That's what I got from this too. Copywriting is about the only thing ChatGTP _is_ good for.
But it's also pretty darned good at doing things that might not be as popularly relatable. There are some teams reporting that GPT4 can outperform Mechanical Turk labelers for many corpus labeling tasks, for example. That might then significantly reduce the cost of developing NLP products that may or may not use a GPT-series network as part of the production implementation.
ChatGPT and GPT4 can also do a pretty darned good job at producing synthetic data. Other methods may be able to produce higher-quality data, but at typically at the cost of a much larger development effort. That could, for example, be useful for privacy-preserving machine learning where you want to develop a model that's tailored to a specific task, but you don't want to (or, if you're lucky enough to live and work in a sensible regulatory regime, maybe even can't) use actual sensitive data to train the model.
The underlying embedding model may or may not outperform BERT for specific tasks. I haven't poked at that issue much, but it wouldn't surprise me.
High risk would be reading legal documents and recommending what the lawyer should do.
If you wanted to jumpstart a competitor to Goodreads for example, GPT can do elaborate book summaries and character discussion (etc). Use that to get a foundation of content (so the first users don't arrive to a nearly empty shell of a site).
This won't be possible for much longer, and bots like GPT will become drastically more difficult to build (regulation, content restrictions, et al.), and also likely more expensive to utilize for valuable purposes (they'll maximize its commercial value, once they understand fully how to bracket all of that and extract properly). GPT got out of the barn before the door was closed (before most realized the door was even open), a few more might make it out before the vast web of restrictions set in (whether content services like Reddit & Stack trying to block use for building services like GPT, or big media empires doing the same, or image/video owners trying to preserve the value of their works, or governments regulating, and so on).
GPT is in the early wild like Uber was, before governments around the world locked down on that premise, which created a captured market in many locations (whether by Uber or a local competitor that beat Uber there). That open situation never lasts. And when other (financially interested) entities see something like GPT's commercial exploitation potential, the barriers go up, they all want a part of what's possible. There won't be many GPTs in fact, for the exact same reason there aren't many Ubers. There will be a few prominent generalized gigantic bots/services that were early (were able to be built up before the lockdown of content access & consequences changed); and then there will be a lot of highly focused, niche bots that splinter the market over time and do more narrow things extraordinarily well. It'll also sort of follow the search market in this regard.
And after the restrictions are put into place, the highly focused (narrow) GPTs will be backed by large piles of venture capital money (which will be necessary, especially in medicine, law, and other similar high value, high regulation, high risk fields). It'll end up costing a lot of money to build the highly focused GPT type services (and they'll outperform the generalized services at what they do; just as Pinterest / Linkedin / Twitter / etc have been better at what they specifically do than Facebook can ever be as a generalized service; and for the same reason Google can never kill Wolfram).
This sort of plan seems just slightly above the level of SEO spam. I've played with ChatGPT book summaries and they're terrible in the sense of being very banal as well as being full of errors.
Plus creating a site that is compendium of answers ChatGPT/etc gives to questions would have to be a poor quality site given that people can just ask the questions directly to ChatGPT.
Explored this myself, and it's a definite no. GPTs cannot do this reliably (we've tested multiple LLMs). Simply put, it's because it's a rule-extraction problem, and not a generative problem, and rule-extraction situations are typically too sensitive for a "99%" solution like LLMs. Basically, that 1% failure rate is enough to screw any gains you get from the 99% successes. For years, all I've ever heard from the DS space was to always be mindful that the model you apply to a problem fits the nature of the problem. The past few months, this has somehow become a lost wisdom. "LLMs for everything"
For this reason, HN comments lately have had my pupils permanently fixated to the top of my eye sockets.
(its real strength as a writer over humans other than being low effort is niche stuff like composing entire sentences with alliteration)
Written by .... Microsoft Research .... owned by ... the company that just invested ~$15b into Openai and their largest partner to date
Conflict of interest level: over 9000
If you're claiming there's been academic fraud in the paper, feel free to cite your sources...