A new UX with AI: LLMs are a Frontend Technology
Halo effect and Reasoning, NVIDIA PoV, History of UIs, and 3 Takeaways on AI and UX
Hello, Cyber Builders community! 🙌
Recently, we’ve explored some pivotal themes in the AI world in AI Meets Cybersecurity: Understanding LLM-based Apps.
I received good comments on dedicating a post on Ethics and another one on use cases for Cyber Builders. Duly noted and added to my to-do list ✅.
The post sparked other excellent conversations, and it seems clear that perceptions of what AI is—and what it's about—are mixed. This is mainly due to what psychologists call "the Halo Effect."
For those unfamiliar with the term, the 'Halo Effect' refers to a cognitive bias where our overall impression of a person influences how we feel and think about their character. It’s part of human nature to assume more positive attributes beyond what we know. Similarly, many are quick to crown AI as a source of reasoning or even some kind of truth—forgetting that its real power lies in delivering a refreshed and new user experience and effortlessly interacting with humans.
This week, we will plunge into the new frontier where LLMs are becoming the new frontend, travel back in time to review how digital communication has evolved, and uncover the threefold impact of LLMs on frontend interaction.
As we take this deep dive together, consider sharing this discussion with your friends, colleagues, or anyone who'd value joining our fast-growing community — a collaborative sharing space for entrepreneurs, security practitioners & industry leaders passionate about the future of cybersecurity.
Stay tuned for an enlightening read!
The New Frontier: LLMs as the New Frontend 🏞️
First, if you've been following AI and Large Language Models (LLMs) discourse, you would have come across Jensen Huang, NVIDIA CEO's exciting and insightful revelations. He envisages a future where LLMs are the first point of contact in virtually all our computer interactions. Watch out for the latest keynote he delivered at SIGGRAPH 2023. I also created a 1-minute clip so you can zoom off the segment where he is discussing LLM as the first computer UI.
The canonical use case of the future is a large language model on the front end of almost everything. Every single application, every single database, whenever you interact with a computer, you will likely first be engaging a large language model. That large language model will figure out your intention, desire, and what you are trying to do—given the context—and present the information to you in the best possible way.
This means LLMs won't simply serve as interaction interfaces. They will be intelligent interpreters of our desires and intentions, mining context from our queries and delivering well-rounded responses. Much like a thinking, understanding companion at our side every time we engage with technology.
Are LLMs potent enough to revolutionize human-computer interactions to this extent? Well, let's consider the runaway success story of ChatGPT.
I am always skeptical when people say that LLMs repeat what they learn or predict the next word. Why? Because most of the time, it misses what you say in your post: with the right fine-tuning « instruct » they can be taught valuable tasks on data: summarizing, mixing database data and questions, and generating code. They are not creative by themselves. They don’t have an abstract representation of the world.
This is why many LLMs, including ChatGPT, will fall short of reason on abstract questions or remember complex historical facts. On this end, I’ve shared previously how Google Bard was wrong at Cryptography (see here).
However, LLMs are potent tools for speeding up many tasks and providing a great user experience.
"A different language is a different vision of life."— Federico Fellini
In the case of ChatGPT, the chat interface is a revolution. Simple and intuitive to use, ChatGPT allows anyone to type into the bot and receive an instantaneous response. The genius lies in this intuitive interaction, transcending complex commands or codes and mirrors human conversation.
With this extraordinary success in mind, it becomes clearer how LLMs possess the potential to eliminate friction in human-computer interaction, serving not merely as tools but as efficient collaborators able to understand and anticipate our needs in a digital environment.
Journey through Digital Time 🕰️: A Capsule of Historical Interactions with Computers
Before we dive into the present, we must journey through the past. Initially, the conversation was one-sided—think of the 20th-century revolution of radio and television as point zero. Humanity was able to broadcast its progress to the world.
The historic moon landing of Neil Armstrong, for instance, was shared worldwide through television. These media channels, however, were mainly live broadcasting platforms, limiting user interaction. There was no recording, replays, personalized feeds, or AI as TikTok or Netflix use today when I interact with them on my tablet.
