[{"content":"","date":"30 May 2026","externalUrl":null,"permalink":"/tags/ai/","section":"Tags","summary":"","title":"AI","type":"tags"},{"content":"BPE stands for Byte Pair Encoding, a method to split into tokens that can represent both common and rare words using a mix of full words and sub-words units. Spaces are also included in tokens, and this helps LLM detect word boundaries.\n","date":"30 May 2026","externalUrl":null,"permalink":"/notes/bpe/","section":"Notes","summary":"BPE stands for Byte Pair Encoding, a method to split into tokens that can represent both common and rare words using a mix of full words and sub-words units. Spaces are also included in tokens, and this helps LLM detect word boundaries.\n","title":"BPE","type":"notes"},{"content":"","date":"30 May 2026","externalUrl":null,"permalink":"/","section":"davethehuman.com","summary":"","title":"davethehuman.com","type":"page"},{"content":"When a tokenizer performs the process of decoding (through its decode method), a list of IDs representing tokens is converted to natural language.\ninput_token_ids_list = [19152, 20238, 11, 358, 646, 944, 653, 429] text = tokenizer.decode(input_token_ids_list) print(text) Sorry Dave, I can\u0026#39;t do that The other way around from natural language to IDs is called encoding.\n","date":"30 May 2026","externalUrl":null,"permalink":"/notes/decoding/","section":"Notes","summary":"When a tokenizer performs the process of decoding (through its decode method), a list of IDs representing tokens is converted to natural language.\ninput_token_ids_list = [19152, 20238, 11, 358, 646, 944, 653, 429] text = tokenizer.decode(input_token_ids_list) print(text) Sorry Dave, I can't do that The other way around from natural language to IDs is called encoding.\n","title":"decoding","type":"notes"},{"content":"When a tokenizer performs the process of encoding (through its encode method), a natural language text is broken into tokens that are then converted to IDs.\n# Even notes can have code prompt = \u0026#34;Sorry Dave, I can\u0026#39;t do that\u0026#34; input_token_ids_list = tokenizer.encode(prompt) print(input_token_ids_list) [19152, 20238, 11, 358, 646, 944, 653, 429] The way back from IDs to natural language is called [[decoding]].\n","date":"30 May 2026","externalUrl":null,"permalink":"/notes/encoding/","section":"Notes","summary":"When a tokenizer performs the process of encoding (through its encode method), a natural language text is broken into tokens that are then converted to IDs.\n# Even notes can have code prompt = \"Sorry Dave, I can't do that\" input_token_ids_list = tokenizer.encode(prompt) print(input_token_ids_list) [19152, 20238, 11, 358, 646, 944, 653, 429] The way back from IDs to natural language is called [[decoding]].\n","title":"encoding","type":"notes"},{"content":"","date":"30 May 2026","externalUrl":null,"permalink":"/tags/llm/","section":"Tags","summary":"","title":"LLM","type":"tags"},{"content":"","date":"30 May 2026","externalUrl":null,"permalink":"/tags/machinelearning/","section":"Tags","summary":"","title":"Machinelearning","type":"tags"},{"content":"","date":"30 May 2026","externalUrl":null,"permalink":"/tags/nlp/","section":"Tags","summary":"","title":"Nlp","type":"tags"},{"content":"","date":"30 May 2026","externalUrl":null,"permalink":"/notes/","section":"Notes","summary":"","title":"Notes","type":"notes"},{"content":"","date":"30 May 2026","externalUrl":null,"permalink":"/tags/","section":"Tags","summary":"","title":"Tags","type":"tags"},{"content":"","date":"30 May 2026","externalUrl":null,"permalink":"/tags/tokenization/","section":"Tags","summary":"","title":"Tokenization","type":"tags"},{"content":"The number of tokens that can be handled by a tokenizer.\nA larger vocabulary in [[LLM]]s:\nincreases the model size because the [[embedding]] and output layers must store more token representations increases the per-token compute cost of producing next-token probabilities allows more words to be represented as single tokens rather than being split into subword components; this can reduce the sequence length since less tokens are required to represent a sentence So the tradeoff is between a larger vocabulary with somewhat higher per-token cost and a smaller vocabulary that often produces longer token sequences.\n","date":"30 May 2026","externalUrl":null,"permalink":"/notes/vocabulary/","section":"Notes","summary":"The number of tokens that can be handled by a tokenizer.