Make sure to have alot of the creative models for tensor code: music (midi, wave), image gen, text gen, TTS and STT, and computer friendly like json and stream

Also python code artifacts can be the output. This allows you to use tensorcode for meta-programming rather than regular programming. Actually this is just algorithmic construction and it is compositional. That is, use template algorithms compositionally (they are away in the options list). Primitives are

• python primitives: if, while
• Extra primitive: do while, repeat,
• The 87 or so

Also have a list of design patterns that it can call to make isomorphic effects.

Select is more than a cfg. It is like a RIM Since he performs intelligent template matching (The call signature is the template)

Add support for Trees,

And don't ignore convenience synonyms like linked list, doubly linked list,

Intro to tc: The python variable scope is a graph and now neural networks can build representations of it, traverse it, and add nodes (define variables and functions)!

• Buy ball in socket joints with slots in them to allow greater freedom for the shoulder joints
• tensorcode.function(x) I know Y is related to X tc.caused_by(x)
• Make sure I've filled out the application for part time service of my circumstances allow it. I don't think i will though

The tensor code frameworkGives me the idea that just like we have English language dictionary's, we should write a python dictionary of all the reasonably possible objects that would be encountered by a machine learning system. Then it's also important to recognize that in the real world most objects don't fall cleanly into just one category, so whenever the tensor code policy creates a new object in its world model to match when it believes exists in the real world, it might not simply create an instance of one of the dictionary entries, but instead it creates an entirely new type and new object with relevant keys assigned based on the salient attributes of the object in the real world. Then there's should also be a is_fuzzy_duck_subclass method that performs an inexact duck type matching between objects (like, for example, an abstract dictionary entry, and a real object). This combines fuzzy logic with highly interpretable modeling

Integrate more natural architectures just as you would for any other deep learning system. The tensor code portion can

1. Directly read observations and write actions
2. Write code that handles io

Teach computation to use the repl to

• long term memory and Kb query
• Add subnetworks for encoding decoding or thinking