Course reference

Sources & summaries

Every link you've added, what it became, the primary sources followed through to, the files saved locally, and every external reference each module cites.

Links added

Saved source files

Unique external references

646

Modules

Module 01: LLM evals with decision models: thresholds, trade-offs and calibration
Module 02: TensorTonic: hands-on ML practice for LLMs and robotics
Module 03: How much to help: human–AI and human–robot interaction
Module 04: Big teacher, small student: distilling fast on-device models for apps and robots
Module 05: LLM architecture at trillion scale: width, depth and length in Kimi K3
Module 06: Ilya's 30 papers: the shared foundations of LLMs and robotics
Module 07: LLM post-training: demonstrations, comparisons and checkers (and the robot equivalents)
Module 08: The LLM research-engineer roadmap: 12 stages, from autograd to open-source teardowns
Module 09: LLM inference engineering: 25 interview problems, worked through with real numbers
Module 10: Self-improving agent harnesses: Recursive Language Models, Continual Harness and Prime Agent
Module 11: The LLM self-study stack: CS336, Build a Reasoning Model, the RLHF Book (and a gym plan)
Module 12: AI training and inference, in nine videos: watch guides for Jia-Bin Huang's playlist
Module 13: 100 days of inference engineering: a learning journal, the book behind it, and a plan you can follow
Module 14: Deep-ML's Deep-Path: 103 concepts from classical ML to GPU kernels, mapped onto this course (with a placement check)
Module 15: Inference Engineering, interactive edition: how to use inferenceengineering.tech alongside this course
Module 16: Hands-On LLM Serving and Optimization: the book's code as a lab track
Module 17: 391 free AI courses, checked: a searchable catalog with link health, provenance, and 2026 updates
Module 18: Build a Robot Brain: 12 weeks, from zero to a robot that understands you (ages 15+)
Module 19: Evals: how to know whether an AI system actually works (a field guide, from one test case to a release gate)
Module 20: Six tiny repos that rebuild modern AI from scratch: micrograd, minGPT, nanoGPT, llm.c, tinygrad, minimalRL (run and tested)
Module 21: The maths behind robots, learned by playing: ten interactive explainers, each one used to make Robo-Helper better
Module 22: Gradient, divergence and curl: three ways a field of arrows moves, and how a robot steers with them

LLMModule 01

LLM evals with decision models: thresholds, trade-offs and calibration

Link you sentx.com/HamelHusain/status/2112588257104216294

AuthorHamel Husain & Antaripa Saha

Date20 Oct 2026

TypeX article

Summary: Set minimum precision and recall first, tune the threshold with CV, verify calibration, then pick on the cost/quality frontier.

Followed through to:

Full post with interactive charts (hamel.dev)

Saved in sources/: 01-hamel-decision-model-evals.json

All 7 external references cited in this module

Reference	Site
full post with interactive charts	hamel.dev
Hamel: Creating an LLM judge that drives business results	hamel.dev
Why binary pass/fail instead of 1–5 scales	hamel.dev
scikit-learn: precision, recall, F-measure	scikit-learn.org
Machine Learning with Python Cookbook , 2nd ed.	oreilly.com
Randy Olson: When to use a decision model instead of an LLM	randalolson.com
X post	x.com

LLM + RoboticsModule 02

TensorTonic: hands-on ML practice for LLMs and robotics

Link you sentwww.tensortonic.com/

AuthorTensorTonic

Datefetched 21 Oct 2026

TypeWebsite

Summary: Drill the maths and learning half of robotics from scratch: transforms, CV, the RL ladder to PPO, ViT/DDPM. Kinematics, control, estimation and planning need other sources.

Followed through to:

Saved in sources/: 02-tensortonic.md

All 53 external references cited in this module

Reference	Site
tensortonic.com	tensortonic.com
dot product & norms	tensortonic.com
matrix multiplication	tensortonic.com
orthogonality	tensortonic.com
least squares	tensortonic.com
SVD	tensortonic.com
eigenvalues	tensortonic.com
Bayes	tensortonic.com
covariance matrices	tensortonic.com
Markov chains	tensortonic.com
Monte Carlo	tensortonic.com
Jacobians & Hessians	tensortonic.com
matrix transpose	tensortonic.com
Apply 4×4 homogeneous transform	tensortonic.com
Gaussian blur	tensortonic.com
Sobel edges	tensortonic.com
bilinear resize	tensortonic.com
image rotation	tensortonic.com
IoU	tensortonic.com
non-max suppression	tensortonic.com
anchor generation	tensortonic.com
Hungarian matching	tensortonic.com
mIoU	tensortonic.com
Dice loss	tensortonic.com
Cracking CV	tensortonic.com
Transformers, ViT, U-Net, DDPM	tensortonic.com
patchify	tensortonic.com
ViT attention block	tensortonic.com
CLIP retrieval	tensortonic.com
quantization	tensortonic.com
KV cache	tensortonic.com
discounted returns	tensortonic.com
Bellman expectation	tensortonic.com
Bellman optimality	tensortonic.com
value iteration	tensortonic.com
policy iteration	tensortonic.com
Monte Carlo evaluation	tensortonic.com
TD(0)	tensortonic.com
SARSA	tensortonic.com
Q-learning	tensortonic.com
ε-greedy	tensortonic.com
UCB1	tensortonic.com
Thompson sampling	tensortonic.com
replay buffer	tensortonic.com
DQN with target net	tensortonic.com
Double DQN	tensortonic.com
REINFORCE	tensortonic.com
with baseline	tensortonic.com
A2C	tensortonic.com
GAE	tensortonic.com
PPO clipped objective	tensortonic.com
ML Math	tensortonic.com
All problems	tensortonic.com

LLM + RoboticsModule 03

How much to help: human–AI and human–robot interaction

Link you sentx.com/LinusEkenstam/status/2112526605189054844

AuthorLinus Ekenstam

Date20 Oct 2026

TypeX post + 2:45 video (not watched)

Summary: Blend robot and human control by calibrated confidence, keep intent beliefs revisable, choose ask/act/wait by expected cost, and keep the struggle where the learning is.

Followed through to:

gakku.ai

Saved in sources/: 03-ekenstam-gakku.json

All 2 external references cited in this module

Reference	Site
gakku	gakku.ai
X post	x.com

LLM + RoboticsModule 04

Big teacher, small student: distilling fast on-device models for apps and robots

Link you sentx.com/MaximeRivest/status/2112156468212289887

AuthorMaxime Rivest

Date19 Oct 2026

TypeX post + chart

Summary: Log a big model on your narrow task, fine-tune the smallest model that meets your bar, evaluate against hand-corrected data split by day, and measure latency on the target device. Milliseconds are metres.

