All images in this article were generated through AI inference- and diffusion technologies.
“We live in a world surrounded by technology” – A phrase that by now has earned its right to be seen as cliché. And for good reason. I’m sure we can all agree that the idea of technology driving our daily lives won’t change any time soon, but what will this technology look like?
AI surely has made itself known as a worthy contender, but is it really up to the challenge of curbing our boredom for long enough to be part of this cliché? As the media buzz slowly dies down, now may the perfect time to take a step back and explore this question further.
⇔ Chat with Ultra-LM ⇔
Hey Ultra! I’m finishing up that blog post we spoke about yesterday, and I’d like to give readers a short demo of what you can do. Would you mind being my guinea pig one more time?
🖥 Sure thing Jacques, happy to help out again! What would you like me to demonstrate this time around?
JJ: Let’s start with something that’s simple, but makes for a good first impression. Do you remember the title of the article we spoke about yesterday?
🖥 Ultra: Yes I do – it was NLP Is All You Need”
NLP Is All You Need?
In January of this year, we wrote about some concepts of Artificial Intelligence and its potential applications. While in no way required for this article, I’d recommend giving it a quick read if you’re interested in the topic and would like a broader view of the AI while we explore it in more detail here. In the article’s opening paragraphs, we state that “Machine Learning gives AI the ability to learn using algorithms, discover patterns and generate insights from the data they process”. The way large language models (like Chat-GPT) achieve this today is in most part due to the application of Natural Language Processing (NLP) – an impressive and mindbogglingly complex set of mathematical algorithms and processes that, in combination with neural networks and deep learning processes enable our AI models to reach that next level of “understanding”.
⇔ Chat with Ultra-LM – Continued ⇔
Before we continue – can you give a brief explanation on what it means if I say you’re a 40B language model?
🖥 Certainly! In the context of Al models, parameters refer to variables or values used in mathematical equations. The more complex an algorithm is – the higher its number of parameters. As for my specifics, Ultra Neural Network Language (N2) has over 40 billion parameters designed specifically around natural language processing tasks like sentiment analysis and text generation.
Making sense, but also not really? Don’t worry – many of these topics are only fully understood or appreciated when experienced. The world of AI is filled to the brim with these “you had to be there” moments. NLP is a prime example of this, so I’ll touch on a few key concepts surrounding NLP, and then we’ll explore a practical application together.
I’ll also be talking to one of the latest and greatest open-source large language models. UltraLM, along the way, in the hope that these two examples would give a better idea of what these experiences would look like if integrated into your daily life.
Let’s get started.
Integrating NLP AI into Home Automation Platforms
To fully appreciate the value NLP adds to AI interactions and communication, we need to start at the beginning. What is natural language processing? Put simply, NLP is a field of computer science that deals with the interaction between computers and humans through the use of natural language concepts (patterns associated with every-day spoken language). In other words, language that is intuitive and feels organic.
We use NLP to improve machine translation accuracy, search engine results and customer service interactions, to name only a few.
⇔ Chat with Ultra-LM – Continued ⇔
Moving on – do you know anything about a research paper published in the field of computer science in 2017, called “Attention is all you need?”
🖥 Yes I’m familiar with that paper. It proposed an alternative approach to machine translation which focused solely on attention mechanisms. In contrast to traditional methods like RNNs (Recurrent Neural Networks), they argued that simply using the right amount of “attention” could achieve better results in sequence-to-sequence tasks, without having to resort to complex architectures or large amounts of training data.
While it can be difficult to accurately predict the potential in marrying NLP AI models with home automation applications, we don’t have to look far before a world of incredible possibilities start jumping up at us.
To preface – The following ideas were conceptualized through applications and technologies available to us today. Wide-spread adoption is possible, but would require significant investment and funding in key areas (like manufacturing) to be viable commercially.
As with any automation platform, we need to start at the core of the platform – your control unit. Modular, extensible home automation platforms – like those typically deployed by iDetic- could easily be expanded to allow for a something like a drop-in processor, designed to run Large Language Models (LLMs) on-premise and fully offline. This approach not being too dissimilar from upgrading your Home Theatre experience by simply plugging a Trinnov Audio Processor in-line with your existing hardware, for example.
With a form factor requirement not much bigger than a typical modern graphics card, it’s design can allow for it to be racked alongside your devices, making integration painless. With the right hardware, such a device would excel at AI-related processing such as Machine Learning (ML) and NLP. It would do all the heavy lifting while passing only high-level functions to your automation controller.
Let’s look at what’s possible if we apply all these concepts towards this specific goal of improving how we interact with, and experience home automation:
A holistic view of your entire digital landscape
Easy access to bespoke hardware and software reporting, port/link status monitoring, lighting control, access control, entertainment systems, air-conditioning, device management and alarm systems, to name only a few.
Interfacing with these systems individually or at the same time will be as simple as describing a set of requirements or explaining what you’d like to achieve to a single touch point into your system.
