AutogenAI publishes primary research on the limitations of reasoning within current artificial intelligence systems and applies findings to helping organisations to win more work.
AutogenAI was the first business in the world to apply large language model technology to solving the problem of writing winning bids and proposals more efficiently and effectively.
We released the first version of our software in August 2022 – six weeks before Chat-GPT launched and we have been continually improving our product since then.
Large language models are remarkable. They are able to read and write and they are able to do so orders of magnitude faster than humans.
But large language models also have some critical limitations when used in isolation. One of these limitations is hallucination – their tendency to produce content that isn’t true. AutogenAI solved this problem in the bids and proposals space in May 2022 using a combination of natural language processing techniques including what would later be popularized as retrieval augmented generation.
AutogenAI’s team of engineers and researchers conduct hundreds of hours of research into the latest artificial intelligence technologies every week. By understanding the technology better than all of our competitors we are able to build better software.
A key part of our research agenda is codifying what makes a winning bid. This complex and multifaceted. One key component is an ability to know when a previous response is not suitable for a new question. Take, for example, the question:
‘Describe your experience of managing stakeholders.’
This needs a very different answer in a a proposal to deliver facilities management services in a hospital than in a proposal to deliver political consultancy services. In the first instance, citing previous experience delivering facilities management in a care home would be good evidence. In a proposal to deliver political consultancy services it would not.
Knowing what makes something relevant or not requires an extremely high level of intelligence. As part of our investigations into the ability of large language models to contribute to this problem our CEO, Sean Williams, and our Chief Research Officer, James Huckle, devised a new test for this as previous benchmarks were not suitable. Their tests present novel logictype problems that mimic the structure of popular logic questions found online but differ significantly in one or more critical aspects. Through a series of examples, they demonstrated a fundamental gap in LLMs’ realworld comprehension capabilities as well as showing considerable performance disparity between popular LLMs.
This research has real-world practical applications for AutogenAI enabling us to build software that helps our customers to win more. The research has been used by AutogenAI to: deploy the best large language models for understanding difference; ensure significant human guardrails in our User Interface when encountering these types of problems; and deploy non-LLM solutions (for example file segregation, meta-tagging and keyword content analysis) where these are more effective.
AI proposal writing solutions need to write winning bids. Understanding what evidence to apply in what contexts is a critical part of this. Only AutogenAI has published primary research in the area and that research has enabled AutogenAI to solve this problem where our competitors have not.
To learn more about AutogenAI and how we can help you transform your bid and proposal writing process, contact us today.