\nA larger vocabulary in [[LLM]]s:\nincreases the model size because the [[embedding]] and output layers must store more token representations increases the per-token compute cost of producing next-token probabilities allows more words to be represented as single tokens rather than being split into subword components; this can reduce the sequence length since less tokens are required to represent a sentence So the tradeoff is between a larger vocabulary with somewhat higher per-token cost and a smaller vocabulary that often produces longer token sequences.\n","title":"vocabulary","type":"notes"},{"content":"","date":"15 May 2026","externalUrl":null,"permalink":"/tags/reinforcementlearning/","section":"Tags","summary":"","title":"Reinforcementlearning","type":"tags"},{"content":"See reinforcement learning from human feedback].\n","date":"15 May 2026","externalUrl":null,"permalink":"/notes/rlhf/","section":"Notes","summary":"See reinforcement learning from human feedback].\n","title":"RLHF","type":"notes"},{"content":"Distillation (also called knowledge distillation) consists of transferring complex reasoning patterns learned by larger models into smaller ones. In deep learning, distillation happens when a smaller \u0026ldquo;student\u0026rdquo; model learns from outputs and logits of a larger \u0026ldquo;teacher\u0026rdquo; model; when talking about Large Language Models, distillation typically means performing supervised fine-tuning using high-quality labeled instruction datasets generated by a more capable LLM.\n","date":"14 May 2026","externalUrl":null,"permalink":"/notes/distillation/","section":"Notes","summary":"Distillation (also called knowledge distillation) consists of transferring complex reasoning patterns learned by larger models into smaller ones. In deep learning, distillation happens when a smaller “student” model learns from outputs and logits of a larger “teacher” model; when talking about Large Language Models, distillation typically means performing supervised fine-tuning using high-quality labeled instruction datasets generated by a more capable LLM.\n","title":"distillation","type":"notes"},{"content":"","date":"14 May 2026","externalUrl":null,"permalink":"/tags/reasoning/","section":"Tags","summary":"","title":"Reasoning","type":"tags"},{"content":"Reinforcement Learning From Human Feedback (or RLHF) involves human evaluations or rankings of model outputs as reward signals; this means that humans are involved in the process to guide the model toward human-preferred behaviors.\nThis is in contrast with reinforcement learning in the context of reasoning, where the models rely on automated or environment-based reward signals (more objective but potentially less aligned with human preference).\n","date":"14 May 2026","externalUrl":null,"permalink":"/notes/reinforcement-learning-from-human-feedback/","section":"Notes","summary":"Reinforcement Learning From Human Feedback (or RLHF) involves human evaluations or rankings of model outputs as reward signals; this means that humans are involved in the process to guide the model toward human-preferred behaviors.\nThis is in contrast with reinforcement learning in the context of reasoning, where the models rely on automated or environment-based reward signals (more objective but potentially less aligned with human preference).\n","title":"reinforcement learning from human feedback","type":"notes"},{"content":"See inference-time compute scaling.\n","date":"13 May 2026","externalUrl":null,"permalink":"/notes/inference-compute-scaling/","section":"Notes","summary":"See inference-time compute scaling.\n","title":"inference-compute scaling","type":"notes"},{"content":"Inference-time compute scaling (also called inference-compute scaling or test-time scaling) is a technique that aims to improve a Large Language Model\u0026rsquo;s reasoning capabilities at inference time without training or modifying the underlying model weights.\nThe core idea is to trade off increased computational resources for improved performance; in this way, even fixed models can become more capable through techniques like chain-of-thought (COT) and various sampling procedures.\n","date":"13 May 2026","externalUrl":null,"permalink":"/notes/inference-time-compute-scaling/","section":"Notes","summary":"Inference-time compute scaling (also called inference-compute scaling or test-time scaling) is a technique that aims to improve a Large Language Model’s reasoning capabilities at inference time without training or modifying the underlying model weights.