Do we need a large language mn

 Modern AI systems are built around what are termed 'large language models (LLM)s.

A Large Language Model (LLM) is an AI system trained on huge amounts of text. It recognises language patterns and uses them to answer questions, write content, summarise information, and hold conversations. The system is trained on vast amounts of data available on the Internet - information to which it may not have 'proper' access.

Of course this information belongs to someone - but is often readily available to trawling software robots that collect, dissect and analyse it ready for access as part of the LLM.

This means that for those who use very specialised datasets, the LLM may contain vast amounts of irrelevant data - and this is partly why LLM-based AI can give spurious results and can 'hallucinate'.

Now, some organisations are wondering  if they are better limiting the data in their language model to their own data which they have verified and can trust.  There will be much less data but since the queries to be asked of the data will likely relate to the data in some way (since those data are part of the output of the organisation's own systems and processes results re likely to be more specific and relevant (though not necessarily more useful?).

Of coursedin practice the organisation would use a general LLM-based system in addition for general purposes but rely on the own-data-model for specific process based enquiries.

Horses for Courses?