Yes. Let's not forget that, while the world and his wife are familiar with the use of OpenAi's GPT "X", they do not have a clue as to how it works, let alone the underlying technology. However, we are seeing frequent examples of the tech world's increasing awareness.
One field of tech where Akida's potential is as yet untapped is in the implementation of LLMs.
Sensors (microphones, cameras, ...) and keyboards are essential input devices, and the input signals need to be converted to a format which the system can interpret. This requires the system management processor to be able to interpret natural language. Doing that purely in software burns a lot of electricity. Akida 2 introduces long range skip connexions:
https://brainchip.com/akida-generations/
W
ith the added support for short and long-range skip connections, an akidaTM neural processor can now accelerate complex neural networks such as ResNet completely in hardware without model computations on a host CPU.
Skip connections are implemented by storing data from previous layers in the akidaTM mesh or in a scratchpad memory for combination with data in later layers.
Key Benefits:
- Complex Network Acceleration: Enables complete hardware execution of non feed-forward model architectures such as ResNet and DenseNet.
- Low Latency: Eliminates CPU interaction during network evaluation which minimizes model latency.
By performing these functions in silicon, Akida reduces both power usage and latency compared with earlier LSTM/Attention software implementations.
Akida 2 provides a much more efficient foundation for implementing LLMs and particularly SLMs (Small Language Models adapted for edge applications). However, for the moment, the cloud-based Chat GPT and its LLMs have the advantage of incumbency despite their unsustainable growth in energy consumption. Nvidia's Jensen Huang is rightfully proud of the 2.5 times improvement in power consumption of Blackwell, but that is not sufficient to mitigate the exponential growth in power consumption to contain it within acceptable limits.
Akida offers orders of magnitude greater improvement, particularly with the use of SLMs at the edge. Akida has the potential to be the foundation stone of all things GPT. The development of SLMs is just some of the lead in Akida's saddlebags.
While Edge GPT is the present Holy Grail, there is a myriad of other applications for Akida which can be implemented in the near term.
ADAS - DMS - UI - lidar/..., space, SDR, Edge Box, ...
For startup companies, the rational basis for assessing share price is the market potential of the product.
The problem is that the share price is the plaything of manipulators who take advantage of the fact that many shareholders and potential shareholders do not fully grasp the potential of this revolutionary technology, and the fact that the pre-income phase has been longer than anyone anticipated. One reason for the delay in revenue is that the technology is evolving at an unprecedented rate, LSTM, Attention, 4-bit weights and actuations, Chat GPT, TeNNs, ViT, ... Remember, Akida 2 is just 1 year old.
Added to that is the fact that the company has made a few mis-steps or changed course over the last few years in an attempt to come to grips with an entirely unheralded market. The company was metaphorically feeling its way in the dark and needed to test the waters to see which way the current flows ... and the current has not flowed smoothly.
So there are a lot of factors to be taken into account which no one could have foreseen a few years ago. This has been a learning experience for the shareholders and the company.