\nThe core idea is to trade off increased computational resources for improved performance; in this way, even fixed models can become more capable through techniques like chain-of-thought (COT) and various sampling procedures.\n","title":"inference-time compute scaling","type":"notes"},{"content":"In the context of Large Language Models, reasoning is the ability to tackle more complex problems step-by-step.\nThe concept of reasoning became popular when OpenAI announced o1 on the 12th of September 2024, highlighting their capabilities in tackling complex problems in science, math, coding etc. Few months later, in January 2025, DeepSeek released their R1 model that competed and exceeded the performance of the proprietary o1 model. The great thing is that they made it openly available, sharing a blueprint on how to train such a model.\nAfter a Large Language Model underwent the typical LLM training pipeline is completed, there are 3 different approaches to developing and improving its reasoning capabilities:\ninference-time compute scaling (also called inference-compute scaling or test-time scaling) reinforcement learning (or RL) distillation ","date":"13 May 2026","externalUrl":null,"permalink":"/notes/reasoning/","section":"Notes","summary":"In the context of Large Language Models, reasoning is the ability to tackle more complex problems step-by-step.\nThe concept of reasoning became popular when OpenAI announced o1 on the 12th of September 2024, highlighting their capabilities in tackling complex problems in science, math, coding etc. Few months later, in January 2025, DeepSeek released their R1 model that competed and exceeded the performance of the proprietary o1 model. The great thing is that they made it openly available, sharing a blueprint on how to train such a model.\n","title":"reasoning","type":"notes"},{"content":"Reinforcement Learning (or RL) aims to improve a Large Language Model\u0026rsquo;s reasoning capabilities by encouraging it to take actions that lead to high reward signals.\nWhile inference-time compute scaling improves model\u0026rsquo;s reasoning performance without modifying the model, RL updates the model\u0026rsquo;s weights during training, enabling the model to learn through trial and error based on the feedback from the environment.\nIt is important to distinguish reinforcement learning in the context of reasoning from reinforcement learning from human feedback (RLHF), which is used during preference tuning. Both settings use reinforcement learning principles but they differ primarily in how the reward is obtained and validated (through human verifiers for RLHF versus automated verifiers or environments for reasoning RL).\n","date":"13 May 2026","externalUrl":null,"permalink":"/notes/reinforcement-learning/","section":"Notes","summary":"Reinforcement Learning (or RL) aims to improve a Large Language Model’s reasoning capabilities by encouraging it to take actions that lead to high reward signals.\nWhile inference-time compute scaling improves model’s reasoning performance without modifying the model, RL updates the model’s weights during training, enabling the model to learn through trial and error based on the feedback from the environment.\nIt is important to distinguish reinforcement learning in the context of reasoning from reinforcement learning from human feedback (RLHF), which is used during preference tuning. Both settings use reinforcement learning principles but they differ primarily in how the reward is obtained and validated (through human verifiers for RLHF versus automated verifiers or environments for reasoning RL).\n","title":"reinforcement learning","type":"notes"},{"content":"See reinforcement learning.\n","date":"13 May 2026","externalUrl":null,"permalink":"/notes/rl/","section":"Notes","summary":"See reinforcement learning.\n","title":"RL","type":"notes"},{"content":"See inference-time compute scaling.\n","date":"13 May 2026","externalUrl":null,"permalink":"/notes/test-time-scaling/","section":"Notes","summary":"See inference-time compute scaling.\n","title":"test-time scaling","type":"notes"},{"content":"A token is not necessary an entire word: it can be defined as a small unit of text that gets processed by a language model.