A Yann LeCun–Linked Startup Charts a New Path to AGI
A New Path to AGI: Logical Intelligence Charts a Different Course
In a recent interview, Yann LeCun, a renowned AI researcher and luminary, declared that Silicon Valley has a groupthink problem. According to LeCun, everyone has been "LLM-pilled," meaning that the focus on large language models (LLMs) has led to a narrow view of what artificial general intelligence (AGI) can achieve. LeCun, who has recently joined the board of San Francisco-based startup Logical Intelligence, believes that a different approach is needed to reach AGI.
Energy-Based Reasoning Models: A Deeper Dive
Logical Intelligence has developed an energy-based reasoning model (EBM), which absorbs a set of parameters and completes a task within those confines. This method is supposed to eliminate mistakes and require far less compute, as there's less trial and error. The startup's debut model, Kona 1.0, can solve sudoku puzzles many times faster than the world's leading LLMs, despite running on just a single Nvidia H100 GPU.
How EBMs Handle Tasks Differently
Unlike LLMs, which rely on predicting the most likely next word in a sequence, EBMs use a true reasoning model. When faced with a task, an EBM can see in multiple directions, choose one, and if it encounters a hole, try another way. This is in contrast to LLMs, which are not allowed to deviate until they complete a task.
Training Data for EBMs
The training data for EBMs can be anything, and the model is able to extract the full data from sparse data. Imagine showing someone how to draw a cat, and they can extrapolate how to draw a dog. This kind of extrapolation is what EBMs are capable of.
Practical Applications
Logical Intelligence is interested in deploying EBMs in various industries, including the energy sector. In real-time, energy distribution requires processing a lot of variables and distributing energy accordingly. EBMs can automate this process, making it more efficient and reducing waste.
Collaboration with AMI Labs
Logical Intelligence is committed to working with AMI Labs, Yann's Paris-based startup, which is developing an alternative model architecture. While their focus is on world models, Logical Intelligence is focused on building the brain. They believe that their approaches complement each other and can lead to a more comprehensive understanding of AGI.
Opening Up the Model
Logical Intelligence has opted not to open-source their model, Kona, yet. They want to ensure that they understand it well enough before releasing it to the public. This is a responsible approach, considering the potential implications of their technology.
The Road to AGI
Logical Intelligence sees AGI as an ecosystem of compatible AI models that serve the world and people in the most productive and safest way. They believe that EBMs represent a route to AGI, as they can self-align, self-assess, and plan. This is a critical step towards achieving AGI, and Logical Intelligence is committed to scaling up their model and exploring different use cases.
Funding and Future Plans
Logical Intelligence is seeking funding to scale up their model and explore different use cases. They want to try different teams and partners to develop new applications of EBMs. They also plan to educate people about the different forms of AI and the potential of EBMs.
Conclusion
Logical Intelligence's approach to AGI is a significant departure from the traditional focus on LLMs. Their energy-based reasoning model has the potential to achieve AGI in a more efficient and effective way. As they continue to develop and refine their technology, they are likely to make significant contributions to the field of AI and the pursuit of AGI.
Source: https://www.wired.com/story/logical-intelligence-yann-lecun-startup-chart-new-course-agi/
