[Beta Testers] AI Writer - GPT-3 Based - Relevant Content
- Thread starter Cognitive
- Start date
- Status
lawrencelau
Supreme Member
- Mar 29, 2008
- 1,310
- 575
I am interested to try your beta.
I am currently using nichesss and looking for app that can create long form content
I am currently using nichesss and looking for app that can create long form content
jobhelperapp
Junior Member
- Feb 16, 2021
- 160
- 83
Hi! Interested to test it.
whats_up_doc
Regular Member
- Jun 19, 2013
- 373
- 210
Interested. I'm using jarvis and ryter... it'll be interesting to see how your tool compares to them.
Thanks.
Thanks.
Cognitive
Regular Member
- Mar 11, 2015
- 401
- 222
Thank you all for applying! We're really excited to see this level of interest in this kind of apps.
We've sent the first batch of 30 invitations and we will be sending more invitations during the next few days.
A few updates:
Some of you have already pointed out a few bugs and we've already discovered several others. We've already fixed some of them and we're working on solving the remaining errors.
### IMPORTANT ###
Some clarifications regarding our app and the way it works (based on the questions we've received on PM):
1. Relevanto
The Relevanto engine is a machine learning algorithm that's constantly "updating" itself. We started it a few months ago and it scans the SERPs, learning lots of keywords, longtails, topics, questions, related searches etc. Also, it scans the main actors in each niche/ topic and learns how they build their content, how they are structuring it, what keywords they are using, what related topics they are approaching in each main article together with lots of other things.
Now, when you start to build an article/ blog post with our app (except the Basic module), the generated title, the intro paragraph and the sub-topics ideas and their variations of intro paragraphs are generated by Relevanto, based on the data and parameters it has learned until now.
The standard generation in each module (when you simply click "Generate") does not use Relevanto. This is mainly because Relevanto is too slow. But, you will be using a GPT-3 type model that is fast and "contains" more parameters, even though it might not "know" specific things in your niche, just like Relevanto does. However, if the context of your article already has data generated by Relevanto (title, intro paragraph, sub-topic idea, sub-topic intro paragraph) a GPT-3 model will be able to continue the text generation with relevant content. Also, it is always a good practice to add your own input after each generation in order to "steer" the AI in the direction of your preferred output, and, if necessary, delete what you consider is not relevant in the text that has just been generated.
We are also working on extending the level of interaction between the Text Generation modules and Relevanto, the scope being to use Relevanto data during each generation. But this will happen at a later stage of this project.
2. Speed
While our app should normally be able to generate 100 words in about 4 seconds, sometimes you might find it slow. This is either because there are a lot of users testing the app, or we are updating/ fixing something and we need to freeze the app while we're uploading something on the servers. Also, please see section 5. Bad text/ words of this post. This could also be a cause of delay.
Aslo, please note that we don't support text streaming (the typewriter effect) when new generated text is being displayed. This effect makes the app look faster when generating new text but it also consumes some hardware resources. We just display the entire text after generation has ended. We might add this feature, too, but not during the beta phase.
3. Empty Rows and spaces
Sometimes these models are generating lots of empty rows or leave trailing empty spaces. While we have active filters meant to delete them, sometimes we miss a few (working on this, as well). Please delete them before generating new text. If you don't delete them, during the next generation, the AI model will think that you intentionally left them, and, because it's a machine learning algorithm, it will add them in its next text generations.
4. Generated Text Length and Creativity
While we set the maximum length of generated text to 200 words, it is always a good idea to ask it to generate a minimum value of words. It may sound counterproductive and slow. However, the more tokens a GPT-3 type model has to generate, the more it will lose its coherency/ relevancy. The same applies to the Creativity setting. In our tests we noticed that with these settings set at a minimum value we were getting better content, allowing us to finish our articles faster.
5. Bad text/ words
Some of the companies behind these AI models are limiting the way a user is generating text and, also, the content he/ she is generating. Therefore, our systems are automatically trying to compensate this by using a different model. It works most of the time but there are situations when the generated content is not related to the main topic, or it is in bad shape and cannot be used. Please delete it and hit the "Generate" button again. If the same happens, please modify the previously generated text (which is considered the context of the next text generation) and try again.
