Comment by Der_Einzige | Hacker News Reader

Comment by Der_Einzige | Hacker News Reader

Wow to be charitable- you rebranded already existing language models (gpt-2, Bert, xlnet, etc) - and turned it into a SaaS... Though those models seem more effective than this service by a long shot.

Worse, even the best NLP models at the moment are clever hanses. Also, the huggingface "write with transformer" has far more hyperparamaters to play with which make it far far far better than this website.

I'm sorry to have to echo the other comments or come off as overly critical, but this service blows. I see zero reasons why I or anyone else would want to use it - especially since your competing against the very best NLP minds.

Next time, try to use models which might actually be effective. Salesforce released one called "CTRL" which if you had productized might have actually given you what you want (effective controllable text generation)

Thanks for the feedback!

Yes, the SaaS does rely on state-of-the-art NLG. I don't know why you think they are far better if we are using them as is. I am not offended at all; just curious where that big difference is.

Who says one can't compete against the best minds? Isn't that the premise of startups and progress?

If you are a fan of "CTRL", you should have liked ours a bit better.

Apps need to simplify and yet provide the desired experience for end users. From your feedback, it seems we need to try way harder.

Thanks!

For one thing, it looks like your decoding is not being done the way SOTA systems do it (usually nucleus sampling). Yours appears to be doing top-1 or top-N sampling, but please correct me if I'm wrong on this. Even if you are doing it with nucleus sampling - you need to expose paramaters to control the generation better. That way you have a lot of cover for it not working as expected in the form of responding to dissenting users with "just tune the parameters better"

The reaction you're getting from this system is so negative because it appears to be less good than write with transformer - which implements exactly what I'm describing.

For what its worth, this mostly isn't your fault. NLP is a crapshoot of hype and implementations which disappoint.

I find that "Clever Hans" is a great metaphor to describe the gulf that exists between current state of the art NLP and human intelligence.

Reply on news.ycombinator.com