Covid and the trillionaire AI

Firstly my apologies for the lack of posts recently. I was lucky enough to catch Covid again. Have had all my shots, but it still knocked on my arse, probably because I am an old bugger.

Anyway, that is my excuse for not posting over this last week or two, it is also my excuse if this post is crap. Moving on.

I did a post recently called "Garlic bread, it's the future" where I am banging on about AI and the idea of firms having LLMs internally for use with the corporate intranet.

This post is an extension of this in a way because I still think I know that internal LLMs running on intranets is one of the next big ideas. I read that a billion-dollar start-up is known as a unicorn, a $10B is decacorn and $100B valuation is a hectocorn. I think this idea is decacorn at the least but more likely it is hectocorn league. LLMs and intranets are going to be big big business.

There are lots of reasons for me to think this based on my years working on corporate intranets. I look at every job I've done of this type and every single one of them would have benefitted from the use of AI as it stands now. These are the obvious points that occur to me straight away.

1 Corporations require enhanced data security and privacy. Small internal LLMs leverage the security of intranets, ensuring sensitive corporate data remains within the confines of the organization. This is massive because if this data were to leak out it could and would have a catastrophic effect on the business. It does not make sense to risk compromising data by having it all in a single super-large LLM.

2 Tailored Solutions for Specific Industries. These LLMs can be customized for specific industry needs, offering more relevant and accurate assistance than their generic counterparts. This customization is key in fields like legal, finance, healthcare, and a hundred other fields. Every corporate intranet I have ever worked on had features that were unique to that company let alone industry. Meaning they would benefit from a tuned LLM.

3 Improved Efficiency and Productivity. By operating on internal networks, these LLMs can quickly process and analyze large volumes of company-specific data, leading to improved decision-making and productivity. This is what I have read, but how exactly does it do this? For data that is archived or unchanging, it is trivial, but what about data that is being updated daily? emails being a prime example. Turns out it is really easy (I'll probably get my arse handed to me for saying that), more on this in a later post.

4 Cost-Effective AI Integration. The idea here is that smaller LLMs require less computational power and resources compared to larger models, making AI integration more accessible and cost-effective for businesses. That sounds great, but how would this work for a company with 30 years + of corporate information that may span 50 million + documents? I have worked on document management systems that handled 5 million+ documents a year. This is not a trivial task by any means. With a relatively modest expenditure, this can be achieved though and the potential gain is terrifying.

5 The Intranet Advantage. Leveraging existing intranet infrastructure, companies can deploy these models without the need for significant additional investments in IT infrastructure. This is sort of true, but if you are a corporate and you want to do some Domain-Adaptive Pretraining you may want some decent GPUs and these may cost upwards of £10k each. The IT guys are going to love you for this because some of those GPUs (a 4090 for example) are going make the gaming experience phenomenal.

OK, I got to 5 then I gave up. I am hoping you are beginning to see that I may be right about LLMs and intranets being something big, but where does the decocorn business come in? Microsoft are currently rolling out CoPilot which promises all kinds of wonders and boosts in productivity and I think it will deliver something amazing over the coming years. It will generate billions in revenue for Microsoft, but this is different to what I am talking about.

NVidia put something up on their website called ChipNeMo. It uses LlaMA2 as a foundation model to create the ChipNeMo foundation model using years and years of design docs and code. They then use supervised fine-tuning to create the ChipNeMo Chat Models. On the Nvidia site it says that they had 24 billion tokens of chip design docs and code and then some fine-tuning. The aim is "to explore the applications of large language models (LLMs) for industrial chip design." What I take this to mean is that all the stored knowledge will now be more accessible to their employees. Surfacing relevant information has always been difficult and continues to be, so this is one area that will definitely benefit everyone.

Every industry is waking up to the idea that they need to figure out how AI is going to help their businesses, they are going to have their equivalent ChipNeMo moment/application. A company is going to evolve that provide this service for corporates. What is super cool though is that the company itself will leverage AI to provide this service. This is only the first part of it though. It is the second part that really blows my mind, get on this little gem...

As part of the service, each user/employee will have the equivalent of their own AI that is tuned to them. It will be tuned using their email, their calendar, the files they access regularly and a whole bunch of data that is relevant to them and their role in the corporation. This application will offer customized assistance, help the employee map out their career path within the company, help manage their mental health and well-being as well as dozens of more obvious uses. By obvious I mean things like enhanced productivity, knowledge management, improved communication etc.

The personalized AI and the corporate intranet/AI is going to be a massive game changer for every business out there, and for those who need to worry about such things, the ROI will be obvious and significant.

I started in web development back in the 90s and everyone in the business knew that it was going to be massive. This is going to be even bigger. The company that is going to be the biggest winner from this technology probably hasn't even been created yet. My prediction is that the world's first trillionaire, will come out of the AI business and will appear in the next 5 years.

That is enough of my rantings for one day, hopefully, normal service will be resumed from next week

Bryan
ps. Once again I wrote all of this, so if it is shite, it is on me and not the AI.