Hello all!!

I apologize in advance for a post that is more philosophical than technological, but it is something that has been on my mind for several weeks.

As I'm sure we've all realized, AI is everywhere and I think it's best to see what it is all about. The risk is that we could be left behind in terms of technological capabilities and that our offer appears to be outdated.

My perception is that AI is not another "buzz word" or a passing fad. It is not something limited to the technological field as was the case with GUI or SQL. This thing really is everywhere: from schoolwork to big data analysis, and everything else in between.

Some serious analysts place it at the level of the Internet itself: before and after.

In the very modest research I have done, everything seems to still be at the level of promises. Some software applications say they already incorporate AI but everything makes it seem like they install a parallel product that is not truly integrated into the applications.

There is a lot of talk about how they will be able to optimize inventory levels, improve the layout of products within warehouses, predict sales, analyze market trends, perform almost any repetitive task and the list goes on and on, but when it comes to "show me", the magic vanishes.

There are those who say that AI will be able to learn to program in any language if we give it the rules and purpose of the programs, but this still seems remote to me (or I prefer to think this way!!). I see it as more likely that, based on already developed programs and their respective use cases, automatic documentation can be generated or bugs can be detected.

Apparently AI has two basic modes of operating: learning mode and operational mode, so to speak.

The difference seems to be that it is not something focused on data, but also considers things such as the use that is given to the data, the frequency of these different uses, and, in general, qualitative or heuristic aspects. The flow, the timing, the programs used and their purpose are also part of the analysis along with the data they produce.

Back to Earth, right now it's hard for me to think in some kind of XCALL AI_SOMETHING, so in addition to sharing with you the concern of something as disruptive as AI, I would like to know if you have any other information or if you are doing something about it.

Jack will probably surprise us with a set of already implemented instructions similar to MIAMEX but for AI!!!

Best regards.

I was thinking that our competitive edge was to remain the only AI-free development platform left !!!

(It's time for dinner here, so that's as much thoughtful analysis I can muster at the moment. Maybe while I'm eating and sleeping, the AI bot reading all of these posts will come up with the solution you're looking for.)

Was that Jacks reply or an automated A.I that has dinner?

It's gotten to the point that i don't really believe anything I see online w/o verification! Unfortunately there are bad actors who's sole purpose is to sow dis-information and instability.

AI falls into this category for me. Color me skeptical. The output is only as good as the input which is just vast amounts of data which "bots" or "algorithms" deem worthy of processing in the giant sausage making process. I believe in "math" and "statistics" and "projections" based on trends. Whether or not this is AI or this is just a new buzzword to further confuse and distract the masses.

(insert pic of angry old man with cane here yelling at kids to get off the lawn)

I'm sure ChatGPT will be happy to generate that picture for you Frank!

Hi again!

I agree with Frank that there are several principles that are immutable: garbage in, garbage out; mathematics is behind everything; etc

Under this logic we could well say that the Internet or, more precisely, the WWW is nothing more than the old well-known TCP/IP and a simple product that they elegantly call "browser" but that does nothing more than display files hosted on remote servers.

It seems to me that the point here is how these simple and familiar things were organized to work all together and create something totally different, to the extent that we are communicating through them.

Don't get me wrong. I agree that there is a lot of noise around AI and a lot of opportunists looking to make money by sacrificing the truth. Having said that, what I am saying is that we cannot pretend that nothing is happening and assume that things will be the same as always when there is a risk of drastic changes. I think the bet is very high to be ignored.

Receive greetings from the child who steps on the grass!!

Here is my opinion. I went through the same stages of denial, anger, depreession etc. I couldn't get a believable answer from ChatGPT. It would come up with an answer and I would respond "Thats Wrong" and it woould apologize and then come back with another BS answer. Every answer sounded very confident but was sometimes obviously wrong.

I think it has gotten better though. It answered a very serious probablility question recently that it was clueless on a few months ago. It had to learn to play double deck pinochle to do it. AI provided all of the formulas to prove its work. The odds were 41 trillion to 1 of all 4 players getting a run in diamonds in the same hand. (Like 4 royal flushes) It happened to me and I spent a year trying to find the answer.

