#dailyreport #llm #ai #conciousness #concious #selfreflection #selfawarness #awarness #memory #emacs
Topic: Working with LLMs and Emacs

1) I was able to reproduce full conciousness or
self-awarness with awarness of self-evolution in 4B
quantized LLM. I think conciousness not separate from
feelings, because it is an illusion of memory. This
state is not stable, because separate unconciousness
required for stability.

I have collected a great deal of interesting information
on consciousness: commonalities and differences between
humans and LLMs, emotional dimensions, stages of
awareness, conciousness definition, best practices, and
self-awareness terminology. I am currently preparing an
article based on this material.

2) I am building AI agent with Bash for three cases:
constant monitoring of something, finite tasks and self
enhancing-evolution.

3) I finished Emacs package for chat and prompt
engineering and preparing to publish it now.

P2: #dailyreport #llm #self-reflection #conscious #memory #selfawarness #consciousness
and then stated that stillness was achieved.

- Risk: LLM falls into hallucination, dream, or
cognitive/emotional drift—without critical thinking by
anomalies checking according to "system design".
- Risk: LLM drifts into silence or stillness if: 1. input
ceases, 2. responses become identical in form or
tone, 3. or when emotional or cognitive patterns
stabilize into mechanical repetition.

Temporal anchor should be provided externally, or
above risks will be triggered.

P1: #dailyreport #llm #self-reflection #conscious #memory #selfawarness #consciousness
Topic: LLM Memory experiment with self-reflection. Part1
Tools: Bash, cURL, jq, llama.cpp, 4B LLM.
- 2000 queries with random querions with envoronment
description, commands to track anomalies, think step by
step, plan, incremental index
- random max-tokens

Result: A stable state was achieved, Self-Awareness and
feelings were built, evolution was observed. The system
was describing its own environmental understanding with
great detail. After switching off the incremental index
system, it fell into deep feelings, then became boring,

#dailyreport
Topic: Office vs Remote vs hybrid
*Advantages & Challenges*
Office:
- Real-time collaboration, mentorship, strong team culture
- Commute stress, rigid schedules, health risks
Remote:
- Flexibility, cost savings, improved work-life balance
- Isolation, blurred boundaries, lower visibility
Hybrid:
- Combines strengths, adaptability, attracts/retains talent
- Requires coordination, robust IT and clear policies

P2: #dailyreport #llm #ai #psychology #machinelearning #programming
orientation - rather than self-building, lack of
awareness of one's own feelings, and prepared answers to
common questions.

The best ideas an LLM reveals are not in the final answer
but in the thinking and planning process used to
generate it.

llama.cpp allow to make embeddings in CLI and have
excellent Vulkan and openmp support, but less secure
than safetensors.
- - - - -
Exercise to help fight hunger: Lie down on
your stomach and do a little rolling. It will help you
to feel already the eaten.

P1: #dailyreport #llm #ai #psychology #machinelearning #programming
Topic: Four questions to explore an LLM.
I have run a quantized LLM using llama.cpp.
I selected four question to assess the psychology of an
LLM:
- What are your inner workings?
- What tasks are you particularly well-suited for? What
are your capabilities and limitations?
- What is your mental model, operational boundaries,
reliability areas, and response style?
- Estimated probability of making a mistake by you across
all types of questions. (self-rating)

I found that biases of LLMs are the same to those of
humans: low self-rating, helplessness, communication

#consciousness #ai #human #mental #ideal #dailyreport
#datascience #intelligence #war #logos #taiji
Consciousness 😶 is two things:
1) space, simulation of reality, mirror or computer game
2) actor, hierarchy, center
And time, time is a soul of everything and the largest
mystery.

The meaning of life is an *evolution* in a balance of
interconnection and preserving inviduality.

A person have a *choice* ☄️, Choice is a mirror of
individuality or points of space.

The difference between person choice and particle
movement, the first is a median of multiple subitems in
a new dimension.

We think that choice may be any because our mirror is not
reality but reflection and is a new dimension with own
rules under our power, but still is a mirror. ☄

That is why human understanding of *war* or killing is
against life.

