The journey of analysis on the selection and hiring of an AI in your company continues, in the previous post a bit of analysis was made on the types of AI-based tools and on the expectations that it is good to have before leaving. However, choosing the perfect role for your AI from a moltitude of variables can be a complex challenge.
In fact, we should consider AI a new type of resource: neither human nor purely software or hardware, and it is important to understand how to introduce it into the gears of our complex business organization.
The issue is significant because I imagine that none of you have ever hired a person without considering the AI's characteristics, skills and the appropriate department to which they should be introduced.
I don't think anyone has adopted software without evaluating the value of the investment, the ROI (Return On Investment), the Opex (Operating Expenses), and the actual utility.
I don't think anyone has hired a person without thinking about what task to assign her.
So let's continue the journey in this phase which I warn you, will necessarily be a bit didactic because you firstly need to understand which AI fits your needs.
There is no one-size-fits-all solution.
With that being said, in the second part, I would like to focus on the fact that AI, especially Generative AI, is not a one-size-fits-all solution. There isn't a single AI that is suitable for all tasks and generalizing in such cases does not provide helpful results.
Small lament...
For example, asking LLMs to do mathematical operations is possible, but we take on the risk of errors and use a lot of unnecessary resources to perform tasks that any PC is able to do without any effort. Asking Advanced Data Analysis of Open AI (The Old Code Interpreter) to write an entire software on its own will not lead to a great result.
So let's see how to proceed, taking into account that, before starting seriously, it is important to zoom in more precisely.
A bit of classification of the AI world
Without getting lost in complex taxonomic classifications, I would like to provide you with my interpretation of very result-oriented classification with the intent to understand what we can do TODAY with AI in your Company.
So let's continue on the path of the role to be given to AI.
Before leaving, let's just remember that we are still in the world of ANI (Artificial Narrow Intelligence). When AGI (Artificial General Intelligence) arrives, I'll have her write a treatise on how AI should hire humans.
I've split what I think is BEST to do with LLM-based AI, using this evocative 🧠 symbol and that can be done with different🎓 AI technologies.
Surely I will not find everyone in agreement: integrate your thoughts in the comments, especially if I have forgotten some. The debate is open.
🧠Processes that use LLMs
The systems that use LLMs are based on Natural Language Process (NLP) 📘✒️ logic, i.e. the ability of models to understand and be understood.
NLP includes two main worlds: NLU and NLG. From these start all the subsequent text-based Generative AIs.
Natural Language Understanding (NLU) 📘
It focuses on understanding the meaning, feeling, and intent behind the words. (Whenever a "GPT Chat" includes you, it's because of this.)
Natural Language Generation (NLG) ✒️
For the production, which we like so much, of coherent and contextually relevant sentences or paragraphs. (The Great Success of Generative AI)
So what do they can do?
1. Question Answering ❓🅰️
Answers questions based on information provided or knowledge learned. ( Included hallucinations if you expect a bad prompt to provide a good result.)
2. Summary 📋
Creation of concise summaries from longer texts. (To allow the worst students to cheat at school and for the best to draw conclusions that were unexpected for them to elaborate on further.)
3. Translation 🌍🔄
Well done translation of texts from one language to another. (It is said that no one taught languages at LLMs but learned them themselves. But that's just a legend.)
4. Sentiment Analysis 🙂🙁
Determination of the feeling or emotion behind a piece of text. (Think about the interpretation of some cryptic emails you receive. For the record, here you can also work with technologies other than LLMs, but maybe it's not worth it anymore.)
5. Code Generation 💻⌨️
Writing or hinting code based on instructions or prompts. (Programmers, especially good ones, will continue to serve always. And they will use these features more and more. But you can begin to understand them and speak their language.)
6. Conversational Agents 💬🤖
Participation in natural conversations with users. (GPT chat and Claude are the most striking examples, they "simply" mix NLP and NLG phases in a dedicated UI while keeping them connected in a context window.)
7. Completing Content 📝➡️
Completing or suggesting the continuation of certain content. (So goes the cat to the lard ... )
8. Didactics / Tutoring 📚
Assistance in educational tasks, providing explanations or answers. Both in school and business. (Here the theme will be increasingly important and intriguing. See also this post dedicated to education to learn more.)
9. Pattern Recognition 🔍🔄
Recognition and understanding of patterns in data. (One of my favorite features. In simple terms: understand how the data behaves even without seeing it represented in the graphs. Non-LLM algorithms also excel at this task.)
10. Character and Role 🎭
Simulation Taking on specific characters or roles in conversational scenarios. (Act as... One of the first commands to put in a prompt.)
11. Text Classification 📂🔖
Categorization of text into predefined groups. (For example, determine whether an email contains requests for information or compensation.)