Followed through to:

own-voice-stt repo (README, evaluation, pitfalls, next steps)

Saved in sources/: 04-own-voice-stt-repo.md, 04-rivest-chart.jpg, 04-rivest-distill-tweet.json

All 6 external references cited in this module

Reference	Site
own-voice-stt repo	github.com
evaluation doc	github.com
pitfalls	github.com
tiny-classifiers	github.com
maivi	github.com
X post	x.com

LLMModule 05

LLM architecture at trillion scale: width, depth and length in Kimi K3

Link you sentx.com/antiAIvo/status/2112510656037761095

Author@antiAIvo, quoting Philip Kiely (Baseten)

Date20 Oct 2026

TypeX post (Chinese) + 1h video (not watched)

Summary: Read architectures along three axes: width (sparse experts, balanced routing), depth (layers that reach back), length (fixed-size memory plus some global attention). Every feature also needs a numerical-stability fix and a serving plan.

Followed through to:

Saved in sources/: 05-antiAIvo-kiely-quote.json, 05-kimi-k3-readme.md, 05-kimi-k3-tech-report.pdf

All 8 external references cited in this module

Reference	Site
Kimi K3 repo and technical report	github.com
Baseten model page	baseten.co
mixture-of-experts	tensortonic.com
KV cache	tensortonic.com
speculative decoding	tensortonic.com
quantisation	tensortonic.com
@antiAIvo's post	x.com
Philip Kiely (Baseten)	x.com

LLM + RoboticsModule 06

Ilya's 30 papers: the shared foundations of LLMs and robotics

Link you sentgithub.com/Ojhaharsh/30u30

AuthorOjhaharsh (30u30)

Datefetched 21 Oct 2026

TypeGitHub repo (CC BY-NC-ND 4.0)

Summary: The toolbox behind robot learning in 30 days: memory (RNN → LSTM → attention), seeing (AlexNet → ResNet), relations and graphs, compression as learning, scale, and learning to act (PPO, RLHF). Start each day with the plain-language box.

Followed through to:

The 30 original papers: linked from each day page

Saved in sources/: 06-30u30-readme.md

All 71 external references cited in this module

Reference	Site
arXiv:1409.2329	arxiv.org
A theoretically grounded application of dropout in RNNs	arxiv.org
Weight Uncertainty in Neural Networks	arxiv.org
arXiv:math/0406077	arxiv.org
arXiv:1405.6903	arxiv.org
arXiv:1512.03385	arxiv.org
arXiv:1603.05027	arxiv.org
arXiv:1511.07122	arxiv.org
arXiv:1706.03762	arxiv.org
arXiv:1409.0473	arxiv.org
arXiv:1511.06391	arxiv.org
arXiv:1410.5401	arxiv.org
arXiv:1506.03134	arxiv.org
arXiv:1706.01427	arxiv.org
arXiv:1806.01822	arxiv.org
arXiv:1704.01212	arxiv.org
arXiv:1512.02595	arxiv.org
arXiv:1611.02731	arxiv.org
arXiv:1811.06965	arxiv.org
arXiv:2001.08361	arxiv.org
Training Compute-Optimal Large Language Models	arxiv.org
Universal Intelligence: A Definition of Machine Intelligence	arxiv.org
arXiv:1707.06347	arxiv.org
GAE paper	arxiv.org
Learning to Walk in Minutes Using Massively Parallel Deep RL	arxiv.org
arXiv:1706.03741	arxiv.org
Training language models to follow instructions with human feedback	arxiv.org
colah.github.io	colah.github.io
cs231n.github.io, ConvNets notes	cs231n.github.io
full notes index	cs231n.github.io
30u30 Day 1	github.com
30u30 Day 2	github.com
30u30 Day 3	github.com
30u30 Day 4	github.com
30u30 Day 5	github.com
30u30 Day 6	github.com
30u30 Day 7	github.com
30u30 Day 8	github.com
30u30 Day 9	github.com
30u30 Day 10	github.com
1001-layer code	github.com
30u30 Day 11	github.com
30u30 Day 12	github.com
30u30 Day 13	github.com
30u30 Day 14	github.com
30u30 Day 15	github.com
30u30 Day 16	github.com
30u30 Day 17	github.com
30u30 Day 18	github.com
30u30 Day 19	github.com
30u30 Day 20	github.com
30u30 Day 21	github.com
30u30 Day 22	github.com
30u30 Day 23	github.com
30u30 Day 24	github.com
30u30 Day 25	github.com
30u30 Day 26	github.com
30u30 Day 27	github.com
30u30 Day 28	github.com
30u30 Day 29	github.com
30u30 Day 30	github.com
Ojhaharsh/30u30	github.com
jmlr.org	jmlr.org
karpathy.github.io	karpathy.github.io
nlp.seas.harvard.edu/annotated-transformer	nlp.seas.harvard.edu
NeurIPS proceedings	papers.nips.cc
scottaaronson.blog	scottaaronson.blog
AMS book page	ams.org
cs.toronto.edu (PDF)	cs.toronto.edu
Alexander Shen's homepage	lirmm.fr
thesis PDF (vetta.org)	vetta.org

LLM + RoboticsModule 07

LLM post-training: demonstrations, comparisons and checkers (and the robot equivalents)

Link you sentx.com/dongxi_nlp/status/2112767071926792566

AuthorDongxi (东锡)

Date21 Oct 2026

TypeX article (Chinese)

Summary: Pre-training sets what a model knows; post-training sets how it uses it. Three teachers: demonstrations (SFT / behaviour cloning), comparisons (RLHF, DPO), checkers (RLVR / success detectors). Always keep a leash and watch the true metric.

Saved in sources/: 07-dongxi-post-training.json, 07-dongxi-post-training.md

All 1 external references cited in this module

Reference	Site
post	x.com

LLMModule 08

The LLM research-engineer roadmap: 12 stages, from autograd to open-source teardowns

Link you sentx.com/suraj_sharma14/status/2112876560629825734

AuthorSuraj Sharma

Date21 Oct 2026

TypeX post

Summary: Twelve skills that turn papers into fast, well-tested models. Learn the internals, the data, the architectures, alignment, systems and inference, then reproduce and publish honestly. Most stages tie back to modules 01–07.

