Hacker News Reader

Comment by bubblewand | Hacker News Reader

This is also what OpenAI’s “safety” angle was all about.

“Ohhhh this is so scary! It’s so powerful we have to be very careful with it!” (Buy our stuff or be left behind, Mr. CEO, and invest in us now or lose out)

View on HN

Anthropic has been the most histrionic about this, with their big blog post about how they need to make sure their models don't feel like they are being emotionally abused by the users being the most fatuous example.

"This is obviously why only we can be trusted with operating these models, and require government legislation saying so."

They're trying to get government to hand them a moat. Spoilers... There's no moat.

To me, Anthropic has done enough sketchy things to be on par with the players from Big Tech. They are not some new benevolent corporation backed by SV

Many users don't want to acknowledge this about the company making their fav ai

I was taken aback when I recently noticed a co-worker thanking ChatGPT for its answer.

LLMs talk like people; there is nothing wrong with this. It's perfectly fine to be nice to something even if it isn't human. It's why we don't go around kicking dogs for fun.

I understand why people don't act polite to LLMs, but honestly I think not thanking them will make people act more dickish to other humans.

I like to think we can perceive a difference between a machine and a living being. I don't thank my bicycle for transporting me or thank my spell-checker for finding typos. I get that we are prone to anthropomorphize but was just something I found a bit surprising.

>I don't thank my bicycle for transporting me or thank my spell-checker for finding typos.

Neither your bicycle nor your spell-checker hold conversations and answer questions, neither of them is being used as therapist or virtual girl/boy friend, and neither's whole shtick is being trained on a ginormous human corpus to convincingly respond like a person.

I like to think we can perceive a difference between a bicycle and an something specifically developed and trained to pass for intelligence...

See, i have absolutely thanked my car. When i was broke driving a beater and it was below zero outside, practically every time it chose to start lol

I don't think it's that weird

I occasionally say please and thank you to ChatGPT for my own sake, not for the LLM's. They're sufficiently similar to humans that allowing myself to be a jerk subtly degrades myself and makes it more likely that I'm a jerk to real people.

From a year ago https://www.vice.com/en/article/telling-chatgpt-please-and-t...

People have been thanking Siri back 15 years ago, it’s just a reflex.

I missed this - which blog post?

https://www.anthropic.com/research/exploring-model-welfare

Gpt2- "too dangerous to release"

Oh funny, I forgot about that. But at the time it didn't seem unreasonable to withhold a model that could so easily write fake news articles. I'm not so sure it wasn't..

Either 3 or 3.5 also had "it tricked our researchers and tried to to escape so we had to dumb it down before release" marketing. What actually happened was someone asked it something like "how would you escape detection" it replied telling the person that, basically just regurgitating a regular sci-fi plot.

Reply on news.ycombinator.com