Talk to Transformer - how neural network completes your text

Anything goes, but keep it seemly...

Talk to Transformer - how neural network completes your text

Postby dobrichev » Sat Nov 16, 2019 7:27 pm

https://talktotransformer.com/
See how a modern neural network completes your text. Type a custom snippet or try one of the examples.


Note that the only rule used for completion of the phrase is finding suitable completion words, one at a time.
No attempt is made to understand the meaning of the phrase submitted.

Here is the result when I put the initial phrase (in bold) from this post.

The BLISS solver is a product of some thoughts I had, following recent conversations with SpAce about the nature of "guessing".

It's not so much that we should be guessing. But we need to make things more explicit and precise, so that we are clear who we are and how we are measuring things in the world.

This is particularly important for things like performance analysis.

For example, if I am measuring the cost of a function, I should be able to specify the precise performance function (or more accurately the precise amount of computation for a given input) and have the solver check whether it agrees with the performance function. I should be able to specify the inputs and outputs (and, in general, the order of operations), and the solver should be able to do any necessary optimization, so that we are using all the data at the right places.

Each run gives different results.

Enjoy!
dobrichev
2016 Supporter
 
Posts: 1784
Joined: 24 May 2010

Re: Talk to Transformer - how neural network completes your

Postby tarek » Sat Nov 16, 2019 9:02 pm

My attempt:
Mathimagics made a reference to the Venezuelan Beaver Fish and the story of the man who hunted it, but he's wrong. The reference is to the Mexican Crayfish, the most common freshwater fish species in the world. The fish that was hunted in the story was indeed a Crayfish, but the story of the hunter and the fish that inspired the story was not the Mexican Crayfish.

This is not an accident. There are numerous references to animals in the book that were not the actual animals that were in the story, for example there is a reference to a "sea dog" (sea lion) and a "catfish" (fisherman's hook). The story in fact mentions no fish at all.

The reference to the "crayfish" is actually to a species of saltwater crayfish known as the "sea dog". The species was originally native to Mexico and was common in the waters off Texas and California. The crayfish is an important

Even the neural network knew that the reference was … :lol:
User avatar
tarek
 
Posts: 3745
Joined: 05 January 2006

Re: Talk to Transformer - how neural network completes your

Postby Mathimagics » Sun Nov 17, 2019 8:36 am

Ok, that's insane! :lol:

But as everyone knows, you can't make cheese with a crayfish! 8-)
User avatar
Mathimagics
2017 Supporter
 
Posts: 1581
Joined: 27 May 2015
Location: Canberra

Re: Talk to Transformer - how neural network completes your

Postby HATMAN » Sun Nov 17, 2019 2:02 pm

I'm a electrical power system engineer so from a proposal document I've just written I put in:

Beneath the Grid Code there are a set of

After about ten attempts I got something that is in context and almost makes sense:


Beneath the Grid Code there are a set of rules that govern how the Grid is going to be built and operated. The first two of these are called "Preferred Route" and "Locating an Alternative Route". These two rules provide the means by which the Grid can be designed in a way that allows for a reasonable amount of grid redundancy. This redundancy is not simply an exercise in trying to provide enough redundancy, but rather a means of making sure that the Grid is a stable and reliable service.
For example, a route that is very close to a single point of failure, and so there is a high probability of a single failure, will have a higher probability of being chosen as the preferred route. On the other hand, a route that is much farther from the single point of failure, and which is likely to be hit by lightning or hurricane force winds on the average day, will be selected as an alternative route to keep things simple and efficient.
The third rule
HATMAN
 
Posts: 271
Joined: 25 February 2006
Location: Nigeria

Re: Talk to Transformer - how neural network completes your

Postby dobrichev » Sun Nov 17, 2019 7:26 pm

This is close to the reality too. Third attempt.

HATMAN's Sudoku contributions to the Open Source community will be featured in the new "Reverse" chapter of our upcoming book "Linux HATman: The Official HATman Reference", to be published by O'Reilly in March 2014.

"Open source HATman"

"Linux HATman"

"Linux HATman Reference"

"Reverse" chapter

Linux HATman (formerly known as HAT), is a complete solution for solving word problems, word problems with multiple answers and a wide range of word problems with multiple levels of difficulty. It can be used for both beginners and experts. Its intuitive interface allows the user to focus on solving the puzzles while a computer solves the problems for him.

The most advanced features are:

Word search:

Multiple answer:

Ladder (one answer per puzzle)

Ladder (multiple answers per puzzle)

Multiple selection:


The only error I found is the resolving of the word problems instead of worLd problems. :D
dobrichev
2016 Supporter
 
Posts: 1784
Joined: 24 May 2010

Re: Talk to Transformer - how neural network completes your

Postby HATMAN » Mon Nov 18, 2019 4:13 pm

One obvious use is to assist schoolchildren doing essays.

A more useful one is: when you are thinking on the edge of an idea, plugging things in here may help with possibilities.
HATMAN
 
Posts: 271
Joined: 25 February 2006
Location: Nigeria


Return to Coffee bar

cron