Tech Roundtable Meeting 3/22/23

March 23, 2023 Tech Roundtable meeting on Zoom

Minutes

Topic of AI continued from the last meeting with an overview of the “new” Bing using Microsoft Edge, the OpenAI outage occurring on Monday 3/20/23, the announcement of Microsoft Bard and my PHP application of the OpenAI API.

I opened the meeting emphasizing that the move to using AI’s is moving very fast now that there are interfaces and API’s to using them.

I also suggested that all businesses and organizations need to evaluate how AI can be used in their workflows, processes and products.

We also posed the question why you might use the ChatGPT or equivalents versus searching the web or asking Siri or Alexa.

John Gbur opened a discussion about the AI impact on jobs. We looked at an article about a job impact study done. It seemed initially it would work its way in as an adjunct to existing jobs but over time apps and technology will be developed to eliminate jobs.

New Bing, Bing Chat and OpenAI GPT4 Model

I demonstrated the use of Bing Chat using Microsoft Edge. Bing Chat uses OpenAI and it is claimed to use the new GPT-4 model.

Microsoft has a big interest in OpenAI. It also has been integrating it into its own applications. One is access to the New Bing via Microsoft Edge.

This is the interface for Bing Chat in Microsoft Edge.

With Microsoft Edge you need to have a Microsoft Account. You can do that on the Bing Web Page.

Once you have that account and are signed into it, you can navigate to the Bing Web Page to use a “Chat” option that appears at the top of the Microsoft Edge Screen.

There are several ways you can access Bing Chat once you have the Microsoft Account:

You can go to bing.com/new and download the new version of Microsoft Edge, which has Bing Chat integrated as a button on the top-right corner.
You can also use Bing Chat on your smartphone via the Edge app or the Bing app. You can use your voice to ask me questions and receive spoken responses.
You can also access Bing Chat from the search bar in Windows 11, which opens a chat window in Microsoft Edge directly from your desktop.

You cannot use the Bing chat in other web browsers.

Screen accessing Bing Chat in other Web Browsers

Bing Chat calls a prompt a New Topic. I demonstrated a few topics.

Some differences from OpenAI’s ChatGPT include embedded hyperlinks in the response. A “Learn More” bar at the bottom of the response. The “Learn More” links are to web pages. Below the “Learn More” is a list of additional topics suggested that will generate another response.

There is also a share and a copy feature.

Before you start your first topic you can choose a “conversational style”.

Once you choose a style it does not appear you can change it for the session. To change you can refresh the page or open a new tab or window.

At this point it does not appear to keep your chat history from sessions to session.

ChatGPT Plus and GPT-4 Model

I demonstrated OpenAI’s ChatGPT plus and the GPT-4 model. ChatGPT Plus is a paid ChatGPT tier.

When you start a new chat, you need to specify the model:

The ChatGPT interface has a disclaimer about the GPT-4 model usage limit:

“GPT-4 currently has a cap of 25 messages every 3 hours. Expect significantly lower caps, as we adjust for demand.”
ChatGPT GPT-4 Model disclaimer

When you open a chat in your chat history the model used is shown at the top if the page.

ChatGPT chat that used the GPT-3.5 model

Once you start a new chat, the model selected cannot be changed for that chat.

Google Bard

Google officially released Google Bard this week. At this point all you can do is get on a wait list to use it.

I asked Bing Chat about Google Bard and here is its response.

Google Bard has a FAQ if you want more information.

OpenAI 3/20/23 Outages

OpenAI ChatGPT was unavailable Monday 3/20/23. The issues also impacted the OpenAI API. Reports indicated that Microsoft’s applications of it were not impacted.

Besides the unavailability of ChatGPT, when it was restored, the chat histories were not available. A problem with the histories being accessible by other accounts, an issue they also had in the past, resurfaced. Chat histories were not available until late Wednesday.

OpenAI maintains a status page that provides information on any issues.

You can subscribe to the page to get email updates of issues as well as postmortems of any issue when resolved.

I brought up the question if mission critical operations depended on the AI, having no fallback could be a problem.

OpenAI API Accessed with PHP

I decided to try my hand at using PHP and a web page to access the OpenAI API. Essentially this what Bing Chat and ChatGPT are doing. But so are many applications being developed.