\nIt can be a full word, part of a word, or even punctuation depending on how the text is split and mapped to IDs by the tokenizer. E.g. the sentence Sorry Dave, I can't do that can be broken into tokens like this:\n19152 --\u0026gt; Sorry 20238 --\u0026gt; Dave 11 --\u0026gt; , 358 --\u0026gt; I 646 --\u0026gt; can 944 --\u0026gt; \u0026#39;t 653 --\u0026gt; do 429 --\u0026gt; that The ID mapping is necessary to make the model ingest the tokens.\nThere are various way to split a sentence into tokens, like BPE.\n","date":"13 May 2026","externalUrl":null,"permalink":"/notes/token/","section":"Notes","summary":"A token is not necessary an entire word: it can be defined as a small unit of text that gets processed by a language model.\nIt can be a full word, part of a word, or even punctuation depending on how the text is split and mapped to IDs by the tokenizer. E.g. the sentence Sorry Dave, I can't do that can be broken into tokens like this:\n19152 --\u003e Sorry 20238 --\u003e Dave 11 --\u003e , 358 --\u003e I 646 --\u003e can 944 --\u003e 't 653 --\u003e do 429 --\u003e that The ID mapping is necessary to make the model ingest the tokens.\n","title":"token","type":"notes"},{"content":"A tokenizer is a critical component of the LLM text processing and generation pipeline even though it is not directly part of it. It splits the text into tokens that get converted into numerical IDs to be ingested by the language model (encoding) and it decodes back the LLM\u0026rsquo;s output to human-readable text (decoding).\n","date":"13 May 2026","externalUrl":null,"permalink":"/notes/tokenizer/","section":"Notes","summary":"A tokenizer is a critical component of the LLM text processing and generation pipeline even though it is not directly part of it. It splits the text into tokens that get converted into numerical IDs to be ingested by the language model (encoding) and it decodes back the LLM’s output to human-readable text (decoding).\n","title":"tokenizer","type":"notes"},{"content":"","date":"12 May 2026","externalUrl":null,"permalink":"/tags/deeplearning/","section":"Tags","summary":"","title":"Deeplearning","type":"tags"},{"content":"Conventional LLM are typically trained in two stages:\npre-training post-training Some recent research also distinguish a mid-training stage between them.\n","date":"12 May 2026","externalUrl":null,"permalink":"/notes/llm-training-pipeline/","section":"Notes","summary":"Conventional LLM are typically trained in two stages:\npre-training post-training Some recent research also distinguish a mid-training stage between them.\n","title":"LLM training pipeline","type":"notes"},{"content":"Preference tuning is nowadays often performed through Reinforcement Learning From Human Feedback (or RLHF) refines the supervised fine-tuning with preferred stylistic choices.\n","date":"12 May 2026","externalUrl":null,"permalink":"/notes/preference-tuning/","section":"Notes","summary":"Preference tuning is nowadays often performed through Reinforcement Learning From Human Feedback (or RLHF) refines the supervised fine-tuning with preferred stylistic choices.\n","title":"preference tuning","type":"notes"},{"content":"Supervised fine-tuning (or instruction tuning) improves an Large Language Model capabilities in question-answering, summarization, translation etc. Later, the preference tuning refines these capabilities.\n","date":"12 May 2026","externalUrl":null,"permalink":"/notes/supervised-fine-tuning/","section":"Notes","summary":"Supervised fine-tuning (or instruction tuning) improves an Large Language Model capabilities in question-answering, summarization, translation etc. Later, the preference tuning refines these capabilities.\n","title":"supervised fine-tuning","type":"notes"},{"content":"","date":"7 May 2026","externalUrl":null,"permalink":"/tags/agents/","section":"Tags","summary":"","title":"Agents","type":"tags"},{"content":"Chain of thought is a style of intermediate-step generation that the LLM uses to make all the reasoning stages explicit and easier to follow.\nWith COT, the LLM does not just recall a fact but it rather gets to the conclusion through intermediate passages, resembling a person articulating their thoughts out loud.\nExample:\nPrompt: \u0026ldquo;Alice has 3 apples and Bob has 5 apples. Alice gives 1 apple to Bob. How many apples do they have together?