12. Entity Recognition 🏷️
Identification and classification of entities named in the text, such as names, places and dates. (Very useful feature for structuring datasets that are not. Better if accompanied by specialized software already existing.)
13. Topic Modelling 🌐📊
Discovery of recurring topics in a set of documents. (For example a targeted press review?)
14. Text-to-speech 🗣️🎙️ Recognition of human-like speech from text. (And we're not just talking about tone of voice but with any timbre, height, intensity, volume.)
🎓Processes using technologies other than LLMs
1. Forecasts and Analysis 📈
Prediction of results and identification of insights through statistical and Machine Learning models.
2. Computer Vision 👀
Image and video processing for the recognition of objects, people, activities. Useful in surveillance, autonomous driving, robotics.
3. Anomaly detection 🚨
Identification of anomalous behaviors and data that differ from the norm. To detect fraud and breakdowns.
4. Simulation 🕹️
Realistic simulation of complex systems such as epidemics, social dynamics, physics.
5. Research and Planning 📜
Find optimal paths and action plans in environments such as logistics, video games, design.
6. Robotics 🤖
Movement planning and manipulation of objects for industrial automation and service.
Is it all clear? Otherwise let me know what you want to deepen.
I know this post will be one of the hardest. If you want, it's time to take a break before resuming and figure out which of these AIs makes sense to hire, and for what tasks.
Which AI makes sense to hire to get started?
Would you hire a programmer to handle client communication? Or a salesman in accounting? And what about a manager for simple routine work?
Ok, our AI assistant is almost omnipotent as seen in the previous post but we must place it in the right place so that it can work well. He can do many things, he is very versatile, but to make him do too many things would be a mistake.
We must first understand in which department to make him work.
I will try to be schematic to stimulate you to think about your company not wanting, and not being able to give an answer here to all the dilemmas you are already experiencing: every company has its own business context, its culture, its reasons, and yes, I know, it is different from all the others. 🙂
For simplicity I analyzed four macro-types of companies and I hypothesized that in these companies there are some main departments that perform their processes.
So I got a table like the one below. I hope you agree.
For ease of reading in the table I have inserted only an indication of the macro-type of AI suggested for each department:
🧠 Large Language Models
and/or
🎓 Other types of AI.
I also added an indication of the priority ⭐ in some cases to indicate a possible start of an AI hiring journey in those areas.
I ask you to look at the following table not as a limit but as a stimulus to think about your company. Remember, the best approach depends on your business' unique qualities and goals.
As you can see there are areas in which it is worthwhile NOT to work with Generative AI and others that leave no doubt.
To go even further into the merits I propose, for each of the priority areas, some process ideas to be developed if your imagination has not already begun to grind ideas.
Here, too, you notice that LLMs are not the best solution for every approach. So those are my suggestions:
Production Companies
Commercial Companies
Service Companies
Financial Companies
So what...
Building on the previous post, we've learned the importance that:
the top management of the company understand what is happening by putting their boots on the ground;
it makes sense to spend bandwidth only on solutions that reduce costs or increase revenues;
Let it be clear that without clean data and defined processes very little can be done;
are ready to start with paradigm shift to AI implementation;
identify small projects to solve small pain points with which to start,
the distinction between Chat Interfaces, Specialized Solutions, API Usage, RAG and Fine Tuning is clear.
Yes, only this for now 🙂
But today we've also learned that is very important to recognize AI as a novel resource, distinct from human or tech.
Focusing on how to dominate the space of this powerful new emerging technology will help us to place them in the right roles, in the right projects, and obtain the best results.
Most of those who know its characteristics are moving very quickly in this way: I interact with a lot of entrepreneurs from all over the world that every day try to leverage AI to augment every aspect of their companies, but I often see that there is too much haste, or greed, that risks defocusing and leading to disappointing results.
So: if you need something "good enough" but not PERFECT then you can consider using generative AI via LLM. Alternatively, you can evaluate other types of AI whose job is to "work with numbers" and provide more precise results.
You have to add the final touch, but first you will have to spend time learning this new world.
"...it’s not so much what you know anymore that counts, because often what you know is old. It is how fast you learn. That skill is priceless." Robert T. Kiyosaki
If you have come this far I need important feedback. Just one click, thank you.
The post on this blog are
Too long
They're fine
Too short
As always, I invite you to reflect, to comment, to spread ideas by sharing this post with the people you think might be interested.
To stay up to date on my content:
Read Glimpse, my novel about AI.
Or contact me here.
See you next time!
Massimiliano Turazzini
Disclaimer: I used Claude and Chat GPT to get help in this post (And obviously Midjourney for Images). The merit for the choice of emojis and descriptions of the different types of AI is therefore to be attributed to these two subjects. The grammatical and syntax errors I made them all 🙂
Comments