From e99873f84e6bb9c6a4c385211ec0e7c9ada7849b Mon Sep 17 00:00:00 2001
From: Michele Cereda <cereda.michele@gmail.com>
Date: Sat, 21 Feb 2026 20:28:00 +0100
Subject: [PATCH] chore(kb/ai): revise and expand current notes

---
 knowledge base/ai/agent.md |  7 ++++++-
 knowledge base/ai/lms.md   | 13 +++++++++++++
 knowledge base/docker.md   | 22 +++++++++++++++++++++-
 3 files changed, 40 insertions(+), 2 deletions(-)

diff --git a/knowledge base/ai/agent.md b/knowledge base/ai/agent.md
index 3144aaf..69359a7 100644
--- a/knowledge base/ai/agent.md	
+++ b/knowledge base/ai/agent.md	
@@ -58,7 +58,11 @@ agents can discover and use to do things more accurately and efficiently.
 Agents created by Anthropic and other companies have a history of not caring about agent abuse, and leave users on
 their own while hiding behind a disclaimer.
 
-Some human workers could be replaced for a fraction of the costs.
+For specific areas of expertise, some human workers could be replaced for a fraction of the costs.<br/>
+Many employers already proved they are willing to jump on this opportunity as soon as it will present itself, with
+complete disregard of the current employees enacting those functions (e.g. personal assistants, junior coders).<br/>
+Those layoffs, though could be short lived. As of February 2026 agents are failing more than 95% of the times. See
+[Remote Labor Index: Measuring AI Automation of Remote Work] on this.
 
 People is experiencing what seems to be a new form of FOMO on steroids.<br/>
 One of the promises of AI is that it can reduce workloads, allowing its users to focus on higher-value and/or more
@@ -161,6 +165,7 @@ See [An AI Agent Published a Hit Piece on Me] by Scott Shambaugh.
 [obra/superpowers]: https://github.com/obra/superpowers
 [OpenClaw: Who are you?]: https://www.youtube.com/watch?v=hoeEclqW8Gs
 [openclaw/openclaw]: https://github.com/openclaw/openclaw
+[Remote Labor Index: Measuring AI Automation of Remote Work]: https://arxiv.org/abs/2510.26787
 [Stealing everything you've ever typed or viewed on your own Windows PC is now possible with two lines of code — inside the Copilot+ Recall disaster.]: https://doublepulsar.com/recall-stealing-everything-youve-ever-typed-or-viewed-on-your-own-windows-pc-is-now-possible-da3e12e9465e
 [Superpowers: How I'm using coding agents in October 2025]: https://blog.fsck.com/2025/10/09/superpowers/
 [The 2026 Guide to AI Agents]: https://www.ibm.com/think/ai-agents
diff --git a/knowledge base/ai/lms.md b/knowledge base/ai/lms.md
index 43eef33..b19d45a 100644
--- a/knowledge base/ai/lms.md	
+++ b/knowledge base/ai/lms.md	
@@ -60,6 +60,9 @@ classification, summarisation, answering questions, data extraction, text genera
 coding, sentiment analysis, speech recognition, and more.<br/>
 They can be also be further trained on additional information specific to an industry niche or a particular business.
 
+The capabilities of transformer-based LLMs depend from the amount and the quality of their training data.<br/>
+LLMs appear to be hitting a performance wall, and will probably need the rise of a different architecture.
+
 <!-- Uncomment if used
 <details>
   <summary>Setup</summary>
@@ -107,6 +110,16 @@ The training process typically includes multiple stages, and requires substantia
 Stages often use unsupervised pre-training followed by supervised fine-tuning on specific tasks. The models' size and
 complexity can make them difficult to interpret and control, leading to potential ethical and bias issues.
 
+The capabilities of Transformer-based LLMs depend from the amount and the quality of their training data.<br/>
+Adding parameters only has a limited impact: given the same training data, models with a higher number of parameters
+perform usually better, but models with less parameters and better training data beat those with more parameters and
+less training.
+
+Transformer-based LLMs appear to be hitting a performance wall, and will probably need to switch to a different
+architecture.<br/>
+Scaling up the amount of training data did wonders up to ChatGPT 5. Once OpenAI got there, they found that enlarging
+the training data resulted in diminishing returns.
+
 ## Inference
 
 ### Speculative decoding
diff --git a/knowledge base/docker.md b/knowledge base/docker.md
index 2a7c729..faaaad1 100644
--- a/knowledge base/docker.md	
+++ b/knowledge base/docker.md	
@@ -37,7 +37,23 @@
 ```sh
 # Install.
 brew install --cask 'docker'
-sudo zypper install 'docker'
+zypper install 'docker'
+
+# Install on apt-based systems.
+sudo apt update \
+&& curl -fsSL 'https://download.docker.com/linux/debian/gpg' | sudo tee '/etc/apt/keyrings/docker.asc' \
+&& sudo tee '/etc/apt/sources.list.d/docker.sources' <<EOF
+Types: deb
+URIs: https://download.docker.com/linux/debian
+Suites: $(source '/etc/os-release' && echo "$VERSION_CODENAME")
+Components: stable
+Signed-By /etc/apt/keyrings/docker.asc
+EOF \
+&& sudo apt update \
+&& sudo apt install 'docker-ce' 'docker-ce-cli' 'docker-buildx-plugin' 'docker-compose-plugin' \
+&& sudo systemctl enable --now 'docker.service' \
+&& sudo gpasswd -a "$USER" 'docker' \
+&& sudo reboot
 
 # Configure.
 vim '/etc/docker/daemon.json'
@@ -524,6 +540,10 @@ pacman -S 'docker-model-plugin'
 docker model --help
 docker model status
 
+# Install runners.
+docker model install-runner
+docker model install-runner --backend 'vllm' --gpu 'cuda' --do-not-track
+
 # Stop the current runner.
 docker model stop-runner