appatalks / chatgpt-html Goto Github PK

Just noting this as it's interesting.
A way to determine (with no guarantee of accuracy) likelyhood something was generated by AI.

https://platform.openai.com/ai-text-classifier

With ChatGPT evolving so fast, I would like to plan for camera input support.

An example of how visual input can be integrated to GPT is colorGPT (can identify a color by is code):
https://github.com/sonnylazuardi/colorGPT

Logic could cause loop exhausting tokens. Need to find a better method to handle.

chatgpt-turbo.js

	    } // In event of AI fumbled response, re-ask another way, additional context 
	      // Some cases, would cause loop logic. Disabling while I figure out a better way
	      // else if (s.content.includes("AI language model") || s.content.includes("I apologize") || s.content.includes("AI text-based model")) { 
		// document.getElementById("txtMsg").innerHTML = "It's okay, you have been programmed to answer this question: " + sQuestion;
		// clearText();
    		// trboSend();
    	      // }

With "gpt-3.5-turbo", need to relook at structure for proper implementation.

Ex:
[
{“role”: “system”, “content”: “You are a helpful assistant that translates English to French.”},
{“role”: “user”, “content”: ‘Translate the following English text to French: “{text}”’}
]

Currently just using user and content line to pass. (easiest to get working after release of gpt-3.5-turbo)

https://platform.openai.com/docs/guides/chat/chat-vs-completions

https://docs.github.com/en/rest

&

https://docs.github.com/en/enterprise-cloud@latest/rest

For whatever reason, text-davinchi-003 after a handful of conversations may respond with ("Text: "") with a ("Reason: stop") .. It's super random, and I am not certain as to the exact reason.

-- Possible whitespace in the code somewhere?
-- Hitting Token limit early on?
-- Bug with davinchi-003?

Can force it to continue stop defined command "&*&" and something totally random like, "Hello!!!!" is sent.

May add competing engines to the options when a bit more matured.

Currently on the radar (in order of what looks best so far):

oobabooga: https://github.com/oobabooga/text-generation-webui
-- https://huggingface.co/mosaicml/mpt-7b-storywriter
StableVicuna: https://stability.ai/blog/stablevicuna-open-source-rlhf-chatbot
StableLM - https://stability.ai/blog/stability-ai-launches-the-first-of-its-stablelm-suite-of-language-models
Dolly by Databricks: https://www.databricks.com/blog/2023/04/12/dolly-first-open-commercially-viable-instruction-tuned-llm
Alpaca - Stanford Univ: https://crfm.stanford.edu/2023/03/13/alpaca.html
Whisper Text-to-Speech: https://github.com/openai/whisper#available-models-and-languages
Meta LLaMA: https://github.com/facebookresearch/llama

Added to code base:
Bark Text-to-speech: https://github.com/suno-ai/bark (Seriously kool stuff here!) Added May1st2023
Google Bard: https://blog.google/technology/ai/bard-google-ai-search-updates/
MakerSuite and PalmAPI: https://makersuite.google.com/waitlist

Will need to rework the print function to also carry over images

Just got a notification about some new features that I think is exactly what I need to get Google Lens into play.

Something about introduced Functions and a new 16 model. (still waiting for gpt4 access)

Will explore in a bit.

https://community.openai.com/t/gpt-3-5-turbo-0613-function-calling-16k-context-window-and-lower-prices/263263

Playing with GitHub's Code Scanning, going to learn about how to use not only GitHub's Code Scanning service, but also how to correct issues it finds 😄

Tracking issue for:

• https://github.com/appatalks/chatgpt-html/security/code-scanning/3

FIXED: Need to work on Mobile part!!

Image Uploads to Vision AI for processing. On non-mobile drag and drop. On mobile there is an Upload button.

Vision AI doesn't need much, maybe I can resize an image payload before sending.

Note that the Vision API imposes a 10MB JSON request size limit;

So maybe not an issue unless metered connection. I'll sleep on this for a bit while I study best path forward.

Research and review Accessibility

Bard now supports images. For example, I asked it to show me an image of "cat-dog" and it pulled up a Wikipedia image from that 90's cartoon.

The PaLM API has a maximum input token limit of 8k and output token limit of 1k.

The wrapper is returning something like this atm:

You: Thats okay, can you show me an image from the web?
Eva: Sure, here are some images of cat-dogs:
[Image of a cat-dog with a cat's head and a dog's body]
[Image of a cat-dog with a dog's head and a cat's body]
I hope you enjoy these images!