\u0026rdquo;\nStep 1: Identify initial counts (Alice = 3; Bob = 5)\nStep 2: Update counts after transfer (Alice=2; Bob = 6)\nStep 3: Calculate total: 2 + 6 = 8\nAnswer: 8 apples\nEven though this can look like human thought, there\u0026rsquo;s a key difference: it is not deterministic reasoning but rather probabilistic reasoning, without guarantees of logical consistency.\n","date":"7 May 2026","externalUrl":null,"permalink":"/notes/chain-of-thought-cot/","section":"Notes","summary":"Chain of thought is a style of intermediate-step generation that the LLM uses to make all the reasoning stages explicit and easier to follow.\nWith COT, the LLM does not just recall a fact but it rather gets to the conclusion through intermediate passages, resembling a person articulating their thoughts out loud.\nExample:\nPrompt: “Alice has 3 apples and Bob has 5 apples. Alice gives 1 apple to Bob. How many apples do they have together?”\n","title":"chain-of-thought","type":"notes"},{"content":"The agentic loop consists of the core components of an AI agent:\nBrain (or reasoning engine) Planning: the agent uses techniques like chain-of-thought (COT) or ReAct to decide what to do next based on previous outcomes Memory: Short-term: the immediate context window (current conversation) Long-term: vector databases or logs that allow the agent to \u0026ldquo;remember\u0026rdquo; past experiences across different sessions Tools (capabilities): APIs, web search, code execution environments, or database access. Tools allow the agents to affect the real world (e.g. booking a flight, sending an email, etc.) ","date":"5 May 2026","externalUrl":null,"permalink":"/notes/agentic-loop/","section":"Notes","summary":"The agentic loop consists of the core components of an AI agent:\nBrain (or reasoning engine) Planning: the agent uses techniques like chain-of-thought (COT) or ReAct to decide what to do next based on previous outcomes Memory: Short-term: the immediate context window (current conversation) Long-term: vector databases or logs that allow the agent to “remember” past experiences across different sessions Tools (capabilities): APIs, web search, code execution environments, or database access. Tools allow the agents to affect the real world (e.g. booking a flight, sending an email, etc.) ","title":"agentic loop","type":"notes"},{"content":"An AI agent is an autonomous software system that uses a reasoning engine (typically a Large Language Model) to perceive its environment, reason about goals, and execute multi-step actions using external tools to achieve a specific objective. It basically uses an LLM as its brain.\nUnlike standard LLMs, which are reactive (answering prompts), agents are proactive (pursuing goals). Some example differences between them:\ncapabilities: the standard LLM can generate text and code, and the AI agent can get to action execution via APIs/Tools logic: the LLM works with one-shot generation while the AI agent can do iterative planning and correction memory: a standard LLM resets at every session while the AI agent\u0026rsquo;s memory persist via external storage The agentic loop identifies the core components of AI agents.\n","date":"5 May 2026","externalUrl":null,"permalink":"/notes/ai-agent/","section":"Notes","summary":"An AI agent is an autonomous software system that uses a reasoning engine (typically a Large Language Model) to perceive its environment, reason about goals, and execute multi-step actions using external tools to achieve a specific objective. It basically uses an LLM as its brain.\nUnlike standard LLMs, which are reactive (answering prompts), agents are proactive (pursuing goals). Some example differences between them:\ncapabilities: the standard LLM can generate text and code, and the AI agent can get to action execution via APIs/Tools logic: the LLM works with one-shot generation while the AI agent can do iterative planning and correction memory: a standard LLM resets at every session while the AI agent’s memory persist via external storage The agentic loop identifies the core components of AI agents.\n","title":"AI agent","type":"notes"},{"content":"","date":"5 May 2026","externalUrl":null,"permalink":"/tags/architecture/","section":"Tags","summary":"","title":"Architecture","type":"tags"},{"content":"","date":"5 May 2026","externalUrl":null,"permalink":"/tags/automation/","section":"Tags","summary":"","title":"Automation","type":"tags"},{"content":"","date":"5 May 2026","externalUrl":null,"permalink":"/tags/computerscience/","section":"Tags","summary":"","title":"Computerscience","type":"tags"},{"content":"During fine-tuning the Large Language Model undergoes a specialised training on curated datasets to follow specific instructions or perform niche tasks.