Calculating
Last page | Imprint | Next page

Beta Software!

But a good one.

Constructs that Java needed

Question-dot.

c ?. setDoubleBuffered(b);

=>

if (c != null) c.setDoubleBuffered(c);

Zero Training Time Neural Networks

New invention.

Paging The Reflex

 

Unix just doesn't understand me...

Huh.

Chess Board Recognition

Coming up.

You can help train desktop image segmentation

Giraffe v3—now stores information

Building a simple translator in agi.blue

The demo is word-based, but we have a function for efficient phrase-based translation too.

An array with O(log n) time for all operations

I invented a new data structure (I think). It's an array (variable-size list) where each add, remove, update and random access completes in O(log n) time.

I call it a Log-N array. The implementation uses an order statistic red-black tree. This could be nice for handling long token lists during JavaX transpilation.

Today I wrote a whole bot

...live on stream. Sadly, frame rate dropped to 6 frames a minute at some point. Hopefully fixed next time. Video here. Code made.

Live Coding For Nerds

Stefan Reich builds agi.blue, the Google-killing AI database & chat bot making platform. New functions built: Inter-slice links, Delete page, Checkboxes.

I did my first live-coding stream

...and now it is lost because I didn't click "Save broadcast". Was fun though, thx to Lemon for participating.

And of course, code got made. Specifically, agi.blue slices now have owners.

"Be Thankful"

Randall Brown & band (including me).

Why AGI.blue beats Google

Because White-Box.

Let's make programs

...that understand everything.

Revamping agi.blue

agi.blue now has a query language that looks like this:

word types / is plural of / $x
$y / is a / $x
return $y

This example queries the database for a random word type.

GBot googles in your Discord

Hey ho, we got our bot approved. You can invite it to your Discord group if you have one.

Coarse Procedures & Power Words

Coarse Procedures. Power Words.

Concepts we need.

AI Does Simple Reading Comprehension

Consider this text:

First, I wake up. Then, I get dressed.
I walk to school. I do not ride a bike.
[...]

Question 1: What happens first?

  1 Get dressed   2 Wake up
  3 Eat lunch     4 Walk to school
  
[+4 more questions]

Program that solves this. (One screen of code as always.)

The engine is pretty generic. It works by correlating words in the text, the question and the answer. The only additional information we put in is a list of "important words"—this time in the form of regular expressions:

"(?<!not )like", "not like", "dinner", "school", "bed"

The first expression matches on "like", but not on "not like" which avoids some confusion.

(Why do we have to put in these regular expressions? Well, the AI is young, so we help it out a little bit. The semantic knowledge the AI gathers in these exercises will eventually allow it to function without human help.)

Now on to more challenges!

Last page | Imprint | Next page