Ska-en-Provence 🎶🎉
[Being Human 3/n]: moving on from previous unmet goals...
I wish I had time to finish: my research on the Evolution of Probalisitic Reasoning in AI Particularly Dempster-Shafer and Bayesian Networks How LLMs and Bayesian networks can be used for Risk Management create an youtube/insta/tiktok vid for my latest post on LLM Agent But I don’t!! So this is me putting it to one side…
[Being Human 2/n] Being scrappy shows we are Human in this Brave New World
Polish is cheap in this Brave New World of AI. Being scrappy is a way of being authentic and, most importantly, Being Human!
[IA Series 7/n] Building a Self-Consistency LLM-Agent: From PEAS Analysis to Production Code
Building a Self-Consistency LLM-Agent: From PEAS Analysis to Production Code - a guide to designing an LLM-based agent.
A refreshing AI-en-Provence 🍦
AI-en-Provence 🤓😂✌🏼✌🏼✌🏼
[IA Series 6/n] A Bayesian Learning Agent: Bayes Theorem and Intelligent Agents
The article discusses how to implement Bayes Theorem in a learning agent that updates its beliefs about an environment based on new evidence, illustrated through a game involving guessing a number derived from a dice throw.
It is not reasoning...
Reasoning vs Stream of Consciousness - the output of a transformer is not reasoned in the way we think it is.
[Being Human Series 1/n] Introspection and the cusp of not knowing
What is knowledge? Wtf am I trying to learn! Claude “thinks” this post is mental masturbation 😆 well even the physical version serves a good purpose! 🤷🏼♂️
[IA Series 5/n] The Evolution from Logic to Probability to Deep Learning: A course correction to Transformers
Introduction In the previous post, I shared my view on “Why Study Logic?”, we looked at the Knowledge Representation and highlighted the importance of Logic and Reasoning in storing and accessing Knowledge. In this post I’m going to highlight a section from the book “Introduction to Artificial Intelligence” by Wolfgang Ertel. His approach with this book was to make AI more accessible than Russel and Norvig’s 1000+ page bible. It worked for me.
[IA Series 4/n] A Big Question: Why Study Logic in a World of Probabilistic AI?
Introduction The purpose of this article is to help me answer the question “Why am I studying Logic?”. If it helps you, that’d be great, let me know! The question comes from a nagging feeling of, why don’t I see logic used more in the ‘real world’. It could be a personal bias as I more easily see the utility of Rosenblatt’s work, where he looked at both Symbolic Logic and Probability Theory to help solve a problem and choose Probability Theory ([NN Series 1/n] From Neurons to Neural Networks: The Perceptron), with that we had the birth of the Artificial Neuron and the rest is history!
"There must be an invisible sun, giving heat to everyone"
“There must be an invisible sun, giving heat to everyone”
[IA Series 3/n] Intelligent Agents Term Sheet
“[IA Series 3/n] Intelligent Agents Term Sheet” breaks down essential AI terminology from Russell & Norvig’s seminal textbook. Learn what makes agents rational (or irrational), understand different agent types, and follow a structured 5-step design process from environment analysis to implementation. Perfect reference for AI practitioners and students. Coming next: how agents mirror human traits. #ArtificialIntelligence #IntelligentAgents #AIDesign
Building an Intelligent Agent
First draft in public 😱 😆🤓 What’s the best way for an agent to build a semantically sound and syntactically correct knowledge base? Dog fooding my course material means the first step is to define the task environment. /Checks notes Task Environment: The description of Performance, Environment, Actuators, and Sensors (PEAS). This provides a complete specification of the problem domain. So how can I implement this 🤔 First I need to think on the domain, something different to the examples (e.
[zero-RL] Summarising what LUFFY offers
Here’s a “standard” progression of training methodologies: PRE-Training - This is where the model gains broad knowledge, forming the foundation necessary for reasoning. CPT (Continued Pre-training) - Makes the model knowledgeable about specific domains. SFT (Supervised Fine-Tuning) - Makes the model skilled at specific tasks by leveraging knowledge it already has. RL (Reinforcement Learning) - Using methods like GRPO, DPO to align model behavior. Reasoning traces play different roles at each stage:
[zero-RL] where is the exploration?
Source: Off Policy “zero RL” in simple terms Results demonstrate that LUFFY encourages the model to imitate high-quality reasoning traces while maintaining exploration of its own sampling space. Authors introduce policy shaping via regularized importance sampling, which amplifies learning signals for low-probability yet crucial actions under “off-policy” guidance. The aspect that is still not clear to me is how there is any exploration of the solution space.
[zero-RL] LUFFY: Learning to reason Under oFF policY guidance
Based on conventional zero-RL methods such as GRPO, LUFFY introduces off-policy reasoning traces (e.g., from DeepSeek-R1) and combines them with models' on-policy roll-outs before advantage computation. … However, naively combining off-policy traces can lead to overly rapid convergence and entropy collapse, causing the model to latch onto superficial patterns rather than acquiring genuine reasoning capabilities. …genuine reasoning capabilities… I am not certain if the implication is that Deepseek-R1 can reason or that it is a reminder that no model cam genuinely reason.
[zero-RL] what is it?
Zero-RL applies reinforcement learning RL to base LM directly, eliciting reasoning potentials using models' own rollouts. A fundamental limitation worth highlighting: it is inherently “on-policy”, constraining learning exclusively to the model’s self-generated outputs through iterative trials and feedback cycles. Despite showing promising results, zero-RL is bounded by the base LLM itself. A key characteristic is that it means a LLM can be trained without Supervised Fine Tuning (SFT).
[zero-RL] When you SFT a smaller LM on the reasoning traces of a larger LM
You are doing Imitation Learning (specifically Behavioral Cloning) because the goal and mechanism involve mimicking the expert’s token sequences. You are doing Transfer Learning (specifically Knowledge Distillation) because you are transferring reasoning knowledge from a teacher model to a student model. You are not doing Off-Policy Reinforcement Learning because the learning process is supervised likelihood maximization, not reward maximization using RL algorithms. Although the data itself is “off-policy” (not generated by the model being trained), the learning paradigm is supervised imitation, not RL.
Notes and links on SVMs (WIP)
Support Vector Machines (SVM) are a mathematical approach for classifying data by finding optimal separating hyperplanes, applicable even in non-linear scenarios using kernel methods.