Social media is fun, right? All those goofy TikTok videos, hilarious Instagram Reels, and brutally nasty sub-tweets.
Yeah - social media sure does remind us that a lot of people are looking for attention and will get it anyway they can.
Fortunately, that’s where moderators come in. Unfortunately, many of those tasked with cleaning up the social landscape often have to carry the burden of dealing with all that toxicity. (For a deep-dive, The Verge ran an excellent series of articles on the topic a few years back.)
OpenAI says it has a fix.
In a blog post, the company claims GPT-4 can be used to analyze platform content policies, and then quickly identify potential violations. OpenAI says another benefit is that the moderation process can be speeded up, and that it might ultimately spare some human labor.
OpenAI says once GPT-4 has ingested a platform’s terms of service, it can be trained using a “golden set of data.” In layman’s terms, that means feeding in examples of what GPT-4 should be watching for so that it can then start tagging such content when it comes across it.
The company notes that, once trained, GPT-4 is able to make policy judgments - even when the set of rules it’s working from could be interpreted as nuanced.
OpenAI also claims platforms that use GPT-4 to call balls and strikes can also ask it why it has made the decisions it has, and then rectify any areas on which it’s falling down.
All together, this is one of those huge if true moments; if GPT-4 really can help with content moderation - particularly at scale - then there may eventually be real benefit for those tasked with keeping an eye on the behavior of those who hide behind a handle.
Even after all these years, platform moderation is still one of the thorniest issues on the social web. It’s arguable that no social media company has gotten the free speech balance right (just look at the Twitter/X situation). If human beings are still struggling with moderating human behavior, how certain can we be that the machines will figure it out?
When Google rolled out its generative AI Search experience (dubbed SGE) as a public beta a few months back, it was pitched as a tool to provide you with informative machine-written responses to your Search queries. You know, so you don’t have to trawl through all those blue links like some kind of neanderthal.
Now the company has announced a new web-based AI tool - SGE while browsing. (Google and its snappy product names…)
As Google explains it, SGE while browsing detects when you’ve navigated to a particularly dense article and offers you some help with digesting it. Think an academic paper or a long article that is absolutely loaded with facts. Basically anything that would take you a long time to read and by the time you get to the end, you’ve probably forgotten much of it.
With SGE while browsing, you are presented with an “i” icon which, once tapped, kicks Google’s generative AI into gear. After a few seconds, Google presents a slide-over summary, featuring all the key points that it thinks you need to know. Underneath that is an Explore section, providing easy shortcuts to various sections of the article.
While much of this can be done with Bard or Claude, the promise of SGE while browsing is that you don’t have to copy and paste reams of text from one place to another. With SGE while browsing, the summary and key points are all right there.
Personally, as someone whose brain taps out when enduring a particularly dry long-form piece, SGE with browsing is the kind AI tool that just makes sense; give me the cliff notes, and then point me to where my attention is really needed. (Can we do the same with Zoom meetings yet?)
For those who want to try it out, SGE while browsing is Initially available as a Search Labs feature via the Google app (iOS and Android). Google has said it will also be making its way to Chrome for desktop in the coming days. There’s no word on whether it will eventually make it to mobile Chrome.
With the return to school right around the corner, there will be plenty of people taking a fresh look at how AI tools can make their lives a little easier. This may well be one of them.
OpenAI's GPT-4 aids in content moderation, potentially easing human moderators' burden, while Google's new tool summarizes dense content for quick comprehension.