Yes, thank you so much. I totally wanted and have time for a scavenger hunt + jigsaw puzzle. Clearly you understand enough of the issue to know that I need to handle it but can’t be bothered to spend 45 seconds to summarize the relevant bits.
Yes, thank you so much. I totally wanted and have time for a scavenger hunt + jigsaw puzzle. Clearly you understand enough of the issue to know that I need to handle it but can’t be bothered to spend 45 seconds to summarize the relevant bits.
Because a person writing a summary actually understands what they read and wrote the summary based on that.
LLMs don’t. It’s literally fancy autocomplete. That’s all it is. It’s really fancy autocomplete, but it doesn’t actually understand anything.
– Frost
And yet it produces better summary output than many humans I’ve had the (dis)pleasure of reading the work of.
You’re assuming because it doesn’t function the same way as a human that the output can’t be equal or better. That’s simply not the case at this point in this type of task.
Autopilot in airplanes doesnt fly like a human, and yet its installed and running on almost every passenger flight in the world.
Autopilot in planes also isn’t an automatic pilot. All it does is the bare minimum of “keep the plane level at this altitude/climb rate and/or heading this direction” while you focus on other things (unless you get distracted in the wrong place and fly into a mountain). Some get a little fancier than that, but still.
It doesn’t replace the pilot and it’s not even trying to replace the pilot.
Oh and you know what happened when pilots got too used to following the line on their display instead of actually using their brains and reacting to the situation? Planes started crashing. They had to actually dial back on the automation somewhat. https://99percentinvisible.org/episode/children-of-the-magenta-automation-paradox-pt-1/
– Frost
They can entirely take off and land at this point, it’s a lot more capable than you think.
You didn’t discuss the actual situation though, LLM models are capable of summarizing email chains just as well or better than a human at this point.
I think you underestimate the importance of “knowing what the words actually mean” if you think a statistical wordsharter can summarize things better than a person.
– Frost
I don’t think, I know. I use these systems regularly. They do a great job of the summaries. I sometimes double check things, but it has proved unnecessary since the later gpt 4 level models.
Your negative feelings towards these systems don’t magically override reality.