OK, there's a lot of negative comments here, but this is a great idea at a high level. It really is hard to figure out where to do a thing and it's also very easy to get phished. If they can figure out how to help people get all the services they're eligible for that'll be an amazing improvement for the people.
I agree in general, but how do you resolve the problem that people will take the output of the chatbot as authoritative even when it could possibly have errors or miss important nuance? People being lead astray with information from the government could lead to very serious consequences, up to and including imprisonment.
Honest question, I had the same issue when my work wanted me to stick a chatbot in front of our HR portal and I never resolved it to my satisfaction.
> but how do you resolve the problem that people will take the output of the chatbot as authoritative even when it could possibly have errors or miss important nuance
You have to measure that problem against the status quo, which is that current government websites are very hard to understand and leave people with very little idea of what’s going on.
no. humans can be held accountable and learn from their mistakes.
they can be taught where they went wrong.
if a pattern emerges they can be moved to a role more fitting for them. or removed from a project entirely.
we can judge their effectiveness from past performance.
we can put them in less important roles and gauge whether or not they should be moved up.
so, no, pretending that the correct route forward is to “remove humans from the equation” for a bot that doesn’t learn, is never held accountable, and not held to the same standard as humans is silly tier thinking.
I have read the opening chapters of the book, and honestly I had to put it down after reading 20 percent of the book. I just don't see any new information that I had already known (rightwing effect, propaganda, Fox/Breitbait nonsense, bla bla).
I used to believe Trump winning the 2016 election was a product of rightwing propaganda and fringe theory, but with the 2024 election, I had to revise my belief. What if this is the mainstream mindset of Americans all along, and Trump simply tried to capitalize on that?
If I instruct an employee to prioritize my profit in a way that opens both of us to criminal prosecution, there is a non zero chance of them whistle blowing or even cooperating with the authorities to prosecute me.
depending on which model one is using, and the error the employee needs to be retrained on, i strongly disagree that you can retrain an llm as easily as you can a person.
particularly most of the commercial models.
again, i’m sure we can all come up with a thousand “hypotheticals”, but i’ll state again, “entirely removing human” from the situation is silly.
particularly as even the ceos of frontier producing models will each and every single one tell you to never trust their model.
“our model is smartest thing in the world…”
…next breath..
“wait, you trusted our model? that was silly of you. always double check it”
We shouldn't use an LLM because...LLMs never improve?
This seems like a great source for domain-specific RL, but even without that, we can expect there to be higher-quality LLMs that can be trivially swapped in within a year, if not a week.
Okay, but I don't understand what you're getting at with them not learning? They acquire more information, get better at providing accurate and relevant information, and acquire new skills. What is the learning they're not doing?
Maybe a human should be available as a backup or something, but not because LLMs don't "learn" -- the system seems to learn in all relevant senses.
(I'd also object to the no accountability, but another subthread is already on that.)
maherbeg · · focus · HN ↗
gthrow12345 · · focus · HN ↗
Honest question, I had the same issue when my work wanted me to stick a chatbot in front of our HR portal and I never resolved it to my satisfaction.
rayiner · · focus · HN ↗
You have to measure that problem against the status quo, which is that current government websites are very hard to understand and leave people with very little idea of what’s going on.
andrewflnr · · focus · HN ↗
kulahan · · focus · HN ↗
toofy · · focus · HN ↗
they can be taught where they went wrong.
if a pattern emerges they can be moved to a role more fitting for them. or removed from a project entirely.
we can judge their effectiveness from past performance.
we can put them in less important roles and gauge whether or not they should be moved up.
so, no, pretending that the correct route forward is to “remove humans from the equation” for a bot that doesn’t learn, is never held accountable, and not held to the same standard as humans is silly tier thinking.
buriram · · focus · HN ↗
As an individual, yes. As a population, no. They elected a conman twice and would rather burn down the country rather than changing their view.
intended · · focus · HN ↗
buriram · · focus · HN ↗
I used to believe Trump winning the 2016 election was a product of rightwing propaganda and fringe theory, but with the 2024 election, I had to revise my belief. What if this is the mainstream mindset of Americans all along, and Trump simply tried to capitalize on that?
intended · · focus · HN ↗
buriram · · focus · HN ↗
olmo23 · · focus · HN ↗
bad bots can be retrained or shut down far more easily
rapidaneurism · · focus · HN ↗
I don't see such a path with an llm.
toofy · · focus · HN ↗
particularly most of the commercial models.
again, i’m sure we can all come up with a thousand “hypotheticals”, but i’ll state again, “entirely removing human” from the situation is silly.
particularly as even the ceos of frontier producing models will each and every single one tell you to never trust their model.
“our model is smartest thing in the world…”
…next breath..
“wait, you trusted our model? that was silly of you. always double check it”
dansquizsoft · · focus · HN ↗
lkbm · · focus · HN ↗
This seems like a great source for domain-specific RL, but even without that, we can expect there to be higher-quality LLMs that can be trivially swapped in within a year, if not a week.
toofy · · focus · HN ↗
i didn’t say this at all. did you respond to the wrong comment? i was responding to the person who said:
> … remove humans from the equation, then.
lkbm · · focus · HN ↗
toofy · · focus · HN ↗
> We shouldn't use an LLM because...LLMs never improve?
lkbm · · focus · HN ↗
Maybe a human should be available as a backup or something, but not because LLMs don't "learn" -- the system seems to learn in all relevant senses.
(I'd also object to the no accountability, but another subthread is already on that.)
rerdavies · · focus · HN ↗