Why AMI Labs’ Alexandre LeBrun won’t call his AI ‘AGI’ or ‘superintelligence’

The Marketing of Miracles

If you have spent more than five minutes on social media lately, you have probably seen the acronym AGI thrown around like it is a settled scientific metric. Every venture-backed startup claims to be six months away from building a god in a box. It is exhausting, and quite frankly, it is mostly noise designed to keep the valuation spreadsheets looking green. That is why hearing from Alexandre LeBrun at AMI Labs is such a breath of fresh air.

LeBrun, who is working alongside industry heavyweights like Yann LeCun, is not interested in the buzzwords. While his contemporaries at OpenAI or Google are fighting over who gets to define Superintelligence first, LeBrun is busy admitting something very few founders will: we do not actually know what those words mean. He is refusing to use the terminologies that have become the primary currency of the AI hype cycle.

For those of us building real products, this distinction matters. When a founder starts talking about AGI, they are usually trying to sell you a dream. When they talk about world models and objective-driven intelligence, they are talking about engineering. One is a magic trick; the other is a roadmap.

The Problem with Artificial General Intelligence

The term AGI—Artificial General Intelligence—is a moving goalpost. Twenty years ago, beating a human at chess or translating a document instantly would have been considered general intelligence. Today, we call that a standard feature in a free smartphone app. LeBrun understands that by labeling a target as AGI, you are basically setting yourself up for a perpetual cycle of disappointment and redefinition.

The issue with words like Superintelligence is that they are comparative, not absolute. They imply a ceiling that we have not reached yet, but they do not provide a technical specification for how to get there. It is a marketing term used to drum up investment from people who are afraid of missing out on the next big paradigm shift. LeBrun’s refusal to play this game suggests that AMI Labs is focused on the architecture rather than the applause.

As builders, we should be skeptical of anyone who claims to be building an all-knowing, all-capable mind. Intelligence is not a monolithic block. It is a collection of specific capabilities, many of which our current Large Language Models still lack, such as reasoning, planning, and a persistent understanding of physical reality.

Building World Models Instead of Chatbots

LeBrun’s focus is on something called world models. This is the core philosophy that Yann LeCun has been championing for years. The idea is that current AI is basically just a very sophisticated autocomplete. It predicts the next token based on statistical patterns, but it does not actually understand that if you drop a glass, it will shatter. It does not have a mental model of how the world works.

To get to the next level of utility, we do not need bigger datasets or more GPUs—at least not in the way we are using them now. We need systems that can simulate outcomes before they act. This is what humans do. We plan. We imagine consequences. We understand causality. If AMI Labs can crack the code on how an AI can build its own internal model of the world, that is infinitely more valuable than a chatbot that can write a mediocre poem in the style of Shakespeare.

For founders in the crypto and AI space, the takeaway is clear: stop trying to build generalists. The value is in the depth of the model's understanding of its specific environment. Whether that is a decentralized financial market or a robotic arm in a warehouse, the world model is the engine that drives actual utility.

The Founder Perspective on Hype

It takes a certain level of confidence to walk into a room of investors and refuse to use the words they want to hear. LeBrun is effectively de-risking his company by being honest. By avoiding the AGI label, he avoids the inevitable backlash when the system inevitably fails at a task a human toddler could do. He is managing expectations while focusing on the high-level engineering challenges that actually move the needle.

We see this same pattern in the crypto world. There are projects that promise to decentralize the entire internet and replace every bank on earth by next Tuesday. They usually end up as footnotes. Then there are the teams that say, we are building a more efficient way to settle cross-border stablecoin transactions. They might not get the same viral headlines, but they are the ones still standing when the hype cycle resets.

LeBrun’s approach is a lesson in intellectual honesty. He knows that his team is working on something groundbreaking, but he also knows that calling it Superintelligence is just a way to hide the fact that we are still in the very early innings of this technology.

Why Builders Should Care

When you are designing your own product or protocol, you have to decide which camp you are in. Are you building for the headline, or are you building for the use case? The headline builds a community of speculators. The use case builds a community of users. LeBrun is clearly building for the latter.

If we strip away the mystical language, what we are really talking about is efficiency and reliability. Can an AI system perform a task with 99.9% accuracy without a human in the loop? That is the real milestone. It does not matter if you call it AGI or a very good script. If it works, it creates value. If it does not work, no amount of fancy terminology will save your burn rate.

• Focus on architecture: The underlying structure of how an AI learns is more important than the size of the training set.
• Avoid the hype trap: Using buzzwords might help with a seed round, but it creates a debt of expectation that is almost impossible to pay back.
• World models over word models: Real-world utility requires an understanding of physics, logic, and cause-and-effect, not just language patterns.

The industry is currently obsessed with the idea of a digital god. But while everyone else is arguing about when that god will arrive, people like Alexandre LeBrun are quietly building the tools that will actually make our software smarter. It is less glamorous, but it is a much better bet for the long term.

The term AGI is a distraction from the actual engineering work required to make machines truly useful in the physical and digital world.

We should all be a little more like LeBrun. Let others fight over the dictionary. We should focus on building things that actually work, regardless of what the marketing department wants to call them. Honest engineering usually beats hype in the end, even if it takes a little longer to get noticed.

Read the original at TechCrunch AI →