belay-labs/gpt-explorer

GPT-3 Explorer

gpt-explorer's Issues

Scripting & Chaining

• Enable commands which can trigger scripts.
• These scripts can then: i. do computation, or ii. trigger API requests to do actions, get more information, call other prompts, etc
• In particular, this scripting interface can chain prompts together.

Remove Requirement to Authenticate with Gòógle

Not everyone who is on GitHub or has an interest in text-completion has an active Google account, whether out of choice or if they're in a situation that makes it difficult to log in. In my opinion, an API key should be enough for us to use the website. I'm perfectly fine logging in through Google, but is Richard Stallman?

If traceability of abusers is what this is for, I would recommend using a simple email/password mechanism. One con of this approach is that you may have to send tonnes of verification emails, which could require setting up an emails server and such.

If a Google account turns out to be necessary, I would suggestthe alternative of disclosing the requirement in the README.

if the objective is to have the site appear more presentable (which is unlikely), I would do away with the whole Google mess.

Prompt Evaluation

GPT's performance is very, very sensitive to prompt design. Thus, Explorer should enable users to iterate on prompt design.

• With built-in tools for prompt evaluation, you can annotate, rank and sort results for the same task across different prompt designs.

Mobile view width problem

Describe the bug
Can't fit to the screen

Smartphone (please complete the following information):

• Device: [e.g. iPhone6] Android
• OS: [e.g. iOS8.1] pie
• Browser [e.g. stock browser, safari] Chrome
• Version [e.g. 22]

Update styling to indicate text provenance [human vs GPT]

Bold: human
Unformatted: GPT-3

Source: https://openai-api.slack.com/archives/C015JAN056F/p1596143417188900?thread_ts=1596140805.188800&cid=C015JAN056F

Token counter in prompt field

https://openai-api.slack.com/archives/C015JAN056F/p1597165057206700

• David Schnurr

https://repl.it/@schnerd/gpt2-tokenizer
huggingface/tokenizers#63

Show LogProbs

Show Log Probs in UI

Add davinci-v2b languageEngine

• The languageEngine list is currently hard-coded.
• OA just released davinci-v2b.
• Add davinci-v2b to that list
• Eventually: get the up-to-date list upon initial page load from https://beta.openai.com/api-ref#list-engines-get

Add `frequency_penalty` and other missing API params

Params to add

• frequency penalty,
• presence penalty,
• model

Q: how to handle n?
This would create downstream UI considerations.
Perhaps they could just load sequentially in main page, s/t a user could page through them -- evaluating each as they go.

Add GitHub link at bottom

Public prompt library + imports

"GitHub for Prompts". Let's make it easy to share and build on each others' work!
requirements:

• easy to add prompts
• easy to discover new prompts

Search through prompts

GUI for the Search API 🔍

OpenAI's Playground doesn't have this yet, so it's a high priority to get a GUI out for non-technical GPT-3 users.

Export history [2]

Allow users to bulk export their history

• Spec output format