Followed through to:

62 mapped courses, docs and repos (verified 22 Oct 2026)

Saved in sources/: 08-suraj-ai-research-engineer.json, 08-suraj-ai-research-engineer.md

All 152 external references cited in this module

Reference	Site
ARENA · Transformer interpretability chapter	arena.education
PyTorch: An Imperative Style, High-Performance Deep Learning Library	arxiv.org
Automatic differentiation in machine learning: a survey	arxiv.org
FlashAttention: Fast and Memory-Efficient Exact Attention with IO-Awareness	arxiv.org
FlashAttention-2: Faster Attention with Better Parallelism and Work Partitioning	arxiv.org
Online normalizer calculation for softmax	arxiv.org
Squareplus: A Softplus-Like Algebraic Rectifier	arxiv.org
Scaling Laws for Neural Language Models	arxiv.org
Training Compute-Optimal Large Language Models	arxiv.org
Chinchilla Scaling: A replication attempt	arxiv.org
Deduplicating Training Data Makes Language Models Better	arxiv.org
DoReMi: Optimizing Data Mixtures Speeds Up Language Model Pretraining	arxiv.org
The FineWeb Datasets: Decanting the Web for the Finest Text Data at Scale	arxiv.org
SGDR: Stochastic Gradient Descent with Warm Restarts	arxiv.org
MiniCPM: Unveiling the Potential of Small Language Models with Scalable Training Strategies	arxiv.org
Outrageously Large Neural Networks: The Sparsely-Gated Mixture-of-Experts Layer	arxiv.org
Switch Transformers: Scaling to Trillion Parameter Models with Simple and Efficient Sparsity	arxiv.org
Efficiently Modeling Long Sequences with Structured State Spaces	arxiv.org
Mamba: Linear-Time Sequence Modeling with Selective State Spaces	arxiv.org
Transformers are RNNs: Fast Autoregressive Transformers with Linear Attention	arxiv.org
Ring Attention with Blockwise Transformers for Near-Infinite Context	arxiv.org
Self-Instruct: Aligning Language Models with Self-Generated Instructions	arxiv.org
WizardLM: Empowering Large Pre-Trained Language Models to Follow Complex Instructions	arxiv.org
STaR: Bootstrapping Reasoning With Reasoning	arxiv.org
Scaling Relationship on Learning Mathematical Reasoning with Large Language Models	arxiv.org
Textbooks Are All You Need	arxiv.org
The Curse of Recursion: Training on Generated Data Makes Models Forget	arxiv.org
Will we run out of data? Limits of LLM scaling based on human-generated data	arxiv.org
Rafailov, Sharma, Mitchell, Ermon, Manning and Finn (2023)	arxiv.org
Xu et al. (2024)	arxiv.org
Azar et al., 2023	arxiv.org
Ethayarajh, Xu, Muennighoff, Jurafsky and Kiela (2024)	arxiv.org
Hong, Lee and Thorne (2024)	arxiv.org
Ouyang et al., 2022	arxiv.org
Gao, Schulman and Hilton (2022)	arxiv.org
Coste, Anwar, Kirk and Krueger (2023)	arxiv.org
Singhal, Goyal, Xu and Durrett (2023)	arxiv.org
Bai et al., 2022	arxiv.org
Shoeybi, Patwary, Puri, LeGresley, Casper and Catanzaro (2019)	arxiv.org
Korthikanti et al., 2022	arxiv.org
Narayanan et al. (2021)	arxiv.org
Rajbhandari, Rasley, Ruwase and He (2020)	arxiv.org
Zhao et al., 2023	arxiv.org
Chen, Xu, Zhang and Guestrin (2016)	arxiv.org
CLIP: Learning Transferable Visual Models From Natural Language Supervision	arxiv.org
SigLIP: Sigmoid Loss for Language Image Pre-Training	arxiv.org
Flamingo: a Visual Language Model for Few-Shot Learning	arxiv.org
Visual Instruction Tuning (LLaVA)	arxiv.org
Improved Baselines with Visual Instruction Tuning (LLaVA-1.5)	arxiv.org
Chameleon: Mixed-Modal Early-Fusion Foundation Models	arxiv.org
Whisper: Robust Speech Recognition via Large-Scale Weak Supervision	arxiv.org
EnCodec: High Fidelity Neural Audio Compression	arxiv.org
RT-2: Vision-Language-Action Models Transfer Web Knowledge to Robotic Control	arxiv.org
OpenVLA: An Open-Source Vision-Language-Action Model	arxiv.org
π0: A Vision-Language-Action Flow Model for General Robot Control	arxiv.org
Fast Transformer Decoding: One Write-Head is All You Need (MQA)	arxiv.org
GQA: Training Generalized Multi-Query Transformer Models from Multi-Head Checkpoints	arxiv.org
Efficient Memory Management for LLM Serving with PagedAttention (vLLM)	arxiv.org
KIVI: A Tuning-Free Asymmetric 2bit Quantization for KV Cache	arxiv.org
KVQuant: Towards 10 Million Context Length LLM Inference with KV Cache Quantization	arxiv.org
Fast Inference from Transformers via Speculative Decoding	arxiv.org
Locating and Editing Factual Associations in GPT	arxiv.org
Dissecting Recall of Factual Associations in Auto-Regressive Language Models	arxiv.org
Does Localization Inform Editing?	arxiv.org
