This Word Does Not Exist

For the casually curious, the reason it frequently pops up real words with fake definitions is (roughly speaking) one of the steps in the ML code is a dictionary check which tries to stop the model from generating existing words, since there’s a lot of combinations of word like phonemes that are actually already words, so the training data is going to naturally generate a lot of existing words without a rejection stage, and the rejection stage is based on a dictionary.

I’ve spent a fair bit of time working with this code (I’m not the original author) as I sort of decided it was a small enough ML project it would be a good way to learn how to work with some real world practical ML code, and it’s been quite educational, mostly in the realm of Torch and other ML dependencies and how important it can be to keep on top of them. My own fork runs well in a current python environment on my M1 MacBook Pro but due to architectural limitations I haven’t yet been able to get the code in a format where I can benefit from the M1s ML acceleration hardware, mostly due to the Transformer library using torchscript and preventing a lot of Tracing/Scripting that Apple’s libraries need to use in order to generate the accelerated model code.

It’s been pretty fun and I’ve actually got a prototype product that uses my code, still noodling at the best business model since charging for “here’s this list of , I’ve done the hard work for you” is a little awkward when you’re preprocessing all the data so that you can serve potential customers in a timely way. There’s a lot of background processing, Trademark Databases, social media scanning, fast DNS based and slower Whois based domain name lookups for the dozens of TLDs gTLDs, and ccTLDs… far too much to do real time after generating a fresh result from the ml model itself.

If anyone is curious feel free to contact me, (info in profile) it’s not yet in a position to link to a website since the UI will naturally depend a fair bit on the eventual business model may or may not come up with.

Neat to see it on the front page again.

Well, yes, to put a finer point on it. I meant, if it has been spoken [in the context of a conversation and understood by more than a single person in said conversation], it is a word. I don't think it needs to be defined in a dictionary or agreed upon by a committee in order for it to become an official word. Even if you don't agree with it.

A dictionary is a good place for someone to confirm the meaning of a word, and in the context of a professional written piece- it's probably wise to stick to words that can be looked up in this manner. But, on the other hand, YOLO.

Claims "mycobiology" is not a word.

slaps own face in disgust