AI Copilots Everywhere; GPT's Trillion Dollar Economy

In the news
16

GPT is not perfect, but humans are flawed as well. And let’s be clear; these examples are super simple and intended only to help frame a general understanding given that you and @RonaldGRuckus use only customer-facing GPT features. I recommend you give it a try - they offer a free trial period as I recall.

The question I believe deserves consideration is whether an AI paired-assistant is likely to improve things. Or make them worse. So far, the data says it’s better](Research: quantifying GitHub Copilot’s impact on developer productivity and happiness | The GitHub Blog). But, each developer needs to really assess that in their own work style.

We can debate how specific copilots may work well or poorly, but my assertions in my post are about copilots in general. I believe they will begin to be proven effective in almost every aspect of business, communications, marketing, sales, research, management, and product development. I think they’re going to change pretty much everything.

17

I think that’s a fair objection. Uninformed readers will make assumptions that may be a stretch for today’s AI models to handle. But isn’t that what a copilot is all about? Shaping the prompts and layering API calls under the covers to achieve that which is very difficult to achieve in a free-form UI?

The first “copilot” was ostensibly the original Microsoft wizard, which was first introduced in Microsoft Publisher in 1991. It was part of the operating system in Microsoft’s Windows 95. The most commonly-used wizard at the time was the Internet Connection Wizard, which was renamed to the “New Connection Wizard” in later versions of Microsoft Windows. But the concept was the same - try to help users compress time and complexity.

LLMs are transcendent; they raise the bar significantly because they can infer meaning through natural language, and they can generate content with meaningful, though, not perfect precision. If you’re old enough, you may recall that Internet Connection Wizard had flaws as well. :wink:

18

It really depends. Optimizing a third-party prompt from what I’ve seen requires some assumptions that can drift away from the original intent. There needs to be a certain level of understanding to write a prompt in a subject without it essentially being noise, or wrong.

There’s so many nuances in building an application, it would almost be the same amount of time to format it for Copilot, as opposed to just writing it myself. As shown in your example, Copilot just “does”. There’s a layer of thought that’s needed which is still absent.

In terms of programming you almost need to already know the architecture, or the end result to actually prompt it correctly. Not all programs need some overly thought out and designed process, obviously. As a learning tool it is wonderful. I absolutely love GPT’s potential for global education.

However for anyone who is starting out and picks it up thinking they can “talk their way to an app” is grossly misinformed.

Again, I completely love using it, I am just disappointed in their statement.

19

I’m sure there are cases where the net gain is zero, even negative. The data, however, indicates overwhelming evidence that CoPilot users are experiencing significant net gains.

20

No, for sure. Again, I don’t use Copilot but I use GPT and it helps me tremendously.

What I am trying to say that in almost no current setting would talking their way to building an app be a good idea. It’s just setting the stage for disappointment.

I’m totally aware it’s named “Copilot” for a reason. It’s just slightly frustrating seeing claims of it being capable of essentially writing out an app through words.

21

This doesn’t seem to suggest what you’re seeing in terms of app development claims.

CleanShot 2023-03-17 at 15.26.57@2x
CleanShot 2023-03-17 at 15.26.57@2x1920×1908 169 KB

22

Yes, in most cases they make it obvious that it’s only intended as an assistant. Again, this was all originally based on the livestream that I watched, and then my attempt using a different framework and concept. I haven’t used Copilot and can’t say much about it.

I do know that Copilot uses GPT, so anything I say is mainly based off my experience with it; which is: it’s a very powerful tool, but can be misused completely, especially with an incorrect understanding and misrepresentation on how it works. So comments such as “building apps through a conversation” can be misleading.

23

Ah, okay. We’re on two very different pages.

Copilot for Microsoft Office 365 was announced Tuesday. It is in the hands of 20 enterprises and it uses a very advance collection of prompts and integrations with OpenAI that only Microsoft has access to. They are able to make this claim because their definition of an app based on PowerApp is deeply constrained.

If you had experience building apps with PowerApp, you would know that this is entirely possible with Microsoft’s variant of the term “copilot”. There are wizards and other frameworks that will do this much the way a no-code app works. PowerApps have a limited ceiling and scope of complexity making it an ideal tool to end-cap with an AI wizard.