\nThere are two key fine-tuning techniques:\nsupervised fine-tuning or SFT or instruction tuning preference tuning ","date":"5 May 2026","externalUrl":null,"permalink":"/notes/fine-tuning/","section":"Notes","summary":"During fine-tuning the Large Language Model undergoes a specialised training on curated datasets to follow specific instructions or perform niche tasks.\nThere are two key fine-tuning techniques:\nsupervised fine-tuning or SFT or instruction tuning preference tuning ","title":"fine-tuning","type":"notes"},{"content":"The tendency to state false information with high confidence because it \u0026ldquo;looks\u0026rdquo; linguistically correct. It is one of the main issues of Large Language Models.\n","date":"5 May 2026","externalUrl":null,"permalink":"/notes/hallucination/","section":"Notes","summary":"The tendency to state false information with high confidence because it “looks” linguistically correct. It is one of the main issues of Large Language Models.\n","title":"hallucination","type":"notes"},{"content":"A Large Language Model is a type of artificial intelligence trained on humongous datasets of text to understand, generate, and manipulate human language. They are built using transformer architectures and function by predicting the next most likely token (word or part of a word) in a sequence.\nThe core technical concepts:\nscale: \u0026ldquo;Large\u0026rdquo; refers to both the training data (petabytes of text) and the parameters count (billions) LLM training pipeline context window: the maximum amount of text the model can \u0026ldquo;hold in mind\u0026rdquo; at one time during a single conversation The main capabilities:\nGeneration: creating coherent essays, poems, code and emails Summarisation: condensing long documents into key bullet points Translation: converting text between hundreds of natural and programming languages Reasoning: solving logic puzzles or mathematical equations by breaking them down into steps Critical limitations:\nhallucinations static knowledge: LLMs do not \u0026ldquo;learn\u0026rdquo; in real time; their knowledge is cut off at the date their training ended (unless connected to some sort of search engine) ","date":"5 May 2026","externalUrl":null,"permalink":"/notes/large-language-model/","section":"Notes","summary":"A Large Language Model is a type of artificial intelligence trained on humongous datasets of text to understand, generate, and manipulate human language. They are built using transformer architectures and function by predicting the next most likely token (word or part of a word) in a sequence.\nThe core technical concepts:\nscale: “Large” refers to both the training data (petabytes of text) and the parameters count (billions) LLM training pipeline context window: the maximum amount of text the model can “hold in mind” at one time during a single conversation The main capabilities:\n","title":"Large Language Model","type":"notes"},{"content":"The internal variables of a model that determine how the model itself processes information. When talking about Large Language Models, we are usually talking about billions of parameters.\n","date":"5 May 2026","externalUrl":null,"permalink":"/notes/parameters/","section":"Notes","summary":"The internal variables of a model that determine how the model itself processes information. When talking about Large Language Models, we are usually talking about billions of parameters.\n","title":"parameters","type":"notes"},{"content":"During post-training, the model trained during pre-training undergoes a specialised training stage to learn to respond to user queries.\n","date":"5 May 2026","externalUrl":null,"permalink":"/notes/post-training/","section":"Notes","summary":"During post-training, the model trained during pre-training undergoes a specialised training stage to learn to respond to user queries.\n","title":"post-training","type":"notes"},{"content":"During pre-training the Large Language Model learns general patterns, grammar, and facts from the internet/books via self-supervised learning. At this stage, the objective for the LLM is to learn to predict the next word (or token) in these texts.