': {'type': 'invalid_request_error', 'message': 'messages.0: tool_result block(s) provided when previous message does not contain any tool_use blocks'}}

The "tool_use" block contains the most recent chunked output.

Let's say that the final output contains ten chunks, does Claude expect me to give only the most recent chunk sent via a tool_use block or all previous tool_use and tool_result blocks? This sequence is not at all clear to me. Requirement is to provide system/user prompts and have output chunked back to us such that the joined chunks form parts of a single whole response (so the model must somehow maintain state through the initial call followed by the call containing the tool result content.

Also, when providing a tool result, do I also need to send back the initial user and system prompts? If so, the scalability of this approach may be in jeopardy.

Wrong output on 10.2 - Prompt Engineering Interactive Tutorial

This doesn't work properly, tried a few times (with temp = 0) and it always returned:

<function_calls>
<invoke name="calculator">
<parameter name="first_operand">5</parameter>
<parameter name="second_operand">3</parameter>
<parameter name="operator">+</parameter>
</invoke>

Use %pip install instead of !pip install in the notebooks

I was following this notebook: https://github.com/anthropics/courses/blob/cf2979dc88626f15c760ce83bc9e9e21015fcac5/prompt_engineering_interactive_tutorial/Anthropic%201P/00_Tutorial_How-To.ipynb

I cloned the repo and started the notebook like this, which was delightfully fast:

uv run jupyter-notebook .

But I got an error trying to import anthropic. It turns out this was because I was running Jupyter from a different environment, so this line:

!pip install anthropic

Installed the package into my default system environment, not the one for the notebook.

The fix is to use this instead:

%pip install anthropic

courses/prompt_engineering_interactive_tutorial/Anthropic 1P/00_Tutorial_How-To.ipynb

Lines 29 to 31 in cf2979d

	"source": [
	"!pip install anthropic"
	]

I just can't get multi turn tool use to work

In multi turn mode API keeps hallucinating a tool use as a standard reply

It is unclear how to encode tool use and result... I am doing this now...

A simple payload or API example that illustrates this (just a messages block with a bunch of history in text form) would be enormously helpful.

XML tags are hidden in the GitHub notebook preview

In the prompt_engineering_interactive_tutorial/Anthropic 1P/04_Separating_Data_and_Instructions.ipynb notebook, for instance, even though this line of code contains XML tags, they are not visible in the GitHub preview:

Since it looks like a GitHub issue I created this discussion in the GitHub community:
https://github.com/orgs/community/discussions/137038

Real World Prompting Feedback

I recently converted Real World Prompting tutorial to TypeScript to make it more accessible (and for my personal learning and enjoyment). Along the way, I made a few adjustments to the content, so I just wanted to share these changes with the course creator.

1. API Parameters

• The tutorial is still referencing the old sonnet model (claude-3-sonnet-20240229).
• The temperature param is not set, even for tasks that could benefit for it (e.g. summarization).

2. Prompt Adjustments

As a result of adjusting the params in 1), I'm seeing way fewer hallucinations and out-of-character behavior (e.g. an assistant referencing <context>), but I noticed some areas where performance declined. For example:

• In chapter 4, I had to explicitly instruct Claude to output the <json> tag after "Generate a JSON object with the following structure". Without that, Sonnet 3.5 would skip wrapping its response in tags (which might be preferable in most cases).
• Also in chapter 4, when dealing with incomprehensible (e.g. "blah blah blah") and empty transcripts, Sonnet 3.5 would fail to output a JSON object. Instead, it responded with a long explanation asking for a valid transcript. I solved this by adding two additional examples to the prompt.

If you'd like more details on these changes, please let me know. For reference, here's the project I've been working on.

Lastly, I appreciate the effort put into creating this course and I'm looking forward to future updates!

Add guidance for parallel/multi tool calling in 1 response