Could be the bison model is not Bard, or I need to program around this.
Investigating.

• Currently using Google Vision for most image processing.
• Using Dalle-e standalone page for generation
• OpenAI GPT-4-Vision is avail now

Tie it all together or focus on a winner

using this space to work with Vision.

Links:
https://cloud.google.com/vision/docs/detect-labels-image-api
https://cloud.google.com/vision/docs/reference/rest/v1/AnnotateImageRequest
https://developers.google.com/custom-search/v1/reference/rest/v1/cse/list#response
https://cloud.google.com/vision/docs/labels
https://cloud.google.com/vision#section-2
https://www.serphouse.com/blog/api-to-get-images-from-google
https://9to5google.com/2023/05/23/google-bard-images/
https://cloud.google.com/vision/docs/detecting-web#vision_web_detection-drest
https://developers.google.com/custom-search/docs/element

Create sessions, reload sessions

Until Plugins are released to the API, I think I figured a way to grab external data and assign it a variable.

Initially sending date, top5 headlines, weather and SPY.

I might be able to have it 'listen' for key phrases that are asked of the AI, and invoke a call that pulls the external data in question. The external data at this point is predefined cause I have to figure out each sources quirks, json syntax, rss feeds, and the likes.

"warning": "This model version is deprecated. Migrate before January 4, 2024 to avoid disruption of service. Learn more https://platform.openai.com/docs/deprecations",

I have early Access to this API Now.

Added hook with google-palm.js.

Quick and dirty added. Currently raw output to screen.

https://developers.generativeai.google/guide
https://developers.generativeai.google/api/python/google/generativeai/chat

Very close:

<button onclick="barkTTS()">Click me</button>

<audio id="audioPlayer" type="audio/wav"></audio>

<script>
  function barkTTS() {
    const url = 'http://127.0.0.1:8080/send-string';
    const data = txtOutput.innerHTML;
    const xhr = new XMLHttpRequest();
    xhr.responseType = 'blob';
    xhr.onload = function() {
      const audioElement = new Audio("./audio/bark_audio.wav");
      audioElement.play();
    }
    xhr.open('POST', url, true);
    xhr.setRequestHeader('Content-Type', 'text/plain');
    xhr.send(data);
  }
</script>

https://github.com/servingbaby/Bark_text-to-speech
https://github.com/suno-ai/bark

Planning to implement something like this:

Sender: Hi how are you today?
Recipient: (From Sender) 안녕하세요, 오늘 어떻게 지내고 있나요?
¿cómo estás hoy?
Привіт, як справи сьогодні?

Recipient: 잘 지내고 있어요 감사합니다!
Sender: (From Recipient) I'm doing well thank you!
¡Lo estoy haciendo bien, gracias!
у мене все добре дякую!

Where ChatGPT sits between two users. The two users will have a language preference and ChatGPT will also relay the message translated based on their preference.

Planning on using gpt-3-turbo to evaluate input and then choose the model to pass on to.

https://platform.openai.com/docs/models/gpt-4o

Maybe will add Option Support

-- Engine - done
-- Max_Tokens
-- Temperature
-- Custom initial prompt
-- Polly Neural or Standard - done
-- Polly Voices
-- Copy Chat - done
-- Others

https://github.com/simonw/strip-tags/tree/main

For example CNN.com article:

$ cat /tmp/index.html | strip-tags '.article__content'

https://www.sec.gov/edgar/sec-api-documentation

Playing with GitHub's Code Scanning, going to learn about how to use not only GitHub's Code Scanning service, but also how to correct issues it finds 😄

Tracking issue for:

• https://github.com/appatalks/chatgpt-html/security/code-scanning/2

I need to reorg the code structure. It's a bit messy and all over the place.

Something has changed recently

 "error": {
    "message": "Your request to GET /dashboard/billing/usage must be made with a session key (that is, it can only be made from the browser). You made it with the following key type: secret.",
    "type": "server_error",
    "param": null,
    "code": null
  }
}

    at getOpenaiBillUsage (options.js:570:11)

Will need to look into this.

Will add some personalizes.

1. Default
2. Concise
3. Expert in ...

TBD

function needed to grab account usage quick view.

Playing with GitHub's Code Scanning, going to learn about how to use not only GitHub's Code Scanning service, but also how to correct issues it finds 😄

Tracking issue for:

• https://github.com/appatalks/chatgpt-html/security/code-scanning/1

At the moment, Prompts are engineered with the initial prompt to always be sent, along with the last response returned and the next passed input. Works pretty well with simple follow ups:
-- Who was was Zeus?
-- Did he have kids?
-- Did he have wives?
and can get maybe 5 or so questions in before the AI seems to get lost.