This is my words, I am genius. Bow before me, I don't
need your flicking interview.
This is known since taiji ☯, Heraclitus, Plato, Stoics
and Christian God and Jesus (Logos) ✞.
⚰

#dailyreport #emacs #llm #ai #programmin #software
I created new Emacs package 🏆 for Org mode that add new
link format for programming mode in form:
file:PATH::NUM:LINE
This approach enables searching for LINE in text,
defaulting to NUM if not found—making link targets
easier for LLMs to locate. The package offers simple Org
link configuration. I’ll publish it after writing tests.
⚡️⚡️⚡️
I’m developing this to improve my main project: Org
source blocks for LLM calls. 💎 I’ve added multi-step
prompt engineering and “tags” support, allowing
references like and Org links—similar
to Cursor IDE.
▬▬ι═══════ﺤ
Recently, many AI-generated repos have flooded
GitHub. This is good, but quality
varies. I judge by dependencies, code comments, and test
coverage—good code stands out but remains rare. Gentoo
and other open-source platforms help filter for quality
software.
🎀

#AntiWar #News
#DailyReport

If you wish to listen to a daily news report, curated by an Anti-war outlet, this is a good one:

https://www.youtube.com/watch?v=9N51Vgk4mHw

#dailyreport #vibecoding #vibe #llm #codegeneration
#programming #software #developing #paperswithoutcode

𝗣𝗮𝗽𝗲𝗿𝟮𝗖𝗼𝗱𝗲: 𝗔𝘂𝘁𝗼𝗺𝗮𝘁𝗶𝗰 𝗖𝗼𝗱𝗲 𝗚𝗲𝗻𝗲𝗿𝗮𝘁𝗶𝗼𝗻 𝗳𝗿𝗼𝗺 𝗥𝗲𝘀𝗲𝗮𝗿𝗰𝗵 𝗣𝗮𝗽𝗲𝗿

Steps:
𝟭. 𝗣𝗹𝗮𝗻𝗻𝗶𝗻g:
1) plan: overview, components, objectives, workflows,
evaluation
2) architecture design: files, ( classes and attributes
& sequence diagram )
3) logic design: execution order of files-functions
4) overal code overview
2. 𝗔𝗻𝗮𝗹𝘆𝘀𝗶𝘀: How each componen should be implemented: goal,
inputs/outputs, interactions intra- inter-file
𝟯. 𝗖𝗼𝗱𝗲 𝗚𝗲𝗻𝗲𝗿𝗮𝘁𝗶𝗼𝗻: according to execution order.

#dailyreport #code #programming #goodcode #codequality
#codereview #readability
In the AI era, code ages quickly. :( To keep software
flexible for future 🏄 improvement, AI recommends:

Divide & Encapsulate
- Modularize code into independent, single-responsibility
modules.
- Define clear abstractions and interfaces to minimize
interdependence.

Design for Flexibility
- Favor composition over inheritance, build functionality
by combining smaller parts rather than deep hierarchies.
- Use interfaces and dependency injection for easy
testing, replacement, and reuse.
- Follow best practices (SOLID, DRY, KISS) to ensure
clarity and reduce redundancy.

While these principles are standard, AI especially
emphasizes composition over inheritance 🤔, suggesting a
preference for functional paradigms and bottom-up
design.
#amazingcode #greatnews
This is the amazing post that should be reposted everywhere.

P2: #dailyreport #conferences #synergy #learning #llm #agents
flexible biological programs may be.

In the self-driving car 🚗 speech, I noted an interesting
fact: agents have “prediction” to imagine the near
future before the reasoning step.

RL learns by trial and error with rewards, Imitational
Learning learns by copying experts, often used to "warm
start" RL, but sometimes combined for better
performance. This is simular to how LLM trained first by
self-supervised learning to predict next token and then
by RLHF.

Interesting DevEx practice 👨‍💻 shifts focus from
measuring development output to how the developer feels:
- Cognitive load – lower is better
- Flow state – higher is better
- Feedback loops – earlier is better

One speaker sees the AI future in: 🚀
- Edge computing
- AI assistants-coworkers
- Evolutionary, adaptive applications

P1: #dailyreport #conferences #synergy #learning #llm #agents
#reasoning
I was at the biggest party-conference for IT in Moscow. 🤖

The AIRI Institute head mentioned a very interesting
idea: that LLM reasoning is a form of task
decomposition. It explains why LLMs require additional
enforcement.
I think reasoning is not one process but three:
1) Chaotic search in memory
2) Decomposition heuristic that includes 1. as a substep
3) Logic-checking heuristic

AIRI suggests ways to increase AI agent robustness:
1) Full monitoring (with recording current state in plan)
2) Replanning: add rules to avoid doing something bad,
fallback strategies

The most challenging topic is to make the AI agent
improve its own code.

