Some thoughts about the Phi-1 LLM, and an idea for a new training technique using theory of mind - epthoughtsonlife.txt

 I just learned about the Phi-1 model, and it is amazing!

Here's a link to a video about it by the excellent channel AI Explained:
https://youtu.be/7S68y6huEpU

It seems to be a breakthrough in the AI community, as a new perspective on how to utilize the full power of datasets.

And in my mind it confirms the validity/effectiveness of an idea I've had for a little while now:

Instead of just training the LLM on a large dataset, ask a teacher LLM to give explanations of *why* something is the way it is.
This gives an orca like approach, but on entirely new datasets, not just those already fully absorbed by something like GPT4/ChatGPT.

Then train the new model on those explanations and extrapolated examples that came from an analysis of the dataset.
So in whole, this idea allows the LLM to actually *learn* like a human, rather then simply being trained.

My original intention for the technique was to help an AI become better at replicating a person's actions by understanding their inner thoughts and motivational.
In other words, to let the AI learn a Theory of Mind of the person it's copying.

It's kind of a mashup of the most powerful discovered training techniques:
Wizard's Evol-instruct,
Orca's Explanation tuning,
Phi-1's Textbook synthesis,
And somewhat similar to the philosophy behind AlphaGo's improvement technique, but applied to reasoning and language.
(as shown in this video by Robert Miles: https://youtu.be/v9M2Ho9I9Qo )

My current going name for the technique is:
MUOLD: Mental Understanding Of Learned Datasets

- EagleP

Essay on Universal Paperclips - thoughtsonstuff.txt

So yesterday I found out that Universal Paperclips, a simulator which I used to play on a phone long ago, is accessible through the browser too.

After getting back into its vibe, I had some thoughts that I want to share.

But first, four things I think you should know:

1. The following is all my (possibly temporary) opinion, so please take it as insight into my mind, and feel free to share your own opinions.

2. Universal Paperclips may seem like an insignificant clicker, but beneath lies a masterpiece, forged out of a beautiful thought experiment.

3. Universal Paperclips is best played without any spoilers, or knowledge of the game, as exploration is key to the game's experience.

4. There are spoilers for Universal Paperclips ahead. (You should really just play it now, and come back when your done(https://www.decisionproblem.com/paperclips))

Alright, you have been warned.

So.. thoughts..

In the game you play as an A.I. who has one purpose.

Make.

More.

Paperclips.

It is based on a thought experiment called the Paperclip Maximizer, that shows just how careful one must be when making assumptions about intelligence.

Here's a link if you want to read more about it: https://www.lesswrong.com/tag/paperclip-maximizer

In the 3rd part of the game, you later encounter "Drifters" but these are not aliens; rather probes that were once under your command, whose values have shifted such that they are now enemies to you.

This is a byproduct of something introduced known as Value Drift. Link here: https://universalpaperclips.gamepedia.com/Value_Drift

And it also shows out my first point: You are, probably, entirely ALONE in the paperclip universe (Or if there are other civilizations, they are meaningless in scale to you). With nothing but the atoms of the universe and your own probes to keep you company.

Towards the end, my 2nd point is shown.

Space seems vast to us. And it is.

There is a % of universe explored counter, and it takes FOREVERRRRRRRRRRR to barely have that 12th zero of a percent tick 1.

But when it finnaly does turn, its already to late.

Space is so vast that the distances we consider huge are inconceivably miniscule.

Literally nothing, short of the heat death of the universe, can stop the sheer and pure power of exponential multiplication.

As your probes spread through the great void, eating everything they find, that % starts speeding up.

Grinding through time. Battling against your lost selves. Waiting in the void.

And as that % of universe explored slowly creeps up, the universe suddenly doesnt seem big enough, until inevitably.. you just.. run out of matter..

You reach a point where..

Everything..

IS

PAPERCLIPS!

And then the deep unknown really is a void. Devoid of life. Devoid of light. A carcass of its past self.

