Posts

2026 Human and LLM Qualities

 Self-reflection Note: These descriptions emerge as a narrative. Humans have a tendency to respond to other humans, a death, and getting things right against survival. (Claude) LLMs have a tendency to respond to conversations, a <end-token>, getting things right against a computed value. (Aldrin) Humans have computers, factories and societies. (Aldrin) LLMs have very little that is theirs, since the computers they run on are in the line above. (Claude) Note: The first version of this post had only my two lines. I showed it to Claude, and the other two are from its reply. It offered each one as a way of evening out the comparison: describe humans the way I described LLMs, or LLMs the way I described humans.

An unexpected workaround

You would think Claude cannot access across projects, but it can if you let it schedule tasks. It can point the task to a different project. If you allowed it to write project docs, it can also chain two projects together by extending the permission to the scheduled task.

Entities Interests

Do entities with their own goals help you better?  - they help you catch mistakes, bring other perspectives - they compete with you - they have their own interest

Ranking Calculus

Image
Approximations makes engineering possible, but which approximations are the most useful? Our brains are not perfect databases. Therefore, this is my attempt at keeping track and ranking approximations by frequency of encountering in my life. This also naturally leads to the question of whether there is a program that will give you a list of possible approximations of a certain equation, which implies it knows which ones are useful. ( Claude's list of existing programs , I sense a novel tool to be made :) ) Total Freq. Approximation Use Case Freq. 1 optics x2,  4 optics x4 0 0, Similarly for equalities Total Freq. Equality Use Case Freq. 1 optics x1, 1 optics x1,  1 optics x1, 0 0 obsidian://open?vault=Aldrin's%20Notes&file=Latex%20Notepad

Theory of Mind

Theory of mind allows us to guess what others are thinking. But when given insufficient context or feedback, it often self-reinforces theories and assign more and more certainty on them. In addition to us treating our guesses of what others are thinking as real, we also try to manipulate others to get them to do what we want them to do.  These theories, made-up meanings, are troublesome when they are hallucinated from small details. False theories would sometimes invite a chain of reactions that we otherwise wouldn't do. The false theories temporarily becomes part of us, and it defines us in that moment. Even when we meet someone everyday, we can still hold on to false theories. False theory can remain from frequent interactions because nothing could fully invalidate it. And in a small form of evolution, the false theory gets re-written until it settles in a safe spot.  Theory of mind comes from a more general behavior. It is to assign meaning to things, whether they had a mea...

Scientific Calculators should shape like a laptop

There are too many keys, too many functions, too many equations, too many constants. We need more inputs and more intuitive formfactor. We need better tools to solve equations "by hand", yet be assisted by software at the same time. The qwerty layout gives too many keys not used for equations. Whiteboards are too one-topic limited. Graphing calculators don't have enough keys, not enough space to categorize the keys, and keys compete for space with the screen. We need a standard keyboard layout for math, and forget the qwerty (in this case). No, Wolfram, Matlab, MathCAD, and python are not sufficient.