By Cyrus Shaoul, CEO, Leela AI
We have received a lot of positive feedback from being profiled in a Forbes article last month. Some of you also had some questions, which I will further explain here.
- What is this new LAM acronym you are using to describe Leela Platform?
- How exactly does Leela address the AI sector’s rising energy use?
The paywall-protected story by Dean DeBiase, which is entitled American Policy To Drive Revival Of Technology And Business Growth, focuses extensively on Leela AI in the second half of the article. Leela AI is profiled as an example of innovative US-based industrial technology provider, as well as a potential solution for skyrocketing AI energy consumption.
When DeBiase asked how Leela can offer faster training and deeper insights into visual information than standard neural networks, I said: “It springs from LAMs (Large Activity Models). When you multiply the number of things people can do with the quantity of things in the environment, you get a very large number of possible actions. The core IP inside Leela enables it to recognize a much larger number of activities than other models and is more accurate than the multimodal models like OpenAI.”
So what the heck is a Large Activity Model? Instead of modeling language patterns like LLMs, our LAMs model activities. Unlike LLMs, LAMs can efficiently analyze video and/or take autonomous action.
The key ingredient in a LAM is a causal (or symbolic) agent combined with neural networks and vision AI algorithms. Our causal agent enables Leela to not only recognize dozens of actions within a video but understand their meaning and context.
Here are some advantages of LAMs:
- More capable of understanding action, cause, and context
- More explainable
- More accurate and reliable
- More computationally efficient for training and inference
NOTE: Some have used the LAM acronym to refer to Large Action Models rather than Large Activity Models. These similarly combine a causal agent with neural nets and vision AI. However, they also typically include a major LLM component and are used as agents that autonomously perform tasks on computers or the Internet.
The impact of this technology is huge for Leela AI’s manufacturing customers. Leela records assembly cycle time for each station, as well as value vs. non-value added time. It tracks tool and equipment use and measures tasks such as picking up, carrying, and using different tools. All these insights, which would be difficult to acquire with a standard neural net and impossible with an LLM, helps our customers compare metrics between stations, shifts, and time periods to see what works and what doesn’t.
As I explained to Forbes: “We deliver a 50% reduction in safety incidents. Customers typically see a boost of 10% to their production capacity. So that’s a lot of business value, if you can see what’s happening and if an AI can help better understand what’s being seen and then take actions based on that.”
Forbes also interviewed Drew Satorius, Global Director of Advanced Manufacturing Technology at one of our favorite customers, BAC. “Leela’s solution has helped us gain a deeper understanding of bottlenecks with insight into operational improvement opportunities,” Satorius told Forbes. “We can now focus more of our time and energy on making progress instead of collecting data. We can mobilize teams to make improvements quickly and, due to the real-time continuous analysis, we can quickly see the results of our AI-augmented decisions.”
As hinted at by Satorius and the LAM bullet points above, Leela offers exceptionally fast AI training, and to a somewhat lesser extent, inference. Thanks to Leela’s LAM – and its causal agent – training requires about 100x less training time and 10x less time-to-value compared to a typical neural network based solution.
As DeBiase suggests in Forbes, greater efficiency means much lower customer costs, reduced power usage, and fewer carbon emissions compared to standard neural nets. And the energy savings are huge compared to LLMs.
I will share more thoughts on LAMs and Leela’s benefits in energy efficiency in the weeks and months ahead.
Any other questions I can answer? Drop us a line on our contact page.