Here we go again; I hope you have studied and are ready to continue reading after understanding.
So far, we have seen What an AI Assistant is, how to power it and how it is made .
We concluded the previous chapter by leaving you with a minimum of suspense on these points that we will see today:
Adopt turnkey solutions created for this purpose by large or small emerging companies.
Build a custom solution.
The 'basic' minimum characteristics of a solution of this type are:
Possibility of choosing the AI model to use
Loading and transforming your data to match that in the model
Vendor reliability and data security. At 360°.
Availability of Analytics to analyze interactions
Ability to manage and segment users and the information they have access to
Keep this in mind while reading!
1) Turnkey AI solutions
The Chat GPT advent has created an explosion of new fauna in this industry: hundreds of new solutions have been released in the last year, especially from small startups.
They are typically cloud-based solutions with a web interface on which you can configure your AI Assistant, provide your data, and do the first tests. Often within minutes.
These solutions allow you to implement solutions based on your knowledge base extremely quickly and without you having to have technical knowledge.
More or less all of them will enable you to:
create an AI Assistant
give a basic prompt to guide the user's search
Attach documents in various formats: TXT, CSV, JSON, PDF (But you remember what we said last time, right?), DOCX, XLS, etc., or upload an entire existing website.
Customize the look and feel of your AI Assistant to bring it to your website.
And with a more or less high monthly amount, you can get a good solution quickly and without spending too much money.
You can find a few hundred examples here or here (and you will often still see it defined as Bot, but now you know it is much more 🙂). Around 300 different solutions are provided by more or less large companies, so there is no shortage of choice.
Which ones do I like the most?
To date, GPT Trainer and Custom GPT (and I don't collaborate with them! 🙂)
Many turnkey solutions do not work with proprietary AI models but mainly use OPENAI and the other Foundation Models available today, such as Mistral or LLaMa. Often, they will ask for your API KEY so you can pay credits directly to Open AI or Google (which is a choice I recommend)
Points of attention:
Lock-in with the manufacturer - Be sure to evaluate the possible dependence on a single supplier and the long-term consequences. We often talk about startups that could suddenly disappear.
Manufacturer reliability - Investigate the strength and reputation of the supplier to ensure a reliable partnership.
Lock-in with the AI model - Consider flexibility in changing or updating the AI model in the future.
Data Privacy - Strictly monitor data security and privacy policies to ensure compliance and protection.
Multimodality - (The ability to handle input and output of text, images, video, and audio) is not always possible - Check if the solution supports various communication formats to meet your specific needs. If you have any. But if you want to start with a simple project, I suggest you postpone this part again.
And then… all the other boring but essential things like the ability to integrate with your systems, scalability, availability of adequate documentation and support, and, of course, the costs, especially the hidden ones!
And the Big Players?
OPEN AI is now reigning supreme with its GPTs and Assistants entering the market. I talked about it extensively in two posts that I recommend you read.
Would I use GPTs for production purposes in the company?
NO!
You can then use them internally for your prototypes; they are exceptional in this, but I don't recommend them for an 'extended' AI Assistant project that needs to go into public production because:
Require the end user to subscribe to GPT-Plus and you to work only with OPENAI.
Accessibility is difficult to control (You cannot decide who accesses what/when)
To date, they do not have Analytics data that allows you to understand if and how they are used.
Anthropic or Google have not yet released 'ready to use' solutions, although both are working hard. (Ps. Claude 3 just came out... but I don't have time to tell you about it now!)
On the other hand, Microsoft is releasing a ton of different and constantly solving things and is shaping up to be the partner of choice for the enterprise world. It has released Copilot Studio and, more recently, Azure AI Studio, advanced and complex solutions that allow, in addition to levels of customization of the interface and behavior of the AI Assistant, a level-to-date unparalleled level of customization of data access rules based on the users who use it.
Not satisfied with all this Microsoft also released Copilot for Finance and Copilot for Sales a few days ago plus, they announced its agreement with Mistral, thus providing easy access even to Models other than OPEN AI.
Unlike the others, unless your IT team is quite strong, they require the intervention of a system integrator to be adopted effectively. I don't mention them here so as not to fill up the whole post but... indeed whether you are working on turnkey or custom solutions, it becomes essential, to develop structured projects, to work with partners who can support you.
Huggingface, a platform specialized in hosting AI models of all kinds, also came out with HuggingChat, a solution that allows you to create your assistants, albeit with many limitations on all fronts (for example, it does not allow you to upload your data, to have analytics, to take actions).
2) Build a custom AI solution.
If this is a strategic project for your company, you can choose the 'homemade' route. That is, create a chatbot "from scratch" using one or more LLMs among the thousands available ( which you can find here ), train it as you like, host it on the servers you prefer, develop it from scratch - or starting from an open-weight solution, by there is still minimal open source - your complete solution.
Did I scare you? You should have some!
To master this type of solution, you will need a highly respectable team, and you will come across solutions that, despite being much more controllable, have decidedly different costs, as seen from this graph.
So? Evaluate the customized solution for Phase 2 and get some experience first!
Points of attention:
Implementation costs - Besides initial developments, consider ongoing maintenance and upgrade costs.
Implementation Time – Creating a custom chatbot can take months, depending on the complexity.
Design Complexity – Managing the integration of various components and ensuring the system functions harmoniously can be a challenge.
Skilled Human Resources – You will need developers, software engineers, and AI experts to build and maintain your chatbot.
Data Security - Protecting user information and ensuring compliance with data privacy regulations is critical.
Certainty of results - I have already seen such projects fail because of their complexity and outputs below expectations. Handle them with care!
As well as the boring but essential points seen before :-)
Homework (yes, this time you have to do it)
I suggest you take an in-depth look at the links you find in these posts and understand what solutions there are and what they can do. Try putting them to the test with some data and see the results.
If you are starting out, moreover, I suggest that in addition to a good workshop, you get some practice with Open AI's GPTs: they are simple and included in your Plus or Teams subscription and will give you a better understanding of what you need.
And as they say in cycling: "Come on, we are almost there!"
📢 If you have thoughts or comments or want to help spread these reflections, share this page with anyone you think might appreciate it. Your opinion matters a lot!
🚀 To always keep you updated on my content:
📝 Subscribe to the Glimpse Blog to never miss an update
📚 Check out 'Glimpse,' my novel about artificial intelligence
🗓️ Contact me if you want to organize an AI Workshop or for any ideas.
See you soon!
Massimiliano
留言