3 ms·
Ai chatbots struggle conforming to facts. This will end badly.
by more_corn 2y ago
Ai chatbots struggle conforming to facts. This will end badly.
- nerdjon 2y agoHopefully they review it very carefully and not well after the event so the memory is fresh. Otherwise, you may forget important details and just think it was correct if you need to reference it later. We know how influential someone's memory can be. There is something to be said about needing to write this stuff down fully to commit it to memory. I agree that this is likely going to end badly.
- isubkhankulov 2y agoUsing LLMs to write a first draft based on your notes, rephrase and reword existing text is completely different from using an LLM to recall facts in its training data.
- russdpale 2y agoIt's using it's training data to rephrase and reword your existing text. Hello, mcfly.
- LargeWu 2y agoYou think AI chatbots are bad with facts, you should see the cops! Honestly I don't think this will be any worse than police filing hand-written reports rife with their own biases and material omissions and sometimes even outright, intentional lies.
- vundercind 2y agoI’d almost support requiring reports to go through an LLM filter that rewrites cop-speak and the exonerative voice into normal, straightforward language.
- visarga 2y agoWhy not have both? A cop report, and a LLM report.
- jrflowers 2y agoBecause the only party to benefit from that would be the company charging for the LLM. If arbitrarily adding more reports is valuable, why stop at just one LLM report? You could include the cop report and a dozen LLM reports with various temperatures and system prompts such as “write this report in the style of Sherlock Holmes as portrayed by Benedict Cumberbatch” or “as portrayed by Johnny Lee Miller” Edit typo
- infecto 2y agoI see this take a lot but I am not sure its entirely true. If we were to blindly ask questions based on a models world knowledge, yes there will be false statements. When asking questions or summarizing data across a provided dataset, I have not seen this to be a problem. Now I think there could be an interesting question on how its summarizing, there may be inherent biases but I don't believe the original issue is in fact a problem.