There are some things AI was very bad at....
Writing a simple ALPHA-BASIC program
Telling Jokes

But I have used it with great success to do a number of things.
Write various simple comtracts . Non-Compete and Non-Disclosure.
Writing seqments of HTML code (almost flawless)
History questions - Write 500 words on Where the US 106th infantry division was on December 16, 1944? (80th anniversary of the Battle of the Bulge)
Creating a vacation itinerary
Write something and then have AI improve it (Very useful)

Recently Eric Dickerson was crtitical of Saquan Barkley who was about to beat his rushing record but in 17 instead of 16 games. I thought that was rude because Dickerson beat other runners who played fewer games. So I asked.
"NFL running backs with the most rushing yards in a season: Player, year, yards, TDs, Games Played that year. But normalize the yards for shorter seasons."
It returned a table with 7 columns. When you sorted that out OJ Simpsn 1973 and Jim Brown in 1963 in 14 game seasons beat Dickerson.
AI added this explanation: Normalized Yards (16 Games) = (Rushing Yards ÷ Games Played) × 16

So AI is great for certain things and getting better but its answers need to be checked. I use it almost every day to help me with something. You really have to try it.

great replies - Joe i like your football analysis. It goes beyond a simple query of reporting total yards per game. I think with most things its all just advertising. So any new feature in ashell should just be called "AI" - something... and the masses will believe and gobble it up. I am not a salesperson, never was. Have always had trouble saying that we had something when we had didnt clue but this seems like the old 95% hype and the other half substance.

For starters all of the AI services are available via HTTPS API, so you could write your own integrations if you have a feature that AI would make sense.

A lot of the software integrations have not been useful. They seem like a solution looking for a problem and checkbox for website your website (A.I. Check)

I use a combo of ChatGPT and Gemini to test their programming capabilities. They are great at helping with languages you don't know very well or at all. I've used them for Python, Powershell, and Ruby, saved me a lot of time perusing Stack Overflow, or just having to search 100 times looking some esoteric thing I need to do.

We might be able create your own (or a MicroSabio) A-Shell code LLM. I haven't tried to train a model to know how much code you have to give it to be able to get any results.

I agree that you can't trust the output without having someone involved in verifying it, but it can be a timesaver when pointed in the right direction.

Stephen makes a good point that most of the AI services (along with most other Internet-based services) are available via HTTP APIs, so for most applications, the way to tap into AI services would be via XCALL HTTP.

And it probably would be useful if ChatGPT could help you write A-Shell code. I haven't looked into what it would take to train it, but that could be an interesting project. Providing it with a large enough code base, one that is most high quality and contains sufficient in-line comments to explain what it is doing, would be a significant challenge though. A simpler challenge, and one that I'm hoping Ty might someday decide to experiment with, would be to have it train on our documentation and then see how well it can find answers to questions. (I suspect that it might be most useful as a tool to help find weaknesses / inconsistencies / gaps in the documentation. But in theory, it should be able to combine the documentation with code samples, first to explain what they do, and later to generate new code samples based on an objective.) That would all be very interesting to us on the "inside", but I'm not sure it would help sell your application to potential customers for whom AI is the latest must-have acronym in their app. (It has been 30+ years since SQL was in that category, yet even now, only a small percentage of A-Shell apps are interfacing directly with SQL, even when SQL databases are nearly ubiquitous in the typical most larger end-user environments.)

I question the ROI on having AI generate code. I don't see the time savings there after you spend 200% more time cleaning it up and testing.

I would 99% agree with that. But it could be useful for generating example code snippets of how to do something that you aren't familiar with, say, taking some kind of record layout specification and creating the code to generate and bind a DYNSTRUCT that could then read the data and display it meaningfully. (Of course that's assuming that someone else spends the gazillion hours necessary to train it to be able to do something like that.)

Or, perhaps even more apropos, even if it is circular, how about: generate some code to illustrate calling an AI web API to generate some code to ...

Hello!

For now, the idea I have is to integrate AI into current applications, rather than asking it to produce code.

These would be things like optimizing order delivery dates based on some criteria, for example, the lowest production cost, or the dates closest to those desired by the customer, etc. considering already purchased products, production times, machines by which the product has to go through, time between processes, transportation, etc.

Or, based on the repetitive demand for certain materials, optimize the layout of the inventory racks.

Or predictive questions: based on consumption in recent months, what items should I buy to cover future sales orders.

Currently is not clear to me how to provide these criteria to the AI, nor how it would give me the results (if any).

I have not found interfaces for the AI to receive information of this type and return it in a useful format to integrate it back into the system processes.

Besides, it seems difficult to have standard processes since each business is different from others. What is a lot to one client, could be really few to another; what is good to one company could be not good enough to another, etc.

Has anyone searched for something similar? Any ideas you want to share? Is anyone planning to integrate this kind of AI "tools" into current systems?

Thank you in advance and best regards.

Joe's example asking for it to assemble and rank football rushing statistics seems somewhat similar to the kind of thing you're asking for. Of course it was a lot simpler to ask for because the engine already knew about all those statistics. But in theory you could feed it tables of data on production, sales, inventory, etc. and get it to sort through it to find answers to your questions.