Interpretability in the Wild: a Circuit for Indirect Object Identification in GPT-2 small	arxiv.org
Towards Automated Circuit Discovery for Mechanistic Interpretability	arxiv.org
Eliciting Latent Predictions from Transformers with the Tuned Lens	arxiv.org
Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena	arxiv.org
Red Teaming Language Models with Language Models	arxiv.org
Tree of Attacks: Jailbreaking Black-Box LLMs Automatically	arxiv.org
Jailbreaking Black Box Large Language Models in Twenty Queries	arxiv.org
Universal and Transferable Adversarial Attacks on Aligned Language Models	arxiv.org
HarmBench: A Standardized Evaluation Framework for Automated Red Teaming and Robust Refusal	arxiv.org
Proving Test Set Contamination in Black Box Language Models	arxiv.org
A Careful Examination of Large Language Model Performance on Grade School Arithmetic	arxiv.org
LiveBench: A Challenging, Contamination-Limited LLM Benchmark	arxiv.org
Language Models are Few-Shot Learners	arxiv.org
Deep Reinforcement Learning that Matters	arxiv.org
Implementation Matters in Deep Policy Gradients	arxiv.org
Do Transformer Modifications Transfer Across Implementations and Applications?	arxiv.org
On the State of the Art of Evaluation in Neural Language Models	arxiv.org
Accounting for Variance in Machine Learning Benchmarks	arxiv.org
Deep RL at the Edge of the Statistical Precipice	arxiv.org
Model Cards for Model Reporting	arxiv.org
Datasheets for Datasets	arxiv.org
Pythia	arxiv.org
OLMo	arxiv.org
Show Your Work	arxiv.org
Improving Reproducibility in ML Research	arxiv.org
fast.ai · Practical Deep Learning	course.fast.ai
distilabel	distilabel.argilla.io
PyTorch docs · Extending PyTorch (custom autograd.Function)	docs.pytorch.org
Autograd mechanics	docs.pytorch.org
PyTorch · Getting started with FSDP	docs.pytorch.org
vLLM docs	docs.vllm.ai
Weights & Biases report	docs.wandb.ai
GPU MODE lectures	github.com
datatrove	github.com
Karpathy · build-nanogpt (reproduce GPT-2 124M)	github.com
Karpathy · nanochat	github.com
Hugging Face smol-course · Synthetic Data unit	github.com
nanoVLM	github.com
ARENA 3.0 materials (GitHub)	github.com
Baukit	github.com
EleutherAI · lm-evaluation-harness	github.com
Hugging Face · Evaluation Guidebook	github.com
Microsoft PyRIT	github.com
NVIDIA garak	github.com
nanoGPT	github.com
30u30 (this course's Module 06)	github.com
MIT 6.5940 · TinyML and Efficient Deep Learning Computing	hanlab.mit.edu
FineWeb: decanting the web for the finest text data	huggingface.co
Hugging Face · Mixture of Experts Explained	huggingface.co
Cosmopedia: how to create large-scale synthetic data	huggingface.co
Hugging Face TRL · DPO / KTO / ORPO trainers	huggingface.co
Hugging Face · The Ultra-Scale Playbook	huggingface.co
Hugging Face · Vision Language Models Explained	huggingface.co
Hugging Face · Robotics Course + LeRobot	huggingface.co
Transformers · SigLIP	huggingface.co
Hugging Face · KV cache explained	huggingface.co
Hub	huggingface.co
Hugging Face Hub · Model cards	huggingface.co
Hugging Face LLM Course	huggingface.co
Hub dataset cards	huggingface.co
UK AISI · Inspect	inspect.aisi.org.uk
Google DeepMind · How To Scale Your Model	jax-ml.github.io
Karpathy · Neural Networks: Zero to Hero	karpathy.ai
LLaVA project	llava-vl.github.io
Hamel Husain & Shreya Shankar · AI Evals for Engineers & PMs	maven.com
A Visual Guide to Mixture of Experts	newsletter.maartengrootendorst.com
A Visual Guide to Mamba and State Space Models	newsletter.maartengrootendorst.com
labml.ai annotated paper implementations	nn.labml.ai
OpenVLA	openvla.github.io
torch.autograd.gradcheck	pytorch.org
Nathan Lambert · RLHF Book	rlhfbook.com
rome.baulab.info	rome.baulab.info
The Annotated S4	srush.github.io
Stanford CS336 · Assignment 1 (Basics) + Assignment 2 (Systems)	stanford-cs336.github.io
Transformer Circuits Thread	transformer-circuits.pub
A Mathematical Framework for Transformer Circuits	transformer-circuits.pub
TransformerLens docs	transformerlensorg.github.io
Triton tutorials	triton-lang.org
Triton docs	triton-lang.org
Stanford CS25 · Transformers United	web.stanford.edu
DeepLearning.AI · Post-training of LLMs	deeplearning.ai
DeepLearning.AI · Reinforcement Learning from Human Feedback	deeplearning.ai
DeepLearning.AI · Efficiently Serving LLMs	deeplearning.ai
DeepSpeed · ZeRO tutorial	deepspeed.ai
LessWrong post	lesswrong.com
Neel Nanda · Getting started in mechanistic interpretability	neelnanda.io
Philip Kiely · Inference Engineering (book)	philipkiely.com
Suraj Sharma's roadmap post on X	x.com

