Show HN: Research AI – AI Writing Assistant

When you're writing anything non-trivial, you ought to have something new and/or interesting to say, otherwise you're just adding noise to the environment.

I can see this being useful in finding ways to rephrase things, improving the structure or flow of a document. But this seems to be suggesting content rather than form or format. Adding content to your writing because it commonly appears in writing similar to yours seems to go against the idea of new and interesting.

That may be good for politicians, but I can see this being misused to 'bulk-up' a small number of ideas or facts into a full paper. On balance, this could easily do far more harm than good.

Now they have my CC number. There is no way to cancel your account. There is just a gmail support email... Let's see if I get my $14.99 back.

> This is Johnson Smith, writing to ask to work as a seasonal employee for Enron. I am currently available, and I think I would be a good fit for your position. You said you were looking for somebody with “strong Linux Unix skills,” so I am qualified. I am mainly self-taught, not having taken any college classes, but I am an avid learner, and I am comfortable using free resources to teach myself the skills I need to reach my goals. I worked as a software engineer at IBM for 8 years, primarily on laser sharks.” I am sure you will find that I am not just trying to fulfill a need, but that I am qualified for this position.

I'm not particularly impressed by the output.

This is impossible to guarantee with current language-generating model technology, and will likely make IP lawyers salivate and review boards cringe. (case in point, the recent GitHub Copilot outputing-unique-code-verbatim snafus)

Not even OpenAI guarantees that generated output from GPT-3 is original, just that they won't enforce copyright.

Or perhaps the inflation of filler text content will make people only refer to their trusted sources - well established publishing authorities.

Besides, the tool seems to sequentially predict sentences given a context. This way of writing contradicts with usual scientific writing, where people usually write the skeleton of the paper first, then add details and adjust the logic, and finally refine the wording.

Considering the reasons mentioned above, I don't know how it can be useful as a research AI.

I'm neck deep in research and ABD on a CS PhD with a focus on language modeling/NLP and human-centered design...and this kind of stuff makes me a little embarrassed to say that. Anyone involved in AI right now needs to gutcheck that they actually think they're doing something to make the world better and not just noisier or more "sticky" or whatever.

This is a service I would love to use, but I'd like to have a free taster before committing my payment details. I know you do a 7-day trial, but at least, for me, there is a big barrier to commit my card details to a random service. Maybe worth taking inspiration from Grammarly by offering a limited free service to get people hooked? Or even letting people sign up for the 7 day trial before giving a way payment details?

Somehow I am not assured.