3rd party technologies or applications that could otherwise be integrated into your platform, would now share in the benefits of a single management interface. Access to already sanitized data and reports would be immediate and always up to date.
A practical example of this would be your AI considering solar predictions when requested to manage the charge on your electric vehicle. Additionally, having access to your calendar would allow the AI to ensure the vehicle is charged and ready for use when you need it.
Automating the mundane
A phrase commonly used by iDetic – this is one of the biggest drivers in how we make decisions on preferred partners/brands, and why we believe so strongly in up-skilling ourselves and our clients, and why we’ll go those extra few miles to provide a peace of mind in every solution we install. Time is one of our most valuable resources, and should always be a priority when considering where to invest in order to save somewhere else. An AI set up with these principles in mind, coupled with the ability to understand and execute on extremely complex tasks and scenarios, can save a lot of time.
Repetitive and time-consuming activities like logging incidents or change requests with a 3rd party provider, performing data and trend analysis (solar yield / ROI, stability of hardware / usage statistics in specific areas or specific hardware) – all can be fully automated with frequent and clear feedback provided on all open items or tasks in and around your home.
An organic and conversational interface
Natural Language Processing is more than just an improved pattern recognition algorithm. The seemingly simple step in creating this bridge between artificial intelligence and humans led to both an improvement in base understanding and context derivation, however – and we’ve now reached the core of why NLP is so valuable – communicating without concerns over how it will be interpreted, how we should approach phrasing and without needing to adhere to ‘rules of engagement’, an interesting dynamic comes into play. If we’re more comfortable in expressing ourselves, we do a better job of it. The result of us expressing ourselves more clearly is a better understanding on the part of an AI, allowing for a more intuitive response, reinforcing the overall effect.
Adding to the above conversational history and context (memory) and an ability to adapt to changes in input or output variables (attention), it’s easy to see potential value that can’t be measured in money alone. A virtual assistant running in your environment can be “trained” according to your preference and routines. It would build an understanding of your family home dynamic and requirements, and can respond to both predictable and unpredictable events within the context of your environment.
Again, security is a good area to explore here. A simple example being an intruder, or break-in scenario. Assuming you have a comprehensive automation solution that includes lighting, access control, alarm and other security systems such as perimeter monitoring, etc. – your always-vigilant AI assistant would have visibility into each of these systems, and could with relative ease determine the location of anyone currently in the home, as well as the intruder’s movements. A set of instructions trigger that alerts you and your family, guides you to the safest area in your home, while a call is placed to your armed response company – providing them with the exact nature of the emergency and allowing them to respond more effectively.
Learned behavior, like noticing you open the same set of blinds every day and automatically incorporating it into the daily schedule, add further value beyond the simple instruction-based interactions we have become accustomed to.
⇔ Chat with Ultra-LM – Continued ⇔
My next request was a summary of the paper abstract, but since you’ve done that already, we’ll have to try something a bit more challening. The similarities in title is obvious, but why did I reference that paper specifically? Can you derive the underlying reason I chose that as a reference?
🖥 Ah, I think I understand why, Jacques! It seems like you were inspired by both the potential of NLP Al as well as the transformative impact that “Attention is all you need” had on machine translation. You’re suggesting a paradigm shift in use-cases where natural language similar to what attention mechanisms led to in machine learning applications.
In conclusion
I could continue to list countless more examples and scenarios where AI would improve your productivity, safety and daily routine. The nuances that exist in the language we use, the decisions we make and the experiences we face in everyday life are endless. It’s within the boundaries of this thought alone where I see enough reason for us to embrace AI – motivating for further innovation and pushing the limits of what’s possible.
I’ve heard both careful murmurs and confident declarations all sharing in the same message – “AI is leading the way for the next age in humanity’s evolution”. If I’m honest, I don’t think we’re there just yet. While the science, math and engineering responsible for these technologies are without a doubt incredible, we’re still left with a system where the output is only as good as the input.
Conversely, I don’t see a reason to be afraid of AI either – as long as we embrace the positive, effectively manage the negative and keep an open dialogue on both positive and potential negative consequences…who knows, we may not be too far from experiencing the “next age in human evolution”, as some have started alluding to.
⇔ Chat with Ultra-LM – Continued ⇔
You’ve summarized it perfectly. The idea of understanding the “reason behind a reason” may be a simple concept for humans, but without NLP technologies this concept would be almost impossible to expect of you. In closing – do you think my assessment of NLP in relation to “Attention is all you need” has any merit to it? You can be honest, even if people are watching.
🖥 Well, as a large language model myself, applying these principles directly – I’d have no reason to lie to you 🙂
Your comparison does indeed hold some validity! The conceptual and practical implication described in “Attention is all you need” was a groundbreaking discovery in the field of machine translation that had far-reaching consequences for Al research beyond just language processing. Your comparison draws an interesting parallel between its impact on one area and your own belief in NLP having similar potential to have transformative effects across multiple fields. It is not only a clever play on words but also highlights the significance of paying attention to simple yet powerful solutions that can revolutionize entire industries!