I believe they will be very successful with this approach, and customers will be far more satisfied than you might assume, especially having never seen it work. One of their test enterprises is my client. They have shared that it truly does build the essential elements of a complete solution.

Is there marketing overreach in their claims? Probably; Microsoft is famous for this. But whatever their claims may be, customers will lap this goodness up in spades and their O365 copilot variant will likely save enterprises millions and win awards.

24

I have absolutely no experience with Copilot…

I have no doubt it’s wonderful technology and helps a lot of people. I just don’t like these types of claims even if it’s exaggerated for the audience. I’m an old man yelling at the clouds, I know.

I do however have quite a bit of experience with GPT and anything I say is based off of that

25

As the title of this post indicates, there will be thousands, perhaps millions of “copilots”; you need to add a predicate when using this term now that there are at least two publicly available “copilots”. ;-).

BTW, I’ve built 21 copilots, and 18 are in production. None of them is intended for anything except internal business use. Each one has a discrete name, and they help domain experts perform many different tasks.

26

I imagine for a lot of cases it is a perfect match.

When I think of application, I think of the whole system. Networking, security, blah blah balh. Which would probably be unnecessary for simple use. So it is a bit of a kick to the shin to see that type of claim when it doesn’t even program well for me in most cases simply by talking, but that’s just my experience with GPT. I understand that Copilot is its own beast

27

Interesting. What domains and expertise? Would love to hear more.

Regarding Copilots. I am not sure if there will be “thousands” or “millions” of them as my crystal ball is as foggy as yours. But I can relate my own experience with Copilot for Visual Studio Code, which I have used nearly daily for a few months.

Screenshot 2023-03-18 at 8.57.49 AM
Screenshot 2023-03-18 at 8.57.49 AM842×390 44.7 KB

I subscribed to VSC Copilot a few months ago and used Copilot nearly daily when writing commercial (paid) Ruby code and sometimes Python as well. It really saved me a lot of keystrokes.

Ofter time, the code completions suggests were a mixed bag. Some were “dead on” and really saved me a lot of time and typos, and others were lines of annoying code-spam. Sometimes I miskeyed and included nonsense companions suggestions which I had to go back and control z which was annoying.

So I had mixed feelings.

Then GitHub notified me that my CoPilot subscription was about to expire and I decided to not renew at the $10 a month charge and let it expire.

Soon after, when coding all those sometimes “very good” and sometimes “very goofy” code completions disappeared (my trial fee subscription expired) and I missed my goofy friend. I realized that I found Copilot very useful at best and a little annoying at worst and even when annoying, often entertaining.

It was like a little bird was coding with me, coming up with ideas, sometime great, sometimes nonsense, but none the less, without the little bird code completions, I felt a bit lonely coding, haha.

So, I reversed my decision and paid the $10 buck a month to keep this companion coder in VSC.

It’s funny to think that I actually “missed” the often goofy auto-completions, the many lines of nonsense; but it’s fun to see the "nonsense’ CoPilot comes up with a times and of course, when it’s “dead on” it’s really, really a productivity booster.

So, It’s easy to image these types of specialized Copilots in all domains, legal, medical, professional writing, marketing, finance and yes the specialization probabilities seem endless.

I hope you @bill.french will post more tech details on the copilots you created and the code,

:slight_smile:

28

Great post; your experience parallels mine and why I tend to ask sceptics - would you feel better, perhaps more confident doing this engineering stuff alone? :wink:

When non-coders get a taste of a copilot experience designed for a specific workflow or information-related task, they will adopt similar sentiments. It is very useful at best, and a little annoying at worst. The net gain will be significant and measurable. They will want more copilots in their work and also in their lives. Ergo - copilots everywhere.

I’ll share two…

Cost Recovery Processing

In a little office in South Carolina, an “agent” is assimilating data from a motor vehicle accident captured by a fire department. It includes a narrative of the accident scene, images of property damage, injuries, and, most importantly, the time, equipment, and materials used to clean up the accident. This information needs to be compiled into a claim and submitted to the drivers’ insurance companies to compensate municipalities for their time and expenses. My “copilot” assists the agent with a variety of time-saving helpers to extract entities from the fire chief’s narrative, entities and data values from the police report,keywords from all of the data, and map geometries from the location address or intersection. The agents doing this work are about 22% more efficient since we implemented this copilot approach. Even in this very early stage and as simplistic as these helpers are, the workers cannot seem to live without them.

