How can I trust OpenAI for my generative AI platform given the NewsCorp deal

Community
1

The newly announced deal with NewsCorp for training OpenAI is deeply troubling. NewsCorp is not a reputable source for journalism. They are hyper partisan and often untruthful. The 3/4 of a billion dollar judgement for libel regarding voter fraud should have told you. This decision makes it likely that we will look for alternate generative AI platforms.

2 Likes
3

The deals OpenAI has made with the various news sources and forums will most likely only be relevant for people who use ChatGPT not the API.

1 Like
4

Most media companies are owned by a handful of companies now.

I’m sure OpenAI will be looking at adding other news organizations.

1 Like
5

I am not sure GenAI for the sake of GenAI is going all that interesting for real world applications. To some extent, you need grounding, and whether you accomplish that with your own RAG data or something else is the question.

All real world applications I look at need some sort of traceable citation for the data. Otherwise, the hallucination risk is too great. This is the case for finance and legal which is what I am working on.

We probably crossed the point of diminishing returns for model size, and with current vocabulary and language generating capabilities, we are probably better off on specialized data/models. One model to rule them all seems like too much of a compute resource waste for little gain.

2 Likes
6

WSJ is considered to be one of the most reputable news sources in the world. Their news division is also considered to be somewhat left learning, if that helps. They hire a lot of elites from prominent universities.

It’s unlikely they will train based on the opinion side of WSJ.

1 Like
7

WSJ was purchased by NewsCorp and that means the reputation of the WSJ is now complete garbage in my view.

8

When it is something like Vox that recently signed on, that publishes an expose of what Sam Altman knew, did, and signed, you have to wonder if this agreement also includes a “non-disparagement” clause…

9

Consider the following: if the model replies with a reference about a news event would you rather have a source informing you about the origin or not?

I clearly prefer the first option and would consider the partnerships with news orgs a trust building measure.
But we have yet to see how this will actually be implemented.

1 Like