6. Save button
Please use this function only after you have finished building your article. It is meant to save your work on our platform so you could download/ copy it later. Do not use it before you have finished building your article.
7. Restart button
Please use it each time you decide to start from scratch. It helps clearing the previous context and lets the AI models focus on the new article/ topic.
I've just realized that I should put this post in the Knowledge section of our website
We've sent the first batch of 30 invitations and we will be sending more invitations during the next few days.
A few updates:
Some of you have already pointed out a few bugs and we've already discovered several others. We've already fixed some of them and we're working on solving the remaining errors.
### IMPORTANT ###
Some clarifications regarding our app and the way it works (based on the questions we've received on PM):
1. Relevanto
The Relevanto engine is a machine learning algorithm that's constantly "updating" itself. We started it a few months ago and it scans the SERPs, learning lots of keywords, longtails, topics, questions, related searches etc. Also, it scans the main actors in each niche/ topic and learns how they build their content, how they are structuring it, what keywords they are using, what related topics they are approaching in each main article together with lots of other things.
Now, when you start to build an article/ blog post with our app (except the Basic module), the generated title, the intro paragraph and the sub-topics ideas and their variations of intro paragraphs are generated by Relevanto, based on the data and parameters it has learned until now.
The standard generation in each module (when you simply click "Generate") does not use Relevanto. This is mainly because Relevanto is too slow. But, you will be using a GPT-3 type model that is fast and "contains" more parameters, even though it might not "know" specific things in your niche, just like Relevanto does. However, if the context of your article already has data generated by Relevanto (title, intro paragraph, sub-topic idea, sub-topic intro paragraph) a GPT-3 model will be able to continue the text generation with relevant content. Also, it is always a good practice to add your own input after each generation in order to "steer" the AI in the direction of your preferred output, and, if necessary, delete what you consider is not relevant in the text that has just been generated.
We are also working on extending the level of interaction between the Text Generation modules and Relevanto, the scope being to use Relevanto data during each generation. But this will happen at a later stage of this project.
2. Speed
While our app should normally be able to generate 100 words in about 4 seconds, sometimes you might find it slow. This is either because there are a lot of users testing the app, or we are updating/ fixing something and we need to freeze the app while we're uploading something on the servers. Also, please see section 5. Bad text/ words of this post. This could also be a cause of delay.
Aslo, please note that we don't support text streaming (the typewriter effect) when new generated text is being displayed. This effect makes the app look faster when generating new text but it also consumes some hardware resources. We just display the entire text after generation has ended. We might add this feature, too, but not during the beta phase.
3. Empty Rows and spaces
Sometimes these models are generating lots of empty rows or leave trailing empty spaces. While we have active filters meant to delete them, sometimes we miss a few (working on this, as well). Please delete them before generating new text. If you don't delete them, during the next generation, the AI model will think that you intentionally left them, and, because it's a machine learning algorithm, it will add them in its next text generations.
4. Generated Text Length and Creativity
While we set the maximum length of generated text to 200 words, it is always a good idea to ask it to generate a minimum value of words. It may sound counterproductive and slow. However, the more tokens a GPT-3 type model has to generate, the more it will lose its coherency/ relevancy. The same applies to the Creativity setting. In our tests we noticed that with these settings set at a minimum value we were getting better content, allowing us to finish our articles faster.
5. Bad text/ words
Some of the companies behind these AI models are limiting the way a user is generating text and, also, the content he/ she is generating. Therefore, our systems are automatically trying to compensate this by using a different model. It works most of the time but there are situations when the generated content is not related to the main topic, or it is in bad shape and cannot be used. Please delete it and hit the "Generate" button again. If the same happens, please modify the previously generated text (which is considered the context of the next text generation) and try again.
6. Save button
Please use this function only after you have finished building your article. It is meant to save your work on our platform so you could download/ copy it later. Do not use it before you have finished building your article.
7. Restart button
Please use it each time you decide to start from scratch. It helps clearing the previous context and lets the AI models focus on the new article/ topic.
I've just realized that I should put this post in the Knowledge section of our website
Last edited:
- Status