A Devil's Bargain with OpenAI | Hacker News Reader

A Devil's Bargain with OpenAI

theatlantic.com

36 pointsfortran7713 comments

I wonder if there's another reason for Sam brokering deals with lots of major news outlet. Surely OpenAI won't try to exert any type of editorial control over these publications. That would be absurd!

I'd be more worried that OpenAI becomes the default method of accessing their stories, and then Sam has complete editorial control.

Pretty sure they had something in the contract that OpenAI can train on their data, so it is an afterwards legitimation of something already performed (allegedly)

A search engine, that returns answers, and cites sources.

You're seeing the tech peak, there is no AGI coming.... we didn't feed it all the works of humanity and have some great problem solving overlord emerge.

Better search.

Sam Altman, leather jacket, motor cycle, tank of sharks.

It's not even a good search engine. I was constantly asking for sources, so I figured I would just add it to the "Customize ChatGPT" to always include the source when possible.

Giving instructions to a model seems to be a monkey's paw simulator. Yeah ok, it always gives me a little section down the bottom with sources now, but they are either unclickable "links" (they are stylized as links but aren't a hyperlink), a fabricated URL (the page doesn't exist and never did according to Internet Archive) or just completely nonsensical and have no relation to the material.

I've experienced all of these today using ChatGPT 4o.

Yeah, source URLs have always been hopeless to get out of ChatGPT. I suspect not because it couldn't, but because they wouldn't: citing sources would open up for copyright infringement lawsuits. So it makes sense that source URLs can be available when quoting the Atlantic now: they have signed the deal and made the pact and can't come complaining.

In the current LLM tech if you need sources reliably, you need to start with search first and then summarise that with LLMs.

ChatGPT without web browsing skips the search part and ends up hallucinating URLs

+1 exactly right, I believe.

Perplexity comes close to what I want, but having a superior product does not, unfortunately, guarantee that they will stay in business.

Perplexity even gives real time status information that it is first searching the web,

as weird as it sounds, tiktok is probably a better search engine for people on that platform from what i understand.

https://archive.is/fved1

Maybe OpenAI should buy news inlets, like the Associated Press, instead of news outlets. OpenAI needs the news collection part, but the distribution part is dead weight.

This piece seemed to start out scathing, but end weak. Hopefully not with the effect of a pressure relief for valid objections.

The 'clear reason no to trust their output' the author uses as an argument against it, is the reason all want it. People know not to trust llms completely but right now these 'news' institutions are trusted too much when in reality they are mouth pieces to special interest or governments, they have no obligation to tell the truth either.

Reply on news.ycombinator.com