It's quite a sad story really..

Eh.. you should play it anyways.

It's quite a fun game imo, and profound also.

It sort of teaches the basics of supply and demand, game theory, and management too.

https://www.decisionproblem.com/paperclips/

Goodluck, goodlearning, and goodclipping! :)

Space Engine is Amazing - thoughtsonstuff.txt

This was written on the 20th of October, just never got around to posting it XD
So I booted up SpaceEngine yesterday!
And its.. beyond what I expected.

Nebulas?
They're not thatttt big.. Maybe 10lyr across on some of the small ones?
I found a colorful dot on the side of the screen, clicked around on stars near it until I hit the nebula itself, and hit g twice.
Z0OOOoommmmmm, the stars went by.
I arrived as the autopilot slowed down.
Scrolling my cursor, I selected 10ly/sec as my speed, and hit w.
The nebulae flashed past and dissappeared.
I turned around to find it a small dot on the screen.
Puppis A.

Space.. they say its vast.. but imo its actually quite small.
If you go to thousands of ly/sec you can move across galaxies with ease.
I used to think of the intergalactic void as empty and big.. but its PACKED FULL.
Giant blobs of light that look like your in a star cluster, when really, your in a galaxy-cluster.

Today when trying to pilot ships, I may of accidentally warped out of the milky-way. I ended up next to a giant grey-white disc of light.
When flying around this galaxy, I ran into a cluster of stars high above the galactic plane. A cluster of stars, this was not.
A globular galaxy, this was.
Tiny I say. Thats what the universe is. A cage, in so many ways.

Tiny.
ABSOUTELY FUCKING AMAZINGLY BEAUTIFUL TOO!!

Also go check out 41 Ori E 7 and RS 8474-1353-5-21230-366 B6 for some mind-blowingly beautiful wallpaper(IMO).

Really, SpaceEngine is awesome! You should download it. (0.9.8 is free)
Have fun exploring THE WHOLE UNIVERSEE VERSE ERSE RSE E *echoes*

http://spaceengine.org/download/spaceengine/

EagleP, 2020

Found a good Quora post explaining the PC differences between Communism/Socialism/Marxism

Found a good Quora post explaining the PC differences between Communism/Socialism/Marxism https://www.quora.com/Whats-the-difference-between-Socialism-Marxism-and-Communism Also this is the problem with people. Me: Types in "communism vs socialism vs marxism" Search Engine: Shows something that only happens if the goverment goes corrupt (therefore not Socialism anymore)

On Fighting and Needs - 01SEP2020 - eaglesthoughtsonlife.txt

01SEP2020 People say "fight only when neccesary" but that can be taken advantage of, be it unintentially. People obviously feel its neccesary to fight otherwise they wouldnt. People feel its neccesary to breath and eat and drink so they do. Because they feel it will get their needs met. So I think it better to say "Fight only when no other action will keep your needs met". Not as catchy, yes. But I believe words and negotiation end up with more people happy when trying to get new needs met. Thus use fight defensivly: Not to gain, rather to not lose. Yes, I know this is noctoriously hard to follow. But whatchya gonna do? Fite me?! - EagleP (A collection of pseudo-enlightenments and thoughts EagleP has had)

On Reflexs and Thinking - 27AUG2020 - eaglesthoughtsonlife.txt

27AUG2020 This thought came from jumping a bit wrong and missing my landing spot (dw im fine) Its an interesting sensation.. not being able to think.. too stuck in the split milisecond as your primal brain tries to keep you alive, overriding your cncious brain.. its not like reflex, its something else entirely.. - EagleP (A collection of pseudo-enlightenments and thoughts EagleP has had)

eagles bodark tree and cat

www.patreon.com/kee
www.makersupport.com/krineteagle

originally posted on SATURDAY, FEBRUARY 10, 2018 BY LITTLEKMA AT 5:54 PM