Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
If you ask Yan to beSilicon Valley has a groupthink problem. Since he left Meta in November, the researcher and AI star has done just that I took aim In the traditional view that large language models (LLMs) will get us to artificial general intelligence (AGI), the threshold at which computers match or exceed human intelligence. Everyone, announced in a The last interviewhe was an “LLM darling.”
On January 21, the San Francisco-based startup Logical Intelligence launched appointed LeCun to its board of directors. Based on theory Envisioned by LeCun Two decades ago, the startup claimed to have developed a different form of artificial intelligence, better equipped to learn, reason, and self-correct.
Logical intelligence has developed what is known as the Energy-Based Reasoning Model (EBM). While LLM experts effectively predict the next most likely word in a sequence, EBMs accommodate a set of parameters—for example, Sudoku rules—and complete the task within those limits. This method should eliminate errors and require less calculations, since there is less trial and error.
The startup’s first prototype, the Kona 1.0, can solve Sudoku puzzles several times faster than the world’s best leading LLM software, despite the fact that it runs on just a single Nvidia H100 GPU, according to founder and CEO Eve Bodnia, in an interview with WIRED. (In this test, LLM holders are prohibited from using cryptographic capabilities that would allow them to “brute force” a puzzle.)
Logical Intelligence claims to be the first company to have built an effective EBM, which so far is just a flight of academic imagination. The idea is for KONA to tackle thorny problems, such as optimizing power grids or automating sophisticated manufacturing processes, in error-tolerant settings. “None of these tasks are related to language,” Bodnia says. “It’s just language.”
Bodnia expects Logical Intelligence to work closely with AMI Labs, a Paris-based startup recently launched by LeCun, which is developing another form of artificial intelligence — a so-called universal model, which aims to recognize physical dimensions, exhibit persistent memory, and predict the outcomes of its actions. Bodnia asserts that the path to AGI begins with layers of these different types of AI: MBAs will interact with humans in natural language, cyberneticians will handle reasoning tasks, while universal models will help robots take actions in 3D space.
Bodnia spoke to WIRED via video from her office in San Francisco this week. The following interview has been edited for clarity and length.
Wired: I have to ask about Yan. Tell me how you met, his role in directing research at Logical Intelligence, and what his role on the board will involve.
Bodnia: Yan is very experienced academically as a professor at NYU, but has been exposed to the real industry through Mita and other collaborators for many years. He saw both worlds.
For us, he is the only expert in energy-based models and different types of structures associated with them. When we started working on this EBM program, he was the only person I could talk to. It helps our technical team navigate in certain directions. It was very practical. Without Yan, I can’t imagine us expanding so quickly.
Yan speaks candidly about the potential limitations of LLM programs and which model architectures are most likely to move AI research forward. Where do you stand?
LLMs are a big guessing game. That’s why you need a lot of calculation. You take a neural network, feed it pretty much all the garbage from the Internet, and try to teach it how people communicate with each other.
When you speak, your language is smart to me, but not because of the language. Language is a manifestation of everything that exists in your brain. My thinking occurs in a kind of abstract space that I translate into language. I feel like people are trying to reverse engineer intelligence by simulating intelligence.