Artificial intelligence is worse than humans in every way at summarising documents and might actually create additional work for people, a government trial of the technology has found.
Amazon conducted the test earlier this year for Australia’s corporate regulator the Securities and Investments Commission (ASIC) using submissions made to an inquiry. The outcome of the trial was revealed in an answer to a questions on notice at the Senate select committee on adopting artificial intelligence.
The test involved testing generative AI models before selecting one to ingest five submissions from a parliamentary inquiry into audit and consultancy firms. The most promising model, Meta’s open source model Llama2-70B, was prompted to summarise the submissions with a focus on ASIC mentions, recommendations, references to more regulation, and to include the page references and context.
Ten ASIC staff, of varying levels of seniority, were also given the same task with similar prompts. Then, a group of reviewers blindly assessed the summaries produced by both humans and AI for coherency, length, ASIC references, regulation references and for identifying recommendations. They were unaware that this exercise involved AI at all.
These reviewers overwhelmingly found that the human summaries beat out their AI competitors on every criteria and on every submission, scoring an 81% on an internal rubric compared with the machine’s 47%.
I mean, what you’re essentially implying is, what if we could do a lot of things that we do today, but faster and less quality.
Imo we have too much things today and very few are worth their salt, so this is the opposite of the right direction.
That’s not what I’m implying. What I’m saying is that wasting time and effort on quality is pointless when the threshold for success is low.
For example, I could use aerospace quality parts (perfectly machined to micron-level tolerances) to build a toaster. However, while this would not increase the performance meaningfully, the cost would be orders of magnitude greater. Instead I can use shitty off-the-shelf parts because it doesn’t really make a difference.
Maybe in other words, engineering tolerances apply to LLMs too. They’re crude devices, but it’s totally fine if you have a crude problem.
Yes and my response to that is for some people maybe, for others they don’t want a low threshold, they want few good articles instead of spam of low quality.
Exactly, I’m saying there is no objective crude problem. You might be okay with simple summaries but I want every single piece of information I consume to have a very high bar.
Sure, go for it. But good luck paying an army of copywriters to summarize every article you read.
What if you’re reading Lemmy, and you don’t really feel like reading the article. Is the headline likely to tell you all you need to know or is the ai summary likely to find more info and without the clickbait?