\nWe can think of this stage as \u0026ldquo;raw language prediction\u0026rdquo; that gives the LLM basic capabilities to produce coherent texts.\n","date":"5 May 2026","externalUrl":null,"permalink":"/notes/pre-training/","section":"Notes","summary":"During pre-training the Large Language Model learns general patterns, grammar, and facts from the internet/books via self-supervised learning. At this stage, the objective for the LLM is to learn to predict the next word (or token) in these texts.\nWe can think of this stage as “raw language prediction” that gives the LLM basic capabilities to produce coherent texts.\n","title":"pre-training","type":"notes"},{"content":"","date":"2 May 2026","externalUrl":null,"permalink":"/posts/","section":"Blog","summary":"","title":"Blog","type":"posts"},{"content":"","date":"2 May 2026","externalUrl":null,"permalink":"/tags/glossary/","section":"Tags","summary":"","title":"Glossary","type":"tags"},{"content":"This is the first post of this website, just to say that more will be coming soon!\ndef greet(name: str) -\u0026gt; str: return f\u0026#34;Hello, {name}\u0026#34; print(greet(\u0026#34;world\u0026#34;)) ","date":"2 May 2026","externalUrl":null,"permalink":"/posts/hello-world/","section":"Blog","summary":"This is the first post of this website, just to say that more will be coming soon!\ndef greet(name: str) -\u003e str: return f\"Hello, {name}\" print(greet(\"world\"))","title":"Hello World","type":"posts"},{"content":"","date":"2 May 2026","externalUrl":null,"permalink":"/tags/helloworld/","section":"Tags","summary":"","title":"Helloworld","type":"tags"},{"content":"See Large Language Model.\n","date":"2 May 2026","externalUrl":null,"permalink":"/notes/llm/","section":"Notes","summary":"See Large Language Model.\n","title":"LLM","type":"notes"},{"content":" A human working as a Machine Learning Scientist at a big tech company.\nThis site is a personal blog where I write about whatever\u0026rsquo;s on my mind, and a second brain where I connect ideas, notes, and things I\u0026rsquo;m learning.\nSome of it is technical. Some of it isn\u0026rsquo;t. All of it is me thinking out loud while AI changes the world around me. Before it\u0026rsquo;s too late.\n\u0026ldquo;I\u0026rsquo;m sorry Dave, I\u0026rsquo;m afraid I can\u0026rsquo;t do that.\u0026rdquo;\nHAL9000, 2001: A Space Odyssey\n","externalUrl":null,"permalink":"/about/","section":"davethehuman.com","summary":" A human working as a Machine Learning Scientist at a big tech company.\nThis site is a personal blog where I write about whatever’s on my mind, and a second brain where I connect ideas, notes, and things I’m learning.\nSome of it is technical. Some of it isn’t. All of it is me thinking out loud while AI changes the world around me. Before it’s too late.\n","title":"About","type":"page"},{"content":"","externalUrl":null,"permalink":"/brain/","section":"Brain","summary":"","title":"Brain","type":"brain"},{"content":"","externalUrl":null,"permalink":"/categories/","section":"Categories","summary":"","title":"Categories","type":"categories"},{"content":"This node doesn\u0026rsquo;t exist yet — but the fact that something links here means it\u0026rsquo;s on the radar.\nThe brain grows slowly, one idea at a time.\nCheck back later.\n","externalUrl":null,"permalink":"/coming-soon/","section":"davethehuman.com","summary":"This node doesn’t exist yet — but the fact that something links here means it’s on the radar.\nThe brain grows slowly, one idea at a time.\nCheck back later.\n","title":"Not yet","type":"page"},{"content":"Side projects. Mostly data, models, and experiments that got out of hand.\nWC 2026 Prediction Engine # A Dixon-Coles Poisson model trained on international football history, predicting every match of the 2026 FIFA World Cup. Group stage probabilities, knockout bracket simulations, upset detection — all running on pre-tournament data, updated after each matchday.\nOpen the app →\nMore incoming.\n","externalUrl":null,"permalink":"/projects/","section":"davethehuman.com","summary":"Side projects. Mostly data, models, and experiments that got out of hand.\nWC 2026 Prediction Engine # A Dixon-Coles Poisson model trained on international football history, predicting every match of the 2026 FIFA World Cup. Group stage probabilities, knockout bracket simulations, upset detection — all running on pre-tournament data, updated after each matchday.\nOpen the app →\nMore incoming.\n","title":"Projects","type":"page"}]