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%.
Llama 2 is insanely outdated and significantly worse than Llama3.1, so this article doesn’t mean much.
L2 it’s one year old. A study like that takes time. What is your point? I bet if they would do it with L3 and the result came back similar, you would say L3 is „insanely outdaded“ as well?
Can you confirm that you think with L3, the result would look completely opposite and the summaries of the AI would always beat the human summaries? Because it sounds like you are implying that.
We know the performance of L2-70b to be on par with L3-8b, just to put the difference in perspective. Surely they models continue to improve and we can only hope the same improvements will be found in L4, but I think the point is that models have improved dramatically since this study was run and they have put in way more attention in the fine-tuning and alignment phase of training, specifically for these kinds of tasks. Not saying this means the models would beat the human summaries everytime (very likely not), but at the very least the disparity between them wouldn’t be nearly as large. Ultimately, human summaries will always be “ground truth”, so it’s hard to see how models will beat humans, but they can get close.
Lemmy users try not to make a strawman argument (impossible challenge)
No, that’s not what I said, and not even close to what I was implying. If Llama 2 scored a 47% then 3.1 would score significantly better, easily over 60% at least. No doubt humans can be better at summarizing but A) It needs someone that’s very familiar with the work and has great English skills and B) It needs a lot of time and effort.
The claim was never that AI can summarize better than people, it was that it can do it in 10 seconds and the result would be “good enough”. People are already doing AI summaries of longer articles without much complaints.
This was not a strawman. Please don’t assume lemmy users make logical fallacies when it’s only you who thinks that.
I guess you missed the part where he said “Oh you said X but you’re actually implying Y? Did you mean Y? Please confirm you actually meant Y.”
That’s my point, from my perspective, there was no switch. Using a one year old model is fine.
My comment was about how people looking at the same thing, one might think it’s a bait and switch while the other one always knew the second item was being implied.
The headline never said all AI or latest AI.
This is pretty much every study right now as things accelerate. Even just six months can be a dramatic difference in capabilities.
For example, Meta’s 3-405B has one of the leading situational awarenesses of current models, but isn’t present at all to the same degree in 2-70B or even 3-70B.
Just a few more tens of millions of dollars, and it’ll be vastly improved to “pathetic” and “insipid”.
You didn’t bother to Read the article. Read the article. Study was conducted last year
I read the article. I’m aware it’s an older study. Point still stands.
And yet your claim is still pointless unlike this study
Did LLama3.1 solve the hallucination problem?
I bet we would have heard if it had, since It’s the albatross hanging on the neck of this entire technology.