saternius

Joined 8 years ago

Karma

Submissions

Member since

Mar 20

Recent submissions

I could see how this comes off as linguistic phrenology. The point is not for this to be used diagnostically, but instead to analyze trends with the labels we use in language.

Using LLama2 to find links in mental disorders and ideological fixation mindgrid.io

4 saternius 3 years ago 2 comments

Comment · 5 years ago

You can check out our best attempt at a question generator at http://quickquiz.me, to get an idea of what type of questions a large language model (T5) can produce. Its the best model my team and I witnessed. We are still collecting data to generate more salient, and well structured questions. Its a work in progress.

Comment · 7 years ago

What we do at https://quillbot.com

Training:

Currently we just use a bunch of beefy desktop workstations for training (using Pytorch).

Deployment:

This is the vast majority of our cost, each time a paraphrase comes in we add it to a queue through google cloud Pubsub. We have a cluster of GPU (T4) servers pulling from the queue, generating paraphrases and then sending the responses back through Redis pub/sub. I think ideally we would have a system that makes it easier to batch sentences of similar length together, but this seems to be the most cost effective way for models that are too computationally expensive for the CPU that is relatively simple to put together.

Comment · 7 years ago

QuillBot (https://quillbot.com) is a full-sentence thesaurus that will rephrase any sentence you give it. Full disclosure, I am a co-founder.

Comment · 7 years ago

As a founder of an startup that uses AI to paraphrase text (https://quillbot.com), I find that this article is hitting a very valid point. As fluency enhancing tools become more prolific and higher quality, submissions will be more standard and thus there will be more unintentional overlap. Plagiarism detection software will become increasingly less reliable.

Comment · 7 years ago

Totally remembered our couch conversation. Appreciate the kind words and support. Hope your startup is doing well and I'm glad you found our website helpful!

Comment · 7 years ago

We considered removing recaptcha, since we are fully aware of how annoying it is, but we had to keep it because of the rampant bot abuse.

Comment · 7 years ago

Like I commented on below, the system is still imperfect. Its about a level 2 safety if compared to a self driving car, and it will not being doing stunt tricks any time soon. That being said, we are hopeful that it is only a matter of time.

Comment · 7 years ago

The system is by no means perfect. If I had to compare it to self driving cars, I'd say its a level 2 on automation safety. You should have your hands on the wheel when you use it, but it still makes driving easier.

Comment · 7 years ago

The main difference between Quillbot and spinners, is that a majority of people we've surveyed say they use QuillBot for suggestions on their writing. I'd give it a good chance on providing a better sentence structure if you are a non native speaker.

Comment · 7 years ago

Glad you agree. I suspect QuillBot is not ideal for fake reviews anyways since you would likely want a diversity of positive opinions rather than the same ones regurgitated. What I'm really excited to see is, where this technology goes in regards to education and writing enhancement. (For clarity, I'm the CEO of Quillbot)