Computer interfaces have indeed evolved a lot in 70 years:
1 - The Era of Command Lines 💻 At the dawn of computer technology, interactions were primarily text-based command lines, following strict syntax one would input into a terminal. Though rudimentary by today's standards - while still used by power users -this began an intimate conversation between human and machine—of command and response.
2 - Arrival of Windows 💼 Fast forward to the ‘80s, and we enter an era defined by Mac and Microsoft Windows innovations (not forgetting the Commodore Amiga or Atari ST. 🧓🏻). This period welcomed features such as windows, drop-down menus, and navigation bars—a majorly intuitive leap from typing in command lines.
3 - The Web Browser & Hyperlinks🌐 Next on our timeline is the Internet revolution in the late 90s and 2000s. The arrival of web browsers introduced hyperlinks, enabling infinite referencing across the web. This allowed users to effortlessly jump from one page to another, making information more accessible.
4 - Mobile Interfaces in Your Pocket 📱 The dawn of the mobile era, especially towards the late 2000s, significantly shrunk interface sizes. With touch-based control—swiping, pinching, zooming—the user can handle a wealth of information on their mobile devices.
5 - Infinite Scroll & Social Media 👆🔄 The rise of social media platforms like Facebook and Twitter in the 2010s gave birth to the infinite scrolling dynamic. This feed-on-the-go continually displays new content. Platforms became infinitely more engaging as they lowered user inputs while multiplying content formats—text posts, images, videos—and personalizing them based on user behavior and connections. Social giants use extensively Deep Learning models. They are training these models to suggest new content based on the user platform interaction (where does he spend time reading a news article) or signals (like retweets, etc.)
Looking back at this whirlwind tour through technological development eras underlines that reducing necessary user inputs while boosting perceived content value is critical to interaction evolution. In simpler terms, we've swiftly moved from command lines to intuitive windows and navigation tools, from hyperlinks to the infinite scroll, aiming for seamless interaction between humans and devices.
Takeaways - LLM's Three-Fold Impact on Frontend Interaction 🚀
The Artificial Intelligence (AI) revolution fundamentally reshapes how we interact with machines, articulate queries, receive responses and interact with the results. Chat GPT and related generative AI tech most remarkably demonstrate this transformation in human-computer interaction. Their breakthrough? An open-ended graphical interface with three distinct features.
1️⃣ Open-Ended Interactions 🔄 The interface is open-ended, allowing users to express their queries with subtlety, nuances, and a chain of thoughts commensurate with their needs. Compared to the 2.0 interface--where the user interacts via a web or mobile application equipped with buttons and keyboards--the open-ended interface is liberating. It isn't forced to operate within predefined limitations but instead allows for a more organic flow of interaction.
2️⃣ Shift in Communication Mediums 📣👆 The medium of interaction will likely lean towards voice commands and specific gesture controls—forms of communication far more intuitive for humans. No one likes to type meeting minutes or write too long documents. AI will help to refactor and format data and ideas provided by the users as brainstorming notes based on their voice.
3️⃣ Powerful Workflows defined by users 🤝 No more limited choices dictated by application designers towards ones that users define; instead of being constrained by workflow determined by the application design, generative AI technology could allow for creating custom workflows, truly tailoring the experience individually.
I remember back in the 2000s years, the “agent-based software” buzz, where researchers were trying to mimic the behaviors of ants to create more complex software. With Generative AI, we might now have a multi-agent architecture based on various APIs that will cross the hurdles currently faced by previous technologies.
These three factors are set to fundamentally change the dynamics of our interaction with technology, reengineering it around our unique needs and preferences.
Conclusion
The intersection of AI, UX, and cybersecurity is a dynamic landscape continuously shifting under the weight of innovation and technological progress—and there's so much more to uncover! 🌎🔍
With Cyber Builders, I would love to hear your insightful thoughts on this sweeping transformation. What do you think about LLMs' role in our digital future? Do you see challenges we might face in harnessing their potential? Or do you foresee unique opportunities and benefits they might offer? As we continue this dialogue, every perspective adds a unique piece to the broader puzzle.
So, let’s continue to explore and build together, my friends! Drop your comments or feedback down below—we're all ears! Remember, this platform is made richer by your participation. So keep the conversation going as we shape the future of cybersecurity build-by-build.
Laurent 💚