You need an account at OpenAI. This is the same account that gives you access to ChatGPT. In that account you need to get an API key. This is free. It is used in the code to authenticate your requests to the API.

The use of OpenAI API does have a cost. The first time you create an account you receive $18 to use within 3 months. So far my endeavors have only used $0.10.

OpenAI provides libraries using Node and Python. Then there are many libraries called Community Libraries for other languages like PHP, C# / .NET, Java, Swift and so on. These are not guaranteed by OpenAI.

PHP had two community libraries to use. I choose the OpenAI PHP API Client library by Tectalic. They also have a library for Javascript.

I used Composer to install Tectalic library and the other libraries that Tectalic used.

I could not use the GPT-4 model. I did however get on the wait list to get access to it. This flummoxed me for a while as the sample code had the gpt-4 model in it. I had to use the gpt-3.5-turbo model. This also flummoxed me for a while as I thought there was a gpt-3 model. However this is more of a category designation of models.

The interface I made was spartan but it could be run a web server with a URL to it.

Lon Hosford’s first PHP OpenAI API project with a web interface

Although we did not discuss code in detail, I will share the PHP code of the first version which did not have a web interface. The OpenAI API key is read from an external file on the line that starts $api_key. You can create your own file with your OpenAPI key or just put your OpenAI API key in the code. Otherwise the code can be run as is on a web server as a URL.

<?php
require "../vendor/autoload.php";

try{
echo '<pre style = "white-space: pre-wrap;">';
//$model = 'gpt-4'; // Will need account access. 
$model = 'gpt-3.5-turbo';
$api_key = file_get_contents('.open-ai-example-01-env');

echo $api_key . PHP_EOL;
//return;
$openaiClient = \Tectalic\OpenAi\Manager::build(
	new \GuzzleHttp\Client(),
	//new \Tectalic\OpenAi\Authentication(getenv('OPENAI_API_KEY'))
	new \Tectalic\OpenAi\Authentication($api_key)
);


$response = $openaiClient->chatCompletions()->create(
	new \Tectalic\OpenAi\Models\ChatCompletions\CreateRequest([
		'model' => $model,
		//'model' => 'text-davinci-003',
		'messages' => [
			[
				'role' => 'user',
				'content' => 'Will using a well designed and supported third party package save time?'
			],
		],
	])
)->toModel();

echo $response->choices[0]->message->content;
}catch(\Tectalic\OpenAi\ClientException  $e){

	echo '$e: ' . print_r($e,true);

}
echo '</pre>';
?>

When it was demoed, the response time was abysmal. That was not the case when I was working on it. We checked the OpenAI status page, but it had no issues. However the next morning I got an email from OpenAI showing there was a latency issue going on at the time I was doing the demo which caused the long delays.

After Hours Segment

In the After Hours segment the impact of using AI was bandied about. For instance was usage in military applications was investigated. The ChatGPT claimed it could not suggest a military strategy for Ukraine to retake lost territory but did provide the various options in a general sense they had.

I discussed the impact on Quora that depends on humans to answer questions. I used ChatGPT to answer some questions on Quora to see how they are received. I did edit and modify my ChatGPT answer slightly before posting. Still I wonder how many people on Quora are using ChatGPT to game their stats.

Bing Chat hypes on its landing page about writing a Haiku. Again we discussed what the impact is on such endeavors and human assessment as to what is a Haiku and its artistic value when an AI assistant is used to write all or part a Haiku or any literature. I had Bing Chat write a Haiku earlier in the day and posted to my Facebook page. I only changed on word.

We also talked about Microsoft 365 Office CoPilot feature which is an AI component to it applications like Word, Excel and Powerpoint. We plan to look at CoPilot next meeting. I brought up the wonder if Apple was considering the same for its Office apps.

We also discussed Nvidia GPUs used in ChatGPT and the impact it had on their bottom line. This was part of a discussion where money could be made in the AI revolution. These GPUs cost $10K to $30K and thousands are used or are needed. Microsoft footed the bill in the tune of $10B for this phase of the project.

Links

These are the links that were discussed, mentioned or relevant to the meeting.