LLMModule 09

LLM inference engineering: 25 interview problems, worked through with real numbers

Link you sentx.com/vishalsingh2972/status/2113830444722741637

AuthorVishal Singh

Date24 Oct 2026

TypeX post

Summary: Serving is arithmetic plus trade-offs: size the KV cache, keep the GPU busy with continuous batching, protect TTFT with chunked prefill, reuse prefixes, pick precision and GPUs by bandwidth, measure tails, and price it all from explicit assumptions.

Followed through to:

NVIDIA product pages (GPU specs) and papers: linked from each page

Saved in sources/: 09-vishal-inference-25.json, 09-vishal-inference-25.md

All 81 external references cited in this module

Reference	Site
Taming Throughput-Latency Tradeoff in LLM Inference with Sarathi-Serve	arxiv.org
DistServe	arxiv.org
Splitwise	arxiv.org
Efficient Memory Management for Large Language Model Serving with PagedAttention	arxiv.org
FlashAttention: Fast and Memory-Efficient Exact Attention with IO-Awareness	arxiv.org
FlashAttention-2: Faster Attention with Better Parallelism and Work Partitioning	arxiv.org
FlashAttention-3: Fast and Accurate Attention with Asynchrony and Low-precision	arxiv.org
Llama 2: Open Foundation and Fine-Tuned Chat Models	arxiv.org
SARATHI: Efficient LLM Inference by Piggybacking Decodes with Chunked Prefills	arxiv.org
Fairness in Serving Large Language Models	arxiv.org
SGLang: Efficient Execution of Structured Language Model Programs	arxiv.org
Mooncake: A KVCache-centric Disaggregated Architecture for LLM Serving	arxiv.org
Fast Inference from Transformers via Speculative Decoding	arxiv.org
Accelerating Large Language Model Decoding with Speculative Sampling	arxiv.org
EAGLE-3: Scaling up Inference Acceleration of LLMs via Training-Time Test	arxiv.org
Medusa: Simple LLM Inference Acceleration Framework with Multiple Decoding Heads	arxiv.org
FP8 Formats for Deep Learning	arxiv.org
LLM.int8(): 8-bit Matrix Multiplication for Transformers at Scale	arxiv.org
SmoothQuant: Accurate and Efficient Post-Training Quantization for LLMs	arxiv.org
GPTQ: Accurate Post-Training Quantization for Generative Pre-trained Transformers	arxiv.org
AWQ: Activation-aware Weight Quantization for LLM Compression and Acceleration	arxiv.org
Root Mean Square Layer Normalization	arxiv.org
ServerlessLLM: Low-Latency Serverless Inference for Large Language Models	arxiv.org
Microscaling Data Formats for Deep Learning	arxiv.org
GShard	arxiv.org
Switch Transformers	arxiv.org
DeepSpeed-MoE	arxiv.org
DeepSeek-V3 Technical Report	arxiv.org
Megatron-LM	arxiv.org
MDN · Using server-sent events	developer.mozilla.org
NVIDIA · Introducing NVFP4 for efficient and accurate low-precision inference	developer.nvidia.com
NVIDIA Nsight Compute	docs.nvidia.com
NVIDIA Dynamo documentation	docs.nvidia.com
Nsight Systems user guide	docs.nvidia.com
Nsight Compute profiling guide	docs.nvidia.com
DCGM feature overview	docs.nvidia.com
CUDA C++ programming guide	docs.nvidia.com
GenAI-Perf	docs.nvidia.com
Python asyncio · tasks and cancellation	docs.python.org
SGLang documentation	docs.sglang.ai
vLLM · OpenAI-compatible server	docs.vllm.ai
vLLM documentation	docs.vllm.ai
vLLM benchmark suite	docs.vllm.ai
vLLM docs · Paged attention kernel design	docs.vllm.ai
vLLM · Optimization and tuning	docs.vllm.ai
vLLM · Automatic prefix caching (design)	docs.vllm.ai
vLLM · Disaggregated prefilling	docs.vllm.ai
vLLM speculative decoding	docs.vllm.ai
vLLM bench serve	docs.vllm.ai
vLLM quantisation	docs.vllm.ai
metrics design page	docs.vllm.ai
vLLM · Metrics	docs.vllm.ai
vLLM · Parallelism and scaling	docs.vllm.ai
flash-attention repository	github.com
TensorRT Model Optimizer	github.com
lm-evaluation-harness	github.com
Kimi K3 repository and technical report	github.com
DeepEP	github.com
HTML Standard · Server-sent events	html.spec.whatwg.org
The Ultra-Scale Playbook	huggingface.co
How To Scale Your Model	jax-ml.github.io
KEDA · Concepts	keda.sh
Kubernetes · Horizontal Pod Autoscaler	kubernetes.io
TensorRT-LLM documentation	nvidia.github.io
OpenTelemetry semantic conventions for generative AI	opentelemetry.io
Prometheus: histograms and summaries	prometheus.io
torch.nn.RMSNorm	pytorch.org
PyTorch profiler	pytorch.org
The Tail at Scale	research.google
Google SRE Workbook · Canarying releases	sre.google
Google SRE Book · Handling overload	sre.google
Triton tutorial: layer normalisation	triton-lang.org
triton.testing.do_bench	triton-lang.org
NVIDIA H100 product page	nvidia.com
H200	nvidia.com
NVIDIA Jetson Orin	nvidia.com
DGX B200	nvidia.com
HGX	nvidia.com
L40S	nvidia.com
Orca: A Distributed Serving System for Transformer-Based Generative Models	usenix.org
Vishal Singh's post of 25 inference-engineering interview problems	x.com

LLM + RoboticsModule 10

Self-improving agent harnesses: Recursive Language Models, Continual Harness and Prime Agent

Link you sentx.com/yacinelearning/status/2113688068352004435

AuthorYacine Mahdid (with Seth Karten)

Date23 Oct 2026

TypeX post + 2h36m video (not watched)

Summary: Don't stuff the context: keep it in variables and query it with code and sub-calls (RLM). Let the agent refine its own harness in small, reversible steps without resets (Continual Harness). Read harness benchmarks as "model + harness".

Followed through to:

Saved in sources/: 10-prime-agent-blog.txt, 10-prime-agent-readme.md, 10-rlm-blogs.txt, 10-yacine-prime-agent-thumb.jpg, 10-yacine-prime-agent.json

All 14 external references cited in this module

Reference	Site
original blog post	alexzhang13.github.io
Prime Agent: A Self-Improving RLM Harness	arxiv.org
Recursive Language Models	arxiv.org
Continual Harness: Online Adaptation for Self-Improving Foundation Agents	arxiv.org
repo	github.com
rlm-minimal	github.com
library	github.com
verifiers	github.com
prime-rl	github.com
project page	sethkarten.ai
launch post	primeintellect.ai
Recursive Language Models: the paradigm of 2026	primeintellect.ai
“Prime Agent: Self-Improving RLM Harness”	youtube.com
X post	x.com

LLMModule 11

The LLM self-study stack: CS336, Build a Reasoning Model, the RLHF Book (and a gym plan)

Link you sentx.com/dongxi_nlp/status/2114597689228698006

AuthorDongxi (东锡)

Date26 Oct 2026

TypeX post (Chinese)

Summary: Build it to understand it: CS336 (pre-training and systems) → Raschka (reasoning, GRPO, distillation) → Lambert (post-training). Plan the hours honestly, and keep your body in the plan.

Followed through to:

Saved in sources/: 11-dongxi-young-people-ai.json, 11-dongxi-young-people-ai.md

All 6 external references cited in this module

Reference	Site
Build a Reasoning Model (From Scratch): code repository	github.com
Raschka's free video course	magazine.sebastianraschka.com
RLHF Book	rlhfbook.com
Stanford CS336, Spring 2025	stanford-cs336.github.io
video collection on Bilibili	bilibili.com
X post	x.com

LLM + RoboticsModule 12

AI training and inference, in nine videos: watch guides for Jia-Bin Huang's playlist

Link you sentwww.youtube.com/playlist

AuthorJia-Bin Huang

Datevideos Oct 2024 – Oct 2026

TypeYouTube playlist, 9 videos (not watched; chapter lists used)

Summary: Watch the mechanisms move: memory and parallelism, LoRA, GRPO, on-policy distillation, optimisers, speculative decoding and serving, each guide tied back to the course's deeper pages.

Followed through to:

Each video's references (ZeRO, PPO, GRPO, LoRA, Muon, SGLang, PagedAttention …): listed on each guide