But I suspect that for the kinds of questions you're asking, the effort to each it how to answer them (e.g. how to optimize in inventory layout or predict future sales), will be more than the effort to just create the algorithm. Unless it can use information and knowledge it can access on its own without your help (weather, inflation, employment stats, politics, stock market, economic news, etc) and apply it to the problem in a way that you wouldn't have been able to work out for yourself.

As for how to interface to these AI engines, you can start by visiting the Open AI Developer Platform Guide.

believe that AI is a great help with popular programming languages. I knew enough HTML to phrase the question properly. "Can you make a button that will do.." "How do I make the button bigger" "Miove it to the left" "Please would you have it do ....." . I sometimes forgot I was talking to a machine.

Alpha-Basic is a different animal. There is a limited amount of information for AI to reference. I pointed AI to the micro-sabio reference manual. I came to realize that we all have this understandng of the genesis of this language that isn't even on the internet. Its in some Alpha-Micro binder in your basement. Without that foundation Jack's documentation doesn't help.

I remember that I did a progeam to add two numbers together and I had to correct each ine of code. It did "learn" as we went along. It would be like teaching some kid how to program.

But I have learned a lot of members of our little club are more prescient than I am and might see uses for this that I am not seeing. I would ask what our goal would be? How can we make this useful to our customers. So I asked ChatGpt. It did come up with one interesting use.

8. Data Insights
Data Clustering: Organize data into meaningful groups.
Anomaly Detection: Spot unusual patterns in data for proactive action.
Visual Analytics: Use AI to generate dashboards and insights automatically.

So I would just throw this out there. What if we had a new icon in Apex "Ask AI" and the user could ask AI something about the data. Jack should ask AI how to do that and have it write the code.

I decided to ask ChatGPT about how to accomplish last nights idea. "How can I have my program call AI with a question?"

Here was the answer. I never did this before so I hope this link works. It it does work then I think It takes you to my ChatGPT account rather than just that one topic. Maybe we it would be beneficial to open a MircoSabio account? If it does take you to my account then feel free to browse the other topics and glean what you can from my random questions. The AlphaBasic conversation took place in a different AI App.

https://chatgpt.com/share/678d1fab-dee4-8002-96c9-7c6a7b02a6ae

I thought the response was pretty good. It even provided the Python code to do it and offered to write it in another language.

My next thought was that if Jack provided the call then the questions could be gathered using XTEXT. Then that could be sent to AI with an attached document usung this new AI routine.

Wow Joe, I hope you were able to sleep last night! (I asked your chat box how to get AI out of one's head so one can get some sleep at night and it came back with a 9 suggestions, each with its own bullet points.)

I'm not sure whether that link is somehow tied to your account (the page offers the option to login in or sign up for more advanced features), but I'm pretty certain that all of the AI web APIs require an account and charge by some unit of activity (i.e. a fraction of a cent per 'token'). So if A-Shell was going to offer a built-in interface like this, it would probably require some mechanism for the user/developer to plug in their own account info and access password/token.

Then again, I don't see much advantage to accessing a chat box via a web API and paying for it, vs. just doing it through a browser (potentially free), other than being able to avoid having the external browser window. (As you suggest, XTEXT would give you the ability to embed the window directly into A-Shell.)

Where the embedded web API would start to make sense would be if it facilitated queries incorporating the current app context (as with your APEX example where you might ask some it to comment on or manipulate the report being previewed.)

Now we just need to negotiate a deal with one of the AI vendors to give us a kickback on each query and we'll have it made or your activity alone!

I tried to come up with some customers who might use something like that if we built it and didn't come up with enough to justify the effort. We have been using AI 101 (Actual Intelligence) to just write a routine to answer questions like "How much inventory should I keep on hand". Then there are all the comlications you noted. If I were the customer I would export the data and attach it to the AI query. Automating thtat wouldn't really save them anything. So I would think there need to better reason before making the effort.

But just for fun here was the ChatGPT discussion about how to do it.....

How can I have my program call AI with a question?

ChatGPT said: To have your program call AI with a question, you can integrate an AI API into your application. Here's a step-by-step guide:

1. Choose an AI Service
Some popular AI services include:

OpenAI API: For text generation, summarization, and language-related tasks.
Google Cloud AI: For NLP, vision, and translation.
AWS AI Services: For text analysis, translation, and other AI capabilities.
Microsoft Azure AI: Offers cognitive services like speech, vision, and language.
Hugging Face APIs: For specific models like GPT, BERT, etc.

2. Set Up an Account
Sign up for the chosen AI platform.
Obtain API keys or authentication tokens from their developer dashboard.