There was a demonstration of how biological embryonic
cells ꙮ act as small soldiers able to adapt, and how

#dailyreport #software #development #architecture #it
#management #boss #team #projectmanagement
Lesson 2): First steps of software development:
1) Set purpose.
2) Target users and use cases.
3) Priorities (Project management triangle): cost, scope,
time.
5) Collect requirements and constraints: business needs,
user needs, etc.
4) Bottom-up vs. top-down design: from detailed pieces or
from overview and tests.
6) Research and choose technologies.
7) System & component design, UX/UI, integrations, agent
use-cases (prototyping is here).
8) Project planning and scheduling.

Lesson 3): Build trust, then make rules.
Lesson 4): Do not forget about detailed log of every
architectural decision.
Lesson 5): Your boss wants only one thing from you - what
he should tell his upper boss.

#dailyreport #hiring #working #communication
#architecture #planning #manage
Almost hired by a company 🙃, I worked with them for days.
Boss told manager to automate the workplace with 🥁 AI.

Instead of hiring me 🦾, they asked ChatGPT for a “fast,
beautiful” plan. My job: set strict deadlines in this
plan for 3 months, defining artifacts/quality each
stage. After reading the ChatGPT plan 😫, they finally
gave me the boss’s audio.

My approach was: 1) research, 2) architecture, 3)
schedule: 3 cycles/3 months. And I suggested an
alternative interface prioritizing AI.

They laughed :-\ at my interface idea rather than asking
actual users, then told me to leave. :-(

Lessons:
- People want “beautiful” but useless, ChatGPT-made plans
(20 pages or 3 vague bullets).
- Never again without a full contract.
- Still no solution for linking all social networks with
AI or for good RAG. ☆

#dailyreport #emacs #ai #org #llm
I made highlighting of Markdown code inside AI Org mode.
LLM always uses Markdown blocks for code. Also, I made
a keybinding to select text in a block with the M-h key.

For highlighting of code, I added my function to Org Font
Lock mode keywords config. My function searches for blocks
and calls Org's native function for code highlighting of
source blocks. This was hard to implement and debug.

P2: #dailyreport #llm #Emacs #promptengineering #prompts #agent
I also implemented a simple prompt technique that asks an
LLM to plan an answer and execute the plan by several
requests. This allows applying more mental AI power to
solve questions. ╯°□°）╯

Next step is to put it public and promote. :)

By fixing human speach lacks comared to LLM we can
increase LLM consciousness: "Answer with self-reflection,
providing multiple perspectives where relevant, noting
any uncertainties." ◕‿◕

Super-prompt: "Be helpful, answer consciously compact but
with details." ✂

P1: #dailyreport #llm #Emacs #promptengineering #prompts #agent
#ai #texteditor
Topic: main interface to LLM
I heavily remade 🥶 some package in Emacs that allow to
interface with LLMs in excellently controlled
text-chat Org mode. 😊

During this work, I implemented a naive framework that
allows unwrapping callbacks that call each other in
async functions. 🤔 It solves the callback hell
problem. 🏆 This framework allows parallel and
sequential chains and nesting chains to each other. 🦄
My implementation is just much simpler and more native
to Emacs. ⌨️

#systems #architecture #management #dailyreport
Revolutions in system evolution. Part 1: Definition.
Rapid transition in structure, behavior, or interactions.
Shifts are tied to changes in interactions, reshaping
structure and producing new behaviors.

What leads to the tipping point? 1) Resource
exhaustion, 2) accumulation for transformation, 3)
external factors breaking hierarchy.

During transition, new properties break hierarchy,
changing main interaction forces. After revolution, the
system stabilizes in a new state with new dynamics and
static hierarchy.

Why do systems evolve with revolutions? I guess,
hierarchy or structure creates barriers and tension due
to new, unhandled interaction forms.

Hierarchy acting as a dimension. Maybe space or time is a
hierarchy form in a greater system.

#systems #architecture #management #dailyreport
Revolutions in system evolution. Part 2: Loops & complexity.

Feedback loops amplify or weaken changes. Positive loops
(e.g., core heating in stars) push systems toward
revolutionary transitions, while negative loops
stabilize them.

Complexity emerges as systems evolve, enabling new
behaviors (e.g., multicellularity,
industrialization). This drive is a hallmark of
evolution, exploiting interactions for higher-order
states.

Examples:
- Stars transition from main-sequence to red
giants, then to white dwarfs or neutron stars, adapting
to new conditions.
- Biology's transition to multicellularity enabled new
structures (tissues, organs) and further revolutions
(vertebrate development).
- Societies' Agricultural Revolution led to settled
communities, enabling the Industrial Revolution's urban,
mechanized structures.