I’m excited to share the “Super Brain” feature I recently built into our Mac app - Elephas
You can bring your own data, set of PDF/Docs documents, blogs, Notion links, and Obsidian or Roam databases and load it into a unique “brain.”
Here’s what it looks like in action, loading a brain and replying to a support email,
Here are the primary use cases,
- Chat with your content. Data could be your books, essays, papers, or legal contracts.
- Reply to emails. Across popular email clients, Elephas can use your content to respond.
- Write official content. Works across apps.
- Write social media posts.
If you decide to try the 30-day free trial, you can use the coupon code “OPENCOMM” to get 25% off. I will appreciate your support.
Note:
- One of our users loaded 1.5 GB files and is using it regularly to write his content
- To upload a 1,700-page PDF file, it costs around 40 cents. Asking questions incur only the regular token charges. In other words, this feature is optimized for cost savings, and relies on your local storage.
3 Likes
Given the user has to also pay for OpenAI, Have you considered a one off payment option for the app?
Hey,
There has always been a lifetime deal, and the coupon code applies to it as well. Additionally, I will soon be introducing limited lifetime deals for the Pro plan.
A bug report:
I have uploaded this one PDF
Then I created a snippet “Please answer this”
Then I asked: Which foods contain serotonine?
And I get this answer:
" Context 1 and Context 5 mention the presence of serotonine in certain foods. In Context 1, serotonine is found in biogenicamines and its content is 1.0 mg/MJ. In Context 5, serotonine is found in salt-waterfish and its content is 7.7 mg/MJ."
What contexts? I have only uploaded one file. Makes no sense like this to me. Also the answer is very bad. Should be like “Tomatoes, etc…”.
Granted. The PDF probably is difficult, because of being mostly tables. Nevertheless… this is ment as a way to improve the app.
1 Like
Click the icon to see the contexts. Contexts are extracted pieces of information from the PDF to answer your question. You can also adjusts the system message to make the brain respond differently, under the Brain settings.
Update: I realized you are not using the Super chat. I will review this.
I looked into your PDF and found it mostly consists of data points. Super brain performs well with text data.
One example I tried, I set up the brain with the poverty paper from the world bank, https://openknowledge.worldbank.org/server/api/core/bitstreams/611fc6f2-140b-551e-9371-468eec64c552/content
and asked it to write “Write a 300-word post on job losses post covid.”
I agree. My PDF is probably a bad example.
Nevertheless I have now fed a super brain with 30+ scientific papers and asked questions in the super-chat about it. The questions are answered so-so, not bad, but a bit incomplete, as if it wouldn’t understand all papers.
Anyhow. The real problem is the “context 1” “context 2” etc. thing.
1: Having no idea where context x is taken from, is a problem.
2: If I click to see the context I see more contexts than ChatGPT specifically mentions in its answer.
Suggestion. I think it would be a good idea to propose a PDF-Naming convention, for GPT to pick up on.
For example:
“2023 - Article - Giancola et al - Dark Triad and COVID-19 vaccine hesitancy- the role of conspiracy beliefs and risk perception.pdf”
“Year, Type of Document - Authors - Title”
This could then be picked up by Chat like: “Giancola et al. (2023)” within the answer. In the context overview the whole filename could be displayed.
THAT would be powerful and very typical for academic writing.
1 Like
Sure, I will add the names of the referred documents.
