I tried to post this to Reddit but for some reason it keeps getting removed by filters. I normally just lurk on Lemmy but I decided to make an account because I’m so over that stupid platform.
ANYWAY
Last night on my flight home I noticed that the person in front of me was using ChatGPT. I got nosy so I looked closer and found out that they were actually using ChatGPT to write the content for a program of a sustainability and climate conference. They specifically included a section about “dirty” energy sources and the harm that results from energy producers supplying data centers. I was floored.
I am not sure if this was the final product or just a draft, but I find it pretty gross and ironic that the AI that is fueling the climate crisis and responsible for the rapid development of data centers that are doing irreparable damage to local communities and ecosystems is being employed to create the materials for a sustainability conference. Maybe start with looking in the mirror first?
I looked up the company, and turns out this program might be part of a Keynote Livestream that is featuring the Mayor of San Fransisco lol.
Oh that could be me… I work at a major organization that directly interacts at the highest levels of climate policymaking. You know what I recently learned from our comms team? Policymakers are asking AI about the options in front of them. So they advised us to feed as much data as we can into these AIs, so that when policymakers ask AI about their options, they will not only get the lobbyist answer. What a world to live in.
That’s not how AI works though. You can’t feed data into an AI in a way that affects the answers other people get. It’s not learning anything in real-time like that.
If you want to “feed data” into an AI system all you can really do is publish that data publicly and wait for it to be scraped up into their training data.
Edit: I’ll admit that I overlooked the fact they do add conversations to training data but a single user is going to be weighted VERY low. Especially if their data looks like an outlier.
Seems pretty unlikely to me that you could really influence the model in a meaningful way like that.
Better off focusing on public/authoritative data that will get weighted a lot heavier.
Accurate, perhaps they were referring to SEO and ensuring that web crawlers from OpenAI, Google, Anthropic, or other corporations stumble upon their perspective on things during the collection process for their next training dataset.
But unless you work at one of these LLM companies, you cannot influence an already trained model directly at a macro level. You can only tailor the answers it provides within your own account settings, which does not affect the output for anyone else.