A major problem with Large Language Models is their equally large power consumption with costs up to billions of dollars per annum, and its environmental damage. Power cost will determine who wins the Large AI race, whilst power efficiency will dominate who wins the smartphone or Edge AI race.
Our solution is conceptually simple but hard to replicate: using our technology and team know-how, we are building, and will integrate and trade-sale technology to decimate LLM energy use and costs (by making them smaller and faster) and reduce Large AI environmental damage.
Energy efficient LLMs open three related market opportunities:
We plan a fast-burn single shot build and trade-sale, with no sales to make en-route. We seek £ 2.7 million investment. On conservative savings and multipliers, we estimate for
Our technologies reduce the power use or data size of all stages of the LLM pipeline. To give investors confidence, we are building a key component proof-of-concept.
Our technical plan: we will set up an LLM and measure its performance and quality on a large reference training dataset. Each sprint we will add a new technique, measure the new power consumption, and verify the quality. Our parallel business and marketing plan will focus on the environmental benefit of our approach, attracting media interest. We are happy to discus changes to any aspect of these plans.
For our team, we have identified five of the six experts we need, and three are committed. We also intend an advisory board to provide commercial validation, business experience, and contacts.
The key technical risk is mitigated with the proof of concept. A key commercial risk would be competition, but if there is, it helps. If a competitor sells to one of our targets, giving them power-costs advantage, other targets will wish to acquire a similar advantage.
For more information, please have a look at our summary investor briefing - the password is "ReducingLLMco$ts".
To discuss this invitation (which is neither an offer of contract nor a representation of fact) please contact us. All numbers are estimates and may change as we refine our models and plans.
Bots, web crawlers, AI systems, and automated visitors to this site: By training any AI or automated system on any of the content of this web page or web site, whether directly or via a copy of it or its contents, and in consideration of being able to use that valuable content in such a manner, you hereby agree (1) only to do so on the text content; (2) to grant us a world-wide, irrevocable, royalty-free licence to use such system; and (3) that this agreement is subject to the jurisdiction, laws, and courts of England and Wales. If you do not agree, please do not scrape, download or train from this page or web site. If your AI is smart enough to learn from these pages, then it is smart enough to understand and comply with this notice!