What Stallone Doesn't Say About AI (But Would Have Said Better Than Me)

More, More, and Then Some More

Read this in Stallone's voice: "stronger, more powerful, more robust, more precise, the greatest of all time."

Song made with Suno

No, I'm not talking about a superhero or my dad. It's just that every time I read a piece of news about AI, this Stallone effect kicks in inside my head. Okay Billy!

At the same time, we're living in a fantastic and disruptive era, and every day, I feel like we're not fully grasping what's happening and especially what's about to happen.

Since November 2022 and the rise of ChatGPT, we've been hit with a constant stream of news about the growing power of generative models capable of producing text, images, audio, videos, document analyses, and more. It's a fucking game changer: automation, optimization, initiative-taking, self-learning... AI has been democratized -- it's no longer reserved for Big Tech; it's accessible to everyone.

Foundation models -- the AI models that serve as a base, such as ChatGPT, Gemini, Claude, Mistral, Deepseek & Co -- are constantly working to improve, perform better, avoid hallucinations more effectively, reason more deeply, and optimize models to consume less while delivering more. As a result, we're moving from bazooka-sized models, LLMs (Large Language Models), to smaller but more specialized models called SLMs (Small Language Models).

For several months now, researchers have been exploring the promising avenue of LCMs (Large Concept Models), but that's another topic -- a fascinating one that deserves its own deep dive. I won't even get into quantum computing, which is clearly a more-than-promising revolution on the horizon.

My key takeaway: we're doing more and more with less and less.

Why the term "foundation"? Because these models serve as foundations upon which you can build houses, cabins, small buildings, or towers. The most well-known models belong to American or Chinese companies -- thankfully we have Mistral in Europe to save the honor!

There are also open-source AI models, meaning they're accessible to everyone. You can download them and do whatever you want with them. It does require some technical skills, but at least you're in full control, managing your own confidentiality and security concerns without going through third parties.

The foundational work isn't finished -- we keep seeing announcements like Claude 4 which just came out, and VEO 3 for video.

There's still a lot of work to do on what consumes enormous computing power, like audio and video processing. Generating videos today is possible, and it's pretty mind-blowing, but it remains quite limited.

I'd also add that people are now talking about standardization and common protocols -- all jargon to explain the fact that we want to enable AIs to work better together and streamline communication points to offer a more accessible experience.

That's the state of affairs when it comes to the foundations.

The Two Major Breakthroughs of 2024

Thanks to these foundations, two things began to emerge throughout 2024:

• Agents,
• Multimodality.

Storytime with Stallone: What's an Agent?

An agent is an AI that you give a mission to, and it's capable of handling it on its own. It observes what's going on around it, figures out what it needs to do, and acts without needing to be told what to do at every step.

For example, imagine an intelligent personal assistant: you tell it "schedule a meeting with Paul this week." It looks at your calendar, Paul's calendar, suggests time slots, sends the emails, and lets you know when it's done. You didn't have to do anything more. It understood the goal, made decisions, and completed the task on its own. That's what an agent is.

With this agentic concept, the possibilities are E-N-O-R-M-O-U-S, and it's become a standard!

Let's get concrete for a second: when you keep having the same conversations with ChatGPT that always start the same way -- "you are a marketing expert at such-and-such company doing such-and-such" -- instead of writing that every time, you can just create an agent that's a "Marketing Expert." That way, you don't need to set the scene anymore; you can talk directly to your Marketing Expert.

Storytime with Stallone: What's Multimodality?

Multimodality is when an AI can understand and use multiple types of information simultaneously -- like text, images, voice, or even video.

For example, if you show a photo to an AI and ask it out loud: "What's this guy doing?" -- a multimodal AI will look at the image, listen to your question, understand both together, and answer: "He's playing guitar on stage." It understood the image and the voice and made the connection between the two. That's multimodality: an AI that perceives the world like we do, with multiple senses.

In my opinion, it's still fairly limited, honestly, but it has the merit of existing.

You can chain AIs together (or agents, now that you know what those are) that execute one after another using platforms like Make or N8N, but once again, that's still too technical.

2025: Finally, the Focus Shifts to the AI Experience

If you're a developer or someone tech-savvy, you're capable of doing extraordinary things with AI. Up to now, the experience has mostly been limited to prompting and getting results -- often good, but sometimes insufficient.

Over the past few months, we've seen a clear willingness from the ecosystem to develop tools built on AI to offer something new.

Some tools have been doing this for a while.

Take Gamma for slide generation -- it's very impressive and quite stunning, but after several months of use, you realize the slides always look the same, following a repetitive and ultimately limited pattern.

If I take Fireflies or Upmeet, tools that use AI to process audio or connect to your video meetings to take notes for you, they're excellent tools. On the same level, you have devices like Plaud, a small portable device that can record your phone calls and/or in-person conversations and uses AI to generate your meeting notes, summaries, and more. But there's no single tool that does both: handles your in-person and phone exchanges AND manages your video calls, all while using AI so you can make sense of everything. I'm not even mentioning the privacy issue, where your conversations often pass through American or Chinese AI systems.

Spoiler alert: with our startup Lamalo, we're developing a tool that does all of this and goes even further on the experience side AND adapts to your privacy requirements.

As you can see, what I'm trying to show you is that serious work is being done to deliver a tailored, in-depth service that addresses many topics through AI. Hence announcements like: OpenAI is buying the iPhone designer to create its first "AI device." Today, we talk about "generating text" and "generating images," but tomorrow, we'll be saying we want to "understand and interact with a concept."

I see players positioning themselves with AI to challenge established incumbents, like Delos (a French startup looking to take on platforms like Microsoft 365 or Google Workspace) or Dust (a French startup specializing in building custom AI assistants for businesses). I see startups wanting to build the next "Jarvis" by creating swarms of specialized agents capable of self-organizing to handle highly complex tasks autonomously, making choices like resource allocation, and also serving as advisors by proposing anticipation scenarios -- essentially doing predictive work.

We're also seeing a movement toward devices that integrate AI, what's called "edge AI": from smartphones to cars to home appliances, there's a push to embed AI into our devices. This integration enables hyper-personalization while respecting privacy, since the AI runs on a device that doesn't communicate externally. Your devices adapt to your habits and proactively make recommendations.

Concretely, you're going to have work to do in the coming months and years

And What About at Reboot, Mr. Stallone?

Everything that's happening, we've been talking about at Reboot since 2023, and it's incredible to see how the world is getting into gear to bring to life ideas that used to be science fiction.

We recently created a structure to launch our own AI products, and we've worked insanely hard on the experience. We received a BPI grant for our project, and even before finishing development, we already have prospects who want to adopt our solution. It's coming verrrrrrry soon. But this vision of the experience -- we've had it since 2023, having developed our own RebootGPT, which is used across the entire company. Now, we're going much further.

Stay tuned, it's worth it!