In a Surprise to No one, GenAI is not good at some tasks

In a Surprise to No one, GenAI is not good at some tasks, knowing my experience with Generative AI large language models and their capability to date, it is not surprising to see the Australian government trial ended with a “not quite good enough” (linked).

LLM’s are good for basic tasks, not people replacing tasks in most cases, sales call outs and call centre tasks excepted – just watch that the LLM doesn’t promise something you can not deliver.

To the point of the article and to not repeat the linked findings (I’m weaving!), I would add that the point of detailed content in submissions, projects etc is to flesh out background, strategy, roadmap, product etc, to have an AI assistant provide a summary of what detail was discussed is not a replacement for being ‘Present’ and absorbing details and nuance.

1. An LLM generated summary does not catch the details that were most important and does not engage the reader of the summary in the same way that the original discussion would and probably misses the actual point of the paper – but you wouldn’t know that the greatest idea in the past 6 months was just uncovered without the full content.
   
   A Generated Summary also does not catch the importance of particular parts of the discussion nor the thought process.
   
   Large Language Models are Word salad makers still, just the same as giving an unlimited number of monkeys a keyboard and eventually you would have the works of Shakespeare writte, just much faster.
   
   Generative AI is an Assistant, not a replacement and the sooner the market settles down to understand this and stop hyping what is essentially an expensive assistant, the better for everyone.