Saved in sources/: 12-jiabin-huang-playlist.json

All 205 external references cited in this module

Reference	Site
ZeRO: Memory Optimizations Toward Training Trillion Parameter Models	arxiv.org
Adam: A Method for Stochastic Optimization	arxiv.org
Mixed Precision Training	arxiv.org
FP8 Formats for Deep Learning	arxiv.org
PyTorch FSDP: Experiences on Scaling Fully Sharded Data Parallel	arxiv.org
Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism	arxiv.org
Reducing Activation Recomputation in Large Transformer Models	arxiv.org
Ring Attention with Blockwise Transformers for Near-Infinite Context	arxiv.org
Online normalizer calculation for softmax	arxiv.org
GPipe: Efficient Training of Giant Neural Networks using Pipeline Parallelism	arxiv.org
PipeDream: Fast and Efficient Pipeline Parallel DNN Training	arxiv.org
Efficient Large-Scale Language Model Training on GPU Clusters Using Megatron-LM	arxiv.org
Zero Bubble Pipeline Parallelism	arxiv.org
DeepSeek-V3 Technical Report	arxiv.org
GShard: Scaling Giant Models with Conditional Computation and Automatic Sharding	arxiv.org
Switch Transformers: Scaling to Trillion Parameter Models	arxiv.org
The Llama 3 Herd of Models	arxiv.org
LoRA: Low-Rank Adaptation of Large Language Models	arxiv.org
LoRA+: Efficient Low Rank Adaptation of Large Models	arxiv.org
QLoRA: Efficient Finetuning of Quantized LLMs	arxiv.org
VeRA: Vector-based Random Matrix Adaptation	arxiv.org
DoRA: Weight-Decomposed Low-Rank Adaptation	arxiv.org
OpenVLA: An Open-Source Vision-Language-Action Model	arxiv.org
Fast Inference from Transformers via Speculative Decoding	arxiv.org
Accelerating Large Language Model Decoding with Speculative Sampling	arxiv.org
SpecInfer: Accelerating Generative LLM Serving with Tree-based Speculative Inference and Verification	arxiv.org
Medusa: Simple LLM Inference Acceleration Framework with Multiple Decoding Heads	arxiv.org
Better & Faster Large Language Models via Multi-token Prediction	arxiv.org
EAGLE: Speculative Sampling Requires Rethinking Feature Uncertainty	arxiv.org
EAGLE-2: Faster Inference of Language Models with Dynamic Draft Trees	arxiv.org
EAGLE-3: Scaling up Inference Acceleration of Large Language Models via Training-Time Test	arxiv.org
DFlash: Block Diffusion for Flash Speculative Decoding	arxiv.org
On-Policy Distillation of Language Models: Learning from Self-Generated Mistakes (GKD)	arxiv.org
MiniLLM: On-Policy Distillation of Large Language Models	arxiv.org
Distilling the Knowledge in a Neural Network	arxiv.org
Sequence-Level Knowledge Distillation	arxiv.org
Scheduled Sampling for Sequence Prediction with Recurrent Neural Networks	arxiv.org
A Reduction of Imitation Learning and Structured Prediction to No-Regret Online Learning (DAgger)	arxiv.org
Qwen3 Technical Report	arxiv.org
Asynchronous Methods for Deep Reinforcement Learning (A3C)	arxiv.org
High-Dimensional Continuous Control Using Generalized Advantage Estimation	arxiv.org
Trust Region Policy Optimization	arxiv.org
Proximal Policy Optimization Algorithms	arxiv.org
DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models (GRPO)	arxiv.org
DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning	arxiv.org
Understanding R1-Zero-Like Training: A Critical Perspective (Dr. GRPO)	arxiv.org
arXiv	arxiv.org
Old Optimizer, New Norm: An Anthology	arxiv.org
Muon is Scalable for LLM Training	arxiv.org
Kimi K2: Open Agentic Intelligence	arxiv.org
Fantastic Pretraining Optimizers and Where to Find Them	arxiv.org
SGLang: Efficient Execution of Structured Language Model Programs	arxiv.org
Efficient Memory Management for Large Language Model Serving with PagedAttention	arxiv.org
Taming Throughput-Latency Tradeoff in LLM Inference with Sarathi-Serve	arxiv.org
DistServe: Disaggregating Prefill and Decoding for Goodput-optimized LLM Serving	arxiv.org
Splitwise: Efficient Generative LLM Inference Using Phase Splitting	arxiv.org
XGrammar: Flexible and Efficient Structured Generation Engine for Large Language Models	arxiv.org
FlashAttention: Fast and Memory-Efficient Exact Attention with IO-Awareness	arxiv.org
DeepSpeed	github.com
vLLM	github.com
SGLang	github.com
The Ultra-Scale Playbook: Training LLMs on GPU Clusters	huggingface.co
DFlash 2: Keep Drafting Parallel	inco.ai
Deriving Muon	jeremybernste.in
Adaptive Subgradient Methods for Online Learning and Stochastic Optimization (AdaGrad)	jmlr.org
Muon: An optimizer for hidden layers in neural networks	kellerjordan.github.io
Simple Statistical Gradient-Following Algorithms for Connectionist Reinforcement Learning (REINFORCE)	link.springer.com
SGLang v0.4: zero-overhead batch scheduler, cache-aware load balancer, faster structured outputs	lmsys.org
SGLang HiCache: fast hierarchical KV caching	lmsys.org
Decoupled Weight Decay Regularization	openreview.net
On the importance of initialization and momentum in deep learning	proceedings.mlr.press
PyTorch FSDP	pytorch.org
On-Policy Distillation	thinkingmachines.ai
Neural Networks for Machine Learning, lecture 6 slides	cs.toronto.edu
Reinforcement Learning: An Introduction	incompleteideas.net
Manim Community	manim.community
Orca: A Distributed Serving System for Transformer-Based Generative Models	usenix.org
Watch on YouTube	youtube.com
0:00	youtube.com
0:55	youtube.com
1:56	youtube.com
11:40	youtube.com
12:40	youtube.com
13:21	youtube.com
16:07	youtube.com
19:27	youtube.com
21:28	youtube.com
Watch on YouTube	youtube.com
0:00	youtube.com
0:33	youtube.com
4:49	youtube.com
6:00	youtube.com
8:44	youtube.com
10:24	youtube.com
14:10	youtube.com
16:16	youtube.com
17:24	youtube.com
20:51	youtube.com
23:03	youtube.com
25:30	youtube.com
27:23	youtube.com
30:08	youtube.com
31:58	youtube.com
32:58	youtube.com
34:15	youtube.com
39:21	youtube.com
Watch on YouTube	youtube.com
0:00	youtube.com
1:32	youtube.com
6:34	youtube.com
9:34	youtube.com
10:21	youtube.com
16:23	youtube.com
18:26	youtube.com
21:23	youtube.com
Watch on YouTube	youtube.com
0:00	youtube.com
0:51	youtube.com
2:08	youtube.com
7:08	youtube.com
9:43	youtube.com
12:44	youtube.com
14:41	youtube.com
16:25	youtube.com
17:19	youtube.com
17:47	youtube.com
19:43	youtube.com
24:06	youtube.com
Watch on YouTube	youtube.com
0:00	youtube.com
1:28	youtube.com
4:11	youtube.com
9:30	youtube.com
12:11	youtube.com
14:08	youtube.com
16:39	youtube.com
19:04	youtube.com
20:30	youtube.com
22:43	youtube.com
24:47	youtube.com
25:52	youtube.com
29:15	youtube.com
30:04	youtube.com
Watch on YouTube	youtube.com
0:00	youtube.com
0:43	youtube.com
2:41	youtube.com
5:11	youtube.com
7:05	youtube.com
7:58	youtube.com
8:31	youtube.com
11:11	youtube.com
12:16	youtube.com
16:48	youtube.com
19:55	youtube.com
21:58	youtube.com
Foundations of Deep RL	youtube.com
DeepMind x UCL · Introduction to Reinforcement Learning	youtube.com
Watch on YouTube	youtube.com
Watch on YouTube	youtube.com
00:00	youtube.com
00:36	youtube.com
02:13	youtube.com
04:24	youtube.com
06:28	youtube.com
08:11	youtube.com
10:35	youtube.com
11:49	youtube.com
15:13	youtube.com
17:24	youtube.com
Watch on YouTube	youtube.com
0:00	youtube.com
1:00	youtube.com
1:56	youtube.com
4:31	youtube.com
6:01	youtube.com
7:53	youtube.com
8:59	youtube.com
9:46	youtube.com
10:49	youtube.com
11:58	youtube.com
13:02	youtube.com
13:40	youtube.com
14:40	youtube.com
15:23	youtube.com
16:40	youtube.com
17:47	youtube.com
19:03	youtube.com
20:45	youtube.com
22:56	youtube.com
24:08	youtube.com
25:28	youtube.com
26:38	youtube.com
27:44	youtube.com
29:14	youtube.com
30:50	youtube.com
32:16	youtube.com
33:18	youtube.com
34:14	youtube.com
35:13	youtube.com
36:13	youtube.com
39:10	youtube.com
40:21	youtube.com
“AI Training and Inference” playlist	youtube.com
FlashAttention	youtu.be

