AGM Q&A 2024
Sean Hehir - We had rough ideas, rough ideas is not good enough. It’s got to be precise, and by the way we’ve got to be iterative because they’re going to demand things so we now have the tools to be responsive and squeeze it down to the smallest area. That gives me confidence.
People only build accelerated hardware for one purpose and one purpose only, to run models better , what was taking us a long time to simulate models to see the performance is not acceptable, we have now automated that process and now when a customer says “how fast .. my model, what’s it going to take, can we do this” we can do in a few minutes. That gives me confidence. And so those are tools but let me tell talk a little bit about the customers themselves because I think that’s even more important.
When I joined a lot of the conversations wed have with customers were like “interesting technology but let’s try to find a use case”. That’s not what we engage with at BrainChip at all. Engagements that we start with is a company that has a defined chip, a defined (inaudible, Doing a formal evaluation, we’ll engage with those. Sure we’ll do some business development but the ones I talk about the potential of winning have defined dates by major ?? With defined criteria.
Radically different profile. So that gives me confidence.
I made the comment in my speech about down selection, we often go into this with a list of 8 or 9 competitors, we’ve made the down select to 2 or 3 competitors in the final rounds of many engagements that we’re in right now. That gives me tremendous confidence.if we were getting dismissed I would say we have a problem with our product , we have a problem with our sales, but we’re not getting dismissed, we are making the down select. That gives me confidence.
I made some comments in the speech that I was personally spending time with C level executives of major companies. These are companies that have revenue measured in billions and multiple-billions or hundreds of millions. They would not consider … or spend 10 minutes with me if they did not think we were a serious contender for their strategic business. All these things give me confidence.
I know the dates that they’re projected to make their decisions. Do those dates move? Absolutely. All strategic decisions go through, they move a little bit, they move a lot, but they’re moving. But we have dates. That’s why I have confidence. All those things that are happening. We know dates, they’re not just science projects. Decisions were made and we think we’re well positioned. Can I guarantee a win on every one? Absolutely not. But listen to my words and listen to me carefully, I think you will hopefully share the same confOdense that I have on wins coming here in the conceivable future. Hopefully that was clear.
Has BrainChip management put on hold technology development while we focus on revenue?
Sean Hehir - Absolutely not. Would say again if you look at our company you got to do all of these things right. We’re doubling down our pace of innovation, this is a market that will not stand still and it’s something that we’re going to continually innovate. We have the right product right now. The markets not going to stand still, we want to win the deals we have now, we want to win potential deals in future years so we absolutely have to do both for sure, so innovation is the lifeblood of this company and we’re going to continue to do it.
With patents currently submitted for review and hopefully be granted by patent offices around the world how many patents have been granted since the last AGM? Also have all new patents or updates to patents announced to investors? If not, why? Also with current patents submitted to be granted have there been any key issues with approval of any patents submitted?
Dr Tony Lewis CTO - 3 or 4 patents since last AGM but don’t hold me to it. We are pretty consistent about giving updates now there’s sometimes you’ll have a patent in Australia or say USA and then you’ll have another patent worldwide and maybe some of those patents are essentially the same IP are not necessarily announced but typically we announce everything, we are very proud of the work we’re doing.
In the example provided, TENNS performance is compared to GPT2 which was released over 4 years ago in 2019. How does TENNS stack up against current state of the art eg. GPT4 ? Do you see us fitting in with other products being released with generative AI on devices such as the new Windows core pilot?
DR Tony Lewis CTO - We’re trying to build a device that’s going to run on something (inaudible) but it’s going to be battery powered so it’s not plugged into the wall so it has to be very low power and so the size of these networks are not going to be GPT4 size, they’re going to be much smaller and they’re going to be able to do simple tasks so we’re not going to be competing with OpenAI or anything like that. What we’re doing right now is we’re comparing ourselves against GPT2 because we have models that are approximately the same size in terms of numbers of parameters as GPT2 so that’s why we made that comparison. We are scaling up, we are making our models larger and larger and as we do so we’re going to compare ourselves to other models of comparable size, but this is a process that takes time and takes alot of compute energy to scale up. But we’re not going to stop at 2 but we’re not going to get all they way up to GPT4.
How is the current competition for silicon impacting future estimated timelines to manufacture and scale?
Dr Tony Lewis CTO - were an IP company so I can’t answer on behalf of customers.
With the Akida power consumption benefits is there a use case to add Akida to data centre architecture?
Dr Tony Lewis CTO - I have a smile on my face, I’d love to say that but we need to prove out our TENNS because we feels that’s a real competitive advantage, we have to see if we can scale up. But it would be wonderful if we could do that .
If TENNS is so good can you give me the number of customer engagements requiring TENNS application?
Dr Tony Lewis CTO - so TENNS is a competitor against existing neural networks, I want to say our customers have a view of the world that is pre-TENNS, they have to be introduced to the technology and they’re being introduced in a staged manner. We’re going out, we’re talking about TENNS, we’re publishing papers, we’re getting people familiarized. I believe that once they’re familiarized with TENNS they going to see that we can be use it for all sorts of things, so we’re proving out how we can use it for audio denoising, but also ASRs on urgent processing of medical samples, and also these large language models. And so we’re proving out what we can do with TENNS and once we have those proof points we can go to customers and say “hey guys you should go with TENNS because it’s going to be a lot more power efficient than what you’re using right now, it’s going to be good for you (inaudible) and learn about our technology “
Sean Hehir - I want to make a comment before the next question comes for a couple of themes that came up. You heard my speech where we’re going to put more resources into TENNS and more development (inaudible) more work coming in the future.
The other thing about patents I would say, and agin Tony is in charge of this, we are absolutely focused on innovation and filing patents and driving the pace of innovation up in the company