Summarizer

Getting started

Just Read's summarizer is powered by OpenAI's chat completion endpoint.

To use the summarizer, you must add your own OpenAI API key to the summarizer options within Just Read's options page. Note that this key is stored in your browser account data and never shared with anyone except OpenAI (not even the Just Read server).

If you already have an OpenAI API key, you can use it if you'd like.

If you don't have an API key or want to use a separate API key specifically for Just Read (which we recommend doing), follow these steps:

1. Go to platform.openai.com/account/api-keys.
2. Log in or sign up for an OpenAI account. If you're new to OpenAI, you can get 3 months free for up to $5.00 in costs. After that, you'll have to pay OpenAI to continue using their services.
3. Click "Create new secret key".
4. Make sure to copy and save this key somewhere safe. OpenAI doesn't let you view a key after it has been created, so make sure to keep track of it. With that being said, you can create multiple keys so if you lose this one you can generate another later.
5. Go to the Just Read options page by right clicking the Just Read icon and clicking "Options".
6. Under "Summarization options", paste your key where it says YOUR_OPENAI_API_KEY_GOES_HERE. Make sure the key you provide has quotation marks around it.
7. (optional) If you'd like, you can change the other options like the prompt. The parameters available to you are covered below.

Parameters

Just Read lets you pass in any options that you can normally send into OpenAI's chat completion endpoint. The only one that is required is the API key. You can change these parameters by going to Just Read's options page.

Here's a list of the options we think you are most likely to potentially change:

• baseUrl (string) - The URL you want the summarizer request to go to (in the case you want to use something other than OpenAI's public endpoint). By default, Just Read provides the baseUrl of https://api.openai.com/v1/chat/completions

• key (string) - Required - Your OpenAI API key. To get yours, follow the steps above. This goes directly from your computer to OpenAI so there's no security risk in using it, following Just Read's privacy statement.

• model (string) - The OpenAI chat completion model to use. We recommend using gpt-3.5-turbo, which is Just Read's default because it's cheapest option while good for this sort of task. But you can use gpt-4 if you'd like. Note: OpenAI restricts gpt-4 to users who have made a payment to them.

  If the selected model does not allow requests large enough to handle the summarization request, Just Read will automatically attempt to use the model that is able to handle a request of that size. For example, if gpt-3.5-turbo is used but the request requires more than 4,086 tokens but less than than 16,384 tokens, gpt-3.5-turbo-16k will be attempted to be used. If the request requires between 16,384 and 32,768 tokens, gpt-4-32k will be attempted to be used. If the request is larger than 32,768 tokens, it's too large for OpenAI and will error.

• prompt (string) - What you want the OpenAI model to do. By default, Just Read provides the prompt Summarize the content you are provided as consisely as possible while retaining the key points. but you can customize this prompt if you'd like. See the prompt customization section below for more details.

• temperature (number) - This is a number between 0 and 1 that affects the "randomness" of the model. The closer to 0, the more likely that GPT will choose the same result. The closer to 1, the more leeway the model will take. It's common to use a temperature around 0.3. This defaults to 0.

• max_tokens (integer) - This is the max number of tokens thats OpenAI is allowed to spend processing the request including the prompt, content, and response. Usually an article summarization request requires around 1,000-5,000 tokens depending on the article length. You can use max_tokens to limit how many tokens a request uses but note that if the max_tokens amount is reached, the summary will likely be cut off mid way or missing completely. By default, Just Read doesn't use max_tokens.

  The main reason why you might want to use this is to disallow Just Read from using a model that supports larger requests. For example, if gpt-3.5-turbo is selected, it can spend up to 4,096 tokens on a given request. If you didn't want a request larger than 4,096 tokens to be passed to a model that supports larger requests, you could set max_tokens to 4096.

  Side note: You can view how many tokens a request took by hovering over the "Summary" title if the "Replace the article content with the summary." setting is off or in the browser's developer console if it is on.

Prompt customization

While Just Read asks for a concise summary, which ChatGPT responds with a paragaph of information, you can prompt it for other things. Note that the response that Just Read expects is either HTML or a string.

Examples

To get it to use bullet points: Summarize the content you are provided using 3 bullet points. Put your response in an HTML format.

To get it to summarize using a different language, you can either tell it to use the language (like Summarize the content you are provided in German as concisely as possible.) or just write the prompt in the language you desire the summary in (like Fassen Sie die Inhalte, die Sie erhalten, möglichst prägnant zusammen.).