Validating AI Performance

I work with a video analytics company that creates real-time highway analytics. The technology uses advanced global shutter cameras and edge devices capable of performing AI to classify, count, and even determine HOV occupancy for vehicles at high speed. This is cutting-edge AI, of course. But there’s another side of this solution that most people assume is 100% manual human effort - validating AI results. Someone has to examine the video to understand and log AI performance. This copilot assists the validation process by using GPT to identify ideal sampling periods, creating reporting narratives from data, and even building key metrics that help the AI team tune their models to improve performance. It compresses the time humans must endure explaining where the AI system is failing or succeeding. It’s a more optimized feedback loop that accelerates knowledge about system performance. Oddly, this copilot is built in Coda.

1 Like
29

As I’ve suggested … AI copilots are just beginning and will be everywhere.

30

I’m having a bit of a problem following your generous use the phrase of “millions of copilots” without more context, to be honest @bill.french .

Do you means millions of people will use copilot ? Of course, this is already happening and this is easy to agree with.

Or do you mean, that you predict there will be millions of different copilot software applications (like the VSC Copilot extension) and these individual copilot software applications and projects will be actively maintained and with a user base?

:slight_smile:

31

That’s what I’m saying. If someone like me - a consultant - is being asked to create a variety of “copilots” (lowercase) for specific domains of work, what can we glean from this early demand and adjacent market observations?

  • Dan Shipper is frequently suggesting where and how copilots (lowercase) will probably emerge.
  • Many of the demos in this community are ostensibly copilots; they serve specific work tasks with GPT.
  • Bardeen is a no-code platform for building AI-enabled data collection processes to serve specific business and professional work tasks. It’s user base is asking for playbooks that utilize GPT to streamline work tasks. Playbooks are ostensibly copilots.
  • Microsoft has demonstrated about a dozen different copilots in its O365 vision; there’s likely to be another dozen before they launch.

We’re in the early days of copilots and the definition is not clear or rigid. Where do you think we’ll be in a year? Three years? Five years? I think it’s safe to predict that copilots of varying complexity and usefulness will be required for every unique work objective to accommodate our demand for AI-assisted productivity.

32

I’ll admit. I am currently trying out visual code with Copilot - might as well. I was using JetBrains for basically my whole toolset so it’s slightly refreshing.

Are you referring to this (GitHub) Copilot, or the Microsoft 365 Copilot? I am slightly confused.

33

This article started by referring only to Microsoft’s new (unreleased) Office 365 Copilot as an example of the rapidly expanding notion of paired AI assistants.

When the comments drifted into another copilot named “Copilot” (from Gihub), which is purposed only for code development, we started referring to that copilot as well.

This is why I tried to suggest we be very clear about the predicates used to describe specific “copilots”. A year ago, there was one Copilot from Github. Today, there are many, and each day this number grows. It seems that the idea of a copilot is deeply wanted by many users and for many domains of work - not just coding. As such, the term Copilot is ambiguous and will remain ambiguous from this moment forward.

There was a time when “car” meant Model T. When a friend asks what you drive, a “car” is not a good answer. It may be whimsical or flippant to respond this way, but it is not an answer that carries much meaning.

1 Like
34

Ah, right. I see.

It’s very confusing. It seems that the phrase “Copilot” is being thrown around as a generic term for AI assistance.

I was initially under the impression that it was the same thing. After looking into it more I see how it is and must say, it all looks exciting. I currently use Google Drive for all my business documents. It’s basically already at the point where I should be able to say “Write a contract using the numbers from this spreadsheet” and probably just some few-shots would suffice. Just imagining the potential is huge. How do you bundle the service and sell it as your own if you don’t mind me asking?

This unreleased version I’m assuming you have early access to?

35

Indeed, much the way Xerox has been bandied about. :wink:

I assume you are referring to the Microsoft O365 Copilot product. No access yet, but I do have a client (one of twenty testing enterprises) who is using it. And they’re deeply bound by an NDA, of course.

Yes, the dots to make that connection are certainly within reach with LLMs.

As you would any other SaaS service. I am building copilots only under a professional service agreement presently. I need many more miles of experience before I might throw down on a specific offering if at all.

1 Like