LLMModule 13

100 days of inference engineering: a learning journal, the book behind it, and a plan you can follow

Link you sentgithub.com/pavelsimo/100-days-of-inference-engineering

AuthorPavel Simo

Datestarted 29 Oct 2026

TypeGitHub learning journal (no licence; summarised, not copied)

Summary: One chapter of Kiely's book per block, one build per chapter, one written note per day, shared publicly. Start with the smallest model that passes your evals.

Followed through to:

Inference Engineering by Philip Kiely (Baseten)

Saved in sources/: 13-100-days-inference-readme.md

All 2 external references cited in this module

Reference	Site
100-days-of-inference-engineering	github.com
Inference Engineering	baseten.co

LLM + RoboticsModule 14

Deep-ML's Deep-Path: 103 concepts from classical ML to GPU kernels, mapped onto this course (with a placement check)

Link you sentwww.deep-ml.com/deep-path/diagnostic

AuthorDeep-ML

Dateread 31 Oct 2026

TypeInteractive learning route + placement test (test needs sign-in; not taken)

Summary: Find your frontier, then drill a few steps a day with spaced review. Read the course page to understand a concept, and do the Deep-Path problems to make it stick. Robotics and control is the thinnest part of the map.

Followed through to:

Saved in sources/: 14-deep-ml-deep-path.json

All 2 external references cited in this module

Reference	Site
Deep-ML Deep-Path placement test	deep-ml.com
Deep-Path map	deep-ml.com

LLMModule 15

Inference Engineering, interactive edition: how to use inferenceengineering.tech alongside this course

Link you sentinferenceengineering.tech/

AuthorPhilip Kiely (book) · digitized by Swapnil

Dateread 3 Nov 2026

TypeInteractive book companion site

Summary: Read the book chapter, run its calculator, then cross-check against the course's own demo and explain any gap. Treat single-setup benchmarks as a starting point.

Followed through to:

Saved in sources/: 15-inferenceengineering-tech-home.html

All 9 external references cited in this module

Reference	Site
inferenceengineering.tech	inferenceengineering.tech
chapters	inferenceengineering.tech
paths	inferenceengineering.tech
calculators	inferenceengineering.tech
cheat sheets	inferenceengineering.tech
glossary	inferenceengineering.tech
benchmarks	inferenceengineering.tech
about	inferenceengineering.tech
Baseten's official book page	baseten.co

LLMModule 16

Hands-On LLM Serving and Optimization: the book's code as a lab track

Link you sentgithub.com/orca3/llm-model-inference

AuthorChi Wang & Peiheng Hu

Daterepo updated Aug 2026

TypeGitHub companion code for an O'Reilly book (no licence file; summarised)

Summary: Build a small server yourself (API, scheduler, worker process), then swap in real engines and optimise one change at a time on realistic data. Batching trades latency for throughput, and continuous batching mostly removes that trade.

Followed through to:

Saved in sources/: 16-orca3-ch03-single-model-readme.md, 16-orca3-ch09-readme.md, 16-orca3-llm-model-inference-readme.md

All 6 external references cited in this module

Reference	Site
vLLM docs	docs.vllm.ai
orca3/llm-model-inference	github.com
llama.cpp	github.com
LMCache	github.com
book website	orca3.github.io
O'Reilly page	oreilly.com

LLM + RoboticsModule 17

391 free AI courses, checked: a searchable catalog with link health, provenance, and 2026 updates

Link you sentx.com/shikamarutsai/status/2088918066310652358

Authorshikamaru (@shikamarutsai)

Date16 Aug 2026

TypeX article (a reproduction of an uncredited GitHub list)

Summary: A huge free list is only useful once it's filtered and verified. Keep the classics for foundations, add post-2022 courses for LLMs and robotics, and prefer courses with assignments over lectures alone.

Followed through to:

Original list: Deep Learning Drizzle (kmario23)

Saved in sources/: 17-courses-with-link-status.json, 17-link-status-raw.json, 17-shikamaru-free-ai-curriculum.json

All 23 external references cited in this module

Reference	Site
fast.ai	course.fast.ai
Stanford CS224W	cs224w.stanford.edu
Stanford CS231n notes	cs231n.github.io
kmario23/deep-learning-drizzle	github.com
LeRobot	github.com
MIT 6.5940 TinyML & Efficient DL	hanlab.mit.edu
Hugging Face LLM Course	huggingface.co
Hugging Face Deep RL Course	huggingface.co
Hugging Face Robotics Course	huggingface.co
Isaac Lab	isaac-sim.github.io
Karpathy, Zero to Hero	karpathy.ai
MIT Robotic Manipulation	manipulation.csail.mit.edu
Modern Robotics	modernrobotics.northwestern.edu
MuJoCo	mujoco.org
Berkeley CS 185/285 Deep RL	rail.eecs.berkeley.edu
RLHF Book	rlhfbook.com
Stanford CS336	stanford-cs336.github.io
MIT Underactuated Robotics	underactuated.mit.edu
Stanford CS224n	web.stanford.edu
Coursera specialisation	coursera.org
CMU 16-385	cs.cmu.edu
DeepLearning.AI, Post-training of LLMs	deeplearning.ai
X article	x.com

LLM + RoboticsModule 18

Build a Robot Brain: 12 weeks, from zero to a robot that understands you (ages 15+)

Link you senttopics/18-robot-brain/index.html

AuthorWritten for this course (from modules 01-17)

DateNov 2026

TypeOriginal hands-on course + Python project

Summary: Learn by building: each idea from first principles, then real code, a self-check and a real-world challenge, growing into one working robot that's honestly tested. Then inference engineering: a tiny LM from zero, latency and cost, caching/quantising/distilling, Claude as planner and agent, and a bake-off of every brain.