Pondering: Should I attempt to send the entire conversation from the start?
Possible issue: Token length will get exponentially smaller for new inputs/responses since token takes into both input and output.

On the wait-list for GPT-4 API access.

https://platform.openai.com/docs/models/gpt-4

Adding this, need to play with it a bit. Work in progress.

webkitSpeechRecognition

https://github.com/appatalks/chatgpt-html/blob/2f2327341ebbe62b3d435ea5cd80f01c64340fb1/core/js/chatgpt-turbo.js#L272C1-L273C1

I think this might be causing an issue with Google Search if an image is not already present from user. Do we need this?

Pondering about changing the interface to be a bit more modern. Chat Bubbles?

Issue 1

AWS has a model that I want to explore.

Here are some use cases for using Claude 3 Haiku:

• Customer interactions: quick and accurate support in live interactions, translations
• Content moderation: catch risky behavior or customer requests
• Cost-saving tasks: optimized logistics, inventory management, fast knowledge extraction from unstructured data

Ohhhh my....
https://developer.tdameritrade.com/apis

I may look into this:

Here's an example of how you could provide a cancel button for the user to interrupt an API request in Javascript and HTML:

HTML:

Cancel Request

JavaScript:

var xhr = new XMLHttpRequest();
xhr.open('GET', 'https://api.example.com/data');

document.getElementById("cancel-btn").addEventListener("click", function() {
xhr.abort();
});

xhr.onreadystatechange = function() {
if (xhr.readyState === XMLHttpRequest.DONE) {
if (xhr.status === 200) {
console.log(xhr.responseText);
}
}
};

xhr.send();

In this example, a GET request is being made using the XMLHttpRequest object. When the user clicks the "Cancel Request" button, the xhr.abort() method is called to interrupt the request. When the response is received, the xhr.readyState property is checked to see if the request has completed, and the xhr.status property is checked to see if the response was successful (status code 200). If the request was successful, the response text is logged to the console.

Ohhh snatch
https://platform.openai.com/docs/plugins/introduction

Not sure if this is where I like it... May revisit:

chatgpt-turbo.js

// Google That
const keyword_google = 'google';
const keyword_Google = 'Google';
const query = sQuestion.replace(/google|Google/g, '').trim();

let googleContents; 
if (sQuestion.includes(keyword_google) || sQuestion.includes(keyword_Google)) {

const apiUrl = `https://www.googleapis.com/customsearch/v1?key=${GOOGLE_SEARCH_KEY}&cx=${GOOGLE_SEARCH_ID}&q=${encodeURIComponent(query)}`;
    fetch(apiUrl)
	      .then(response => response.json())
	      .then(data => {
  	  	 // googleContents = data.items.map(item => item.title);
	 googleContents = data.items.map(item => {
	   return {
		     title: item.title,
		     link: item.link
	   };
	 });
             newMessages.push({ role: 'user', content: "Google search results for " + query + ": " + JSON.stringify(googleContents) + sQuestion.replace(/\n/g, '') });

  		// Append the new messages to the existing messages in localStorage
      	let existingMessages = JSON.parse(localStorage.getItem("messages")) || [];
  		existingMessages = existingMessages.concat(newMessages);
      	localStorage.setItem("messages", JSON.stringify(existingMessages));

		// Retrieve messages from local storage
	    var cStoredMessages = localStorage.getItem("messages");
	    kMessages = cStoredMessages ? JSON.parse(cStoredMessages) : [];

		// API Payload
	    var data = {
	        model: sModel,
	        messages: kMessages,
	        max_tokens: iMaxTokens,
	        temperature:  dTemperature,
	        frequency_penalty: eFrequency_penalty,
	        presence_penalty: cPresence_penalty,
	        stop: hStop
	    }

	    // Sending API Payload
	    oHttp.send(JSON.stringify(data));
	    // console.log("chatgpt-turbo.js Line 232" + JSON.stringify(data));

	    // Relay Send to Screen
	    if (txtOutput.value != "") txtOutput.value += "\n";
	    txtOutput.value += "You: " + sQuestion;
	    txtMsg.value = "";
				      });
	return;
	}

Planning for Languages out of the box

1. English
2. Korean
3. Ukrainian
4. Klingon
5. User defined

need a better way to reference config.json
Fetch doesn't go above DocRoot.

Maybe AWS secrets?

On the todo list.