Issue with HTML formatting of email body in send email node

MindStudio IDE Bug Reports
1

Hello,

First of all, I’d like to say that the feature in the Send Email node, which automatically formats the email body into HTML, is a great feature. Thank you for that.

However, I noticed that when formatting to HTML, it does not include the full text.

Here’s my use case:

I send a video URL via email to Mindstudio and get back an email with 3–5 bullet points as well as a summary of the video of about 500 words.

This is the generated text:

<email_body>
Here is the summary of the following video:

WIRTSCHAFTSDATEN auf der BLOCKCHAIN?! :man_facepalming: USA scheitern am ORACLE-PROBLEM!

The US government is publishing economic data on various blockchain networks, which according to the speaker doesn’t make much sense.

The main criticism is the oracle problem: data on the blockchain can be wrong, and putting it on the blockchain does not make it more true.

From a technical perspective, blockchain is inefficient and, apart from Bitcoin, hardly useful, since its functions can also be implemented with other technologies.

Publishing economic data on the blockchain is, in the speaker’s view, more symbolic and does not solve any real problems.

The speaker sees blockchain technology as overrated and criticizes the lack of understanding among many actors in the crypto sector.

Summary:

The text criticizes the use of blockchain technology, especially on networks like Ethereum. The main point is that storing data on the blockchain does not automatically guarantee its truthfulness.

This is illustrated with the example of the U.S. Department of Commerce: the agency publishes economic data (GDP) on different blockchain networks such as Bitcoin and Ethereum. The speaker argues that this provides no real added value, because:

  • The data may be wrong or inaccurate
  • Blockchain technology is inefficient
  • Economic data can be revised afterwards anyway

The speaker considers blockchain outside of Bitcoin largely useless. He stresses that alternative technologies can provide similar functions. Blockchain only solves the time-server problem for cryptocurrencies but is otherwise a poor technology.

In conclusion, he describes the action of the Department of Commerce as a symbolic gesture without practical benefit – a kind of “Matrix” or “Central Bank 2.0” that does not really understand the true value of blockchain technology.
</email_body>

And this is the generated HTML email. As you can see, not the full text is included:

(german)

Did I do something wrong, or is this a bug?

thanks

2

Hi @mefrankyt,

Could you share the Debugger logs for the run? To do that, open the Debugger, select the run, click Share, and reply with the URL:

image
image777×616 78.9 KB

Please make sure there’s no sensitive information in that run, as the shared log will be public.

3

I wanted to do it, but I didn’t for exactly this reason, because this file contains personal email addresses. What can we do?

4

What are the model settings you are using to generate the email? My hunch is that this might be an issue with the max response size of the model truncating the text. Also, could you please share the prompt instructions you are using in the block?

image
image790×676 44.7 KB

5

Thank you very much for the feedback.

I’m using the model settings of the workflow (3.5 Haiku with a maximum response size of 8192).

I left the instructions empty since they seem to be optional.

6

Thank you so much for the hint, and sorry… Sometimes the solutions are so simple, and I just don’t think of them.

I switched to Sonnet 3.7 for testing with a maximum response size of 128,000, and now it works. So that must have been the issue.

Many thanks for your help.

… still a bit puzzling though, since Haiku’s response size really should have been sufficient :man_shrugging:

7

It’s possible that the full length of the prompt, as well as the HTML it generated (more tokens than just the text, if you click view source on that HTML you shared), started to run up against the constraints. Often it seems the models are hesitant to actually “max out” their full supported response size, especially older models like 3.5 seem to require a lot of coaxing to really use their full output. Sometimes you need to specify “Always include the full text of the summary” or something similar in the prompt to keep it from paraphrasing.

1 Like
8

Alright, thank you very much.