Followed through to:

Project code: homework, solutions, check.py

Saved in sources/: none

All 2 external references cited in this module

Reference	Site
LeRobot	github.com
robotics course	huggingface.co

LLMModule 19

Evals: how to know whether an AI system actually works (a field guide, from one test case to a release gate)

Link you sentx.com/sermakarevich/status/2106453816757354947

AuthorSergii Makarevych

Date3 Oct 2026

TypeX article (34 min read)

Summary: An eval is input, system, grader, number. Read outputs before writing graders, use code for anything definite, and trust an AI judge only after measuring it (kappa, not raw agreement), with the evidence in front of it. Every score gets an error bar; agents get pass^k.

Followed through to:

Saved in sources/: 19-makarevych-evals.json

All 8 external references cited in this module

Reference	Site
Adding Error Bars to Evals	arxiv.org
Who Validates the Validators?	arxiv.org
Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena	arxiv.org
τ-bench	arxiv.org
RAGAS	arxiv.org
Your AI Product Needs Evals	hamel.dev
AI Evals FAQ	hamel.dev
X article	x.com

LLM + RoboticsModule 20

Six tiny repos that rebuild modern AI from scratch: micrograd, minGPT, nanoGPT, llm.c, tinygrad, minimalRL (run and tested)

Link you sentgithub.com/karpathy/micrograd

AuthorAndrej Karpathy; tiny corp; Seungeun Rho

Datefetched 7 Nov 2026

TypeSix GitHub repos (all MIT), cloned and run

Summary: Read tiny code that does the real thing, run it, then use it to try to make your own robot better, and measure whether it did. Bigger models need more data; a fast brain on a laptop can be too slow on a robot; learned control is great until the world changes.

Followed through to:

Saved in sources/: 20-llm.c-readme.md, 20-micrograd-readme.md, 20-minGPT-readme.md, 20-minimalRL-readme.md, 20-nanoGPT-readme.md, 20-nanochat-readme.md, 20-tinygrad-readme.md

All 34 external references cited in this module

Reference	Site
Attention Is All You Need	arxiv.org
Gaussian Error Linear Units (GELU)	arxiv.org
GPT-3 paper	arxiv.org
tinygrad docs	docs.tinygrad.org
microgpt	gist.github.com
karpathy/micrograd	github.com
karpathy/minGPT	github.com
openai/gpt-2	github.com
openai/image-gpt	github.com
huggingface/transformers	github.com
karpathy/nanoGPT	github.com
karpathy/nanochat	github.com
original post (13 Oct 2025)	github.com
"Beating GPT-2 for <<$100: the nanochat journey" (1 Feb 2026)	github.com
modded-nanogpt	github.com
karpathy/llm.c	github.com
Discussion #481: reproducing GPT-2 (124M) in llm.c	github.com
tinygrad/tinygrad	github.com
examples/beautiful_mnist.py	github.com
PyTorch	github.com
commaai/openpilot	github.com
seungeunrho/minimalRL	github.com
REINFORCE.py	github.com
dqn.py	github.com
ppo.py	github.com
Gymnasium migration guide	gymnasium.farama.org
OpenWebText	huggingface.co
llmc-starter-pack	huggingface.co
Neural Networks: Zero to Hero	karpathy.ai
explainer post	karpathy.github.io
PyTorch	pytorch.org
Tiny Shakespeare	raw.githubusercontent.com
tinygrad.org	tinygrad.org
GPT video	youtube.com

LLM + RoboticsModule 21

The maths behind robots, learned by playing: ten interactive explainers, each one used to make Robo-Helper better

Link you sentx.com/0x_meden/status/2119507343348212198

AuthorMartin (@0x_meden)

Date8 Nov 2026

TypeX post: 10 interactive explainers (all links checked 9 Nov 2026)

Summary: Play with each idea first, then use it on your own robot and measure it. Estimation (Bayes, Kalman, Fourier) beats plain averaging once the robot moves; a model of the world only helps if the world has patterns; π0 and GR00T are a scene-understanding Transformer plus a noise-to-motion head.

Followed through to:

Saved in sources/: 21-meden-pi0-groot.json, 21-meden-robot-math.json

All 17 external references cited in this module

Reference	Site
Andrej Karpathy: REINFORCEjs	cs.stanford.edu
Gabriel Goh: Why Momentum Really Works (Distill)	distill.pub
openpi	github.com
Isaac-GR00T	github.com
LeRobot	github.com
Immersive Math: “Immersive Linear Algebra”	immersivemath.com
Polo Club (Georgia Tech): Transformer Explainer	poloclub.github.io
Polo Club: Diffusion Explainer	poloclub.github.io
Seeing Theory (Brown University)	seeing-theory.brown.edu
Setosa: Markov chains, explained visually	setosa.io
Tim Babb: How a Kalman filter works, in pictures	bzarg.com
Jez Swanson: An Interactive Introduction to Fourier Transforms	jezzamon.com
A* page	redblobgames.com
Stanford AA203 lecture 15	youtube.com
Berkeley CS 285 lecture 2	youtube.com
X post	x.com
“How π0, GR00T and Optimus actually think”	x.com

LLM + RoboticsModule 22

Gradient, divergence and curl: three ways a field of arrows moves, and how a robot steers with them

Link you sentx.com/matheorems/status/2119505986381901899

AuthorMatheorems

Date8 Nov 2026

TypeX post with an animation

Summary: Gradient = uphill, divergence = spreading out or draining in, curl = swirl. A robot can drive by following a field, but pure gradients get stuck in dips; a little curl around obstacles gets it round. Measure from many starts, not one.

Followed through to:

3Blue1Brown: Divergence and curl

Saved in sources/: 22-matheorems.json

All 3 external references cited in this module

Reference	Site
Multivariable calculus	khanacademy.org
Divergence and curl: the language of Maxwell's equations, fluid flow, and more	youtube.com
X post	x.com

Generated by tools/build_sources.py from the module pages and the course index. Re-run it after each ingest. · Back to course index