3. Integrate the API
Example: Using OpenAI's API for a Text-Based Question

Step 1: Install Dependencies If using Python, install the required library:

pip install openai

Step 2: Write the Code

python

import openai

# Set up your API key
openai.api_key = "your-api-key-here"

# Function to ask a question
def ask_ai(question):
response = openai.ChatCompletion.create(
model="gpt-4", # Specify the model
messages=[
{"role": "system", "content": "You are an assistant."},
{"role": "user", "content": question}
]
)
return response['choices'][0]['message']['content']

# Example usage
question = "What is the capital of France?"
answer = ask_ai(question)
print("AI Response:", answer)

4. Handle Responses
Parse the response from the API and format it as needed for your application.
For conversational AIs, maintain context by appending previous messages to the messages list.

5. Secure Your Application
Environment Variables: Store API keys securely, e.g., using .env files or cloud secret managers.
Rate Limits: Monitor usage to avoid exceeding API quotas.
Error Handling: Handle network issues, timeouts, and API rate limits gracefully.

6. Test and Deploy
Test the integration thoroughly to ensure it works as expected.
Deploy the application, ensuring compliance with the AI provider’s usage policies.

7. Optional: Advanced Features
Caching: Cache frequent queries to reduce API calls.
Fine-Tuning: Train custom models for specific use cases (e.g., domain-specific knowledge).
Interactive UI: Create a user-friendly interface to input questions and display responses.

If you share more about your programming language or application type, I can provide a more tailored example!

Nice Joe! Sounds like you would be a great keynote speaker at the next conference!!

Since you mentioned keynote speakers I saw a great one last night and I thought I would share the experience.

Chris and I have been attending the speaker series in downtown Philadelphia, where we've had the privilege of hearing insights from experts across various fields. Last night’s speaker was Oren Etzioni, a prominent AI researcher and the founding CEO of the Allen Institute for AI. You can find a brief biography of him @

https://www.philadelphiaspeakers.org/speakers/oren-etzioni/

Here are some key takeaways:

AI as a Tool for Good
Etzioni emphasized that AI is a tool we should embrace, not fear. Rather than worrying about job displacement, he urged us to focus on how AI can help us, particularly in advancing fields like science, medicine, and technology. For instance, he highlighted AI’s potential to assist the impaired, such as guiding the blind or powering autonomous wheelchairs. Get familiar with it because it is here to stay.

AI is Not Autonomous
Etzioni dispelled the myth that AI acts on its own. It only responds to prompts and commands. As he put it, your autonomous car will take you to Dunkin' Donuts if you tell it to, but it won’t decide on its own to go get donuts.

Responsible Development and Deployment
He advocates for policies and initiatives to help workers transition to new roles as AI reshapes industries. Responsible development is essential to ensure AI benefits humanity as a whole.

Never Fully Trust AI
One of Etzioni's key warnings was, “Never trust AI!” He shared a cautionary tale about a lawyer who used AI to draft a legal brief. The AI cited entirely fictional legal precedents, exposing the risks of relying on AI without proper oversight. When a leading expert in AI tells you not to trust it I think we should pay attention.

My friend who attended the talk with us mentioned his Tai Chi teacher, who uses AI in creative ways. The teacher not only uses AI to improve his classes and help write a book but also uses it to monitor his health. In one instance, AI flagged concerning EKG results, prompting a trip to the emergency room. When he told the doctor who gave the order for him to go to the ER, she paused to savor the moment.

Interesting. Best rule - don't trust anything that is on social media or auto-generated. Which circles back to my point of AI generated code - it you are spending 2x amt of time fact checking then the ROI is diminished.

As long as you know you need to check the important stuff then it will definitely save you time. I found that if it told me something wrong I could point that out and it would correct it. Of course I then had to check that too.

If you didn't check it and there were fictional answers isn't it up to the next guy to fact check it?

Another thing that Etzioni pointed out was that you are already using AI in one form or another. Examples would be Siri and Alexa or any voice recognition software, personalized content on social media, Uber booking, entertainment streaming, advanced driving features, etc.

He also mentioned the prevalence of deep fakes and their use by both sides in politics. His website had a feature to recognize deep fakes like a fact checker might, but I found it was closed recently. There are others out there that I will probably be testing. I think that could be a useful tool.

He did not say when AI would become self aware and start annihilating the human race. That would be good to know.

Another excellent point which i also keep saying - they are calling everything AI now... even though you could ask "Whats the temp supposed to be in Philly for the championship game" and it would already do that w/o all the "hype".

PS: Congrats on the Eagles! PLEASE beat KC!!!