The first two lectures of the Scaling Up unit in CMU 10-423 Spring 2026 answer two questions. The second half of L15 covers scaling laws: Kaplan 2020 says 8x more parameters needs only about 5x more data, Chinchilla says scale both equally, and the Phi models and data-filtering scaling laws add data quality as a third axis. L16 covers MoE: feed-forward layers hold most of GPT-3's parameters, so split them into experts and send each token through only the top k. Memory follows total parameters, compute follows active parameters, and the price is load balancing and training stability. No programming homework covers this half of the course; quizzes, practice exam question 13, and the final project do.
Lecture 10 of 11-868 uses Lei Li's own LightSeq and LightSeq2 as the case study and breaks them into four techniques: fuse every small operation outside matrix multiplication into one kernel, rewrite the LayerNorm and Softmax formulas to cut thread synchronizations, store parameters and gradients in FP16 but compute updates in FP32, and reuse memory based on backward-pass dependencies. The slides report 1.4-3.5x training speedups on WMT14 English-German. There is no recording; this guide works from slide page numbers and the two papers.
CMU 11-868's first lecture spends 51 slides on one argument: the LLM bottleneck isn't only the model, it's computing larger LLMs on bigger datasets with fewer GPUs, less memory, and less power, faster. It breaks a Transformer into four low-level operators (matrix multiply, reduction, map, memory movement), sorts the hard problems into kernel, framework, and distributed-system layers, and warns that fast computation isn't enough because moving data takes time too.
CMU 11-868 is Lei Li's graduate course on LLM systems: it goes from CUDA kernels and your own MiniTorch framework to distributed training, SGLang serving, and RLHF. All 28 Spring 2026 slide decks, 7 assignment pages, and 7 starter-code repos are public, which rates it A3. What's missing: videos, GPUs and a PSC account, the quizzes, and any official statement of which two assignments are optional.
In Lecture 12, 6.5940 switches from CNNs to Transformers. The lecture doesn't dwell on theory. It points to where memory and compute go. Attention is O(N²). If Llama-2-70B used MHA, its KV cache at batch 16 and length 4096 would take 160GB. GQA shrinks that 8x and MQA shrinks it 64x. MoE adds total parameters while keeping per-token compute flat. This post bridges into Lecture 13 on LLM deployment.
L04 reduces a large language model to one sentence: look at the preceding words, score every word in the vocabulary, turn the scores into probabilities with softmax, and sample the next word. To give the model a memory of what came before, the lecture covers RNNs and then gives a first look at Transformer Q/K/V. GPT-2's 1.5 billion and GPT-3's 175 billion parameters illustrate scale; temperature and top-p explain why every answer comes out different. The second half covers running open models locally and estimating VRAM. Week 4 homework: write test prompts on a topic you know well and compare at least two LLMs.
The first half of L06 is about ethics. Yen-Lung Tsai quotes Karpathy's line that hallucination is a feature of LLMs, then works through plagiarism, whether your data gets used for training, and DeepSeek's censorship and corpus skew, and closes with seven principles of responsible use. The second half is about applications: give the model the right information and clear instructions, and one system prompt becomes a Lucky Vicky positivity generator, a social-media copywriter, or a biased college-major counselor. The week-6 assignment moves that prompt into an OpenAI-compatible API with a Gradio front end: a chatbot with a persona.
When an LLM is too big to fine-tune in full, the LLM Adaptation slides of NTU ADL Fall 2025 offer three ways to change only a small part of it: insert small Adapter modules into the Transformer, represent the weight update with low-rank matrices (LoRA), or learn only a prefix or soft prompt (prompt tuning). The slides conclude that no single method fits every task. For HW2, the only public information is its title, "LLM Tuning and Prompt Tuning for Classical Chinese Translation"; the data, baseline, and grading have no written spec.
A pre-trained model can continue text, but that does not mean it follows instructions. The Post-Training slides of NTU ADL Fall 2025 fix this in two steps. Instruction tuning (FLAN, T0) teaches the model to read task descriptions. RLHF then pulls its outputs toward human preference. Three limits of instruction tuning connect the two steps, a reward model and pairwise comparisons solve two practical RL problems, and InstructGPT's SFT → reward model → PPO pipeline ties it all together. ChatGPT runs the same pipeline on multi-turn dialogue.
LLMs cannot memorize long-tail facts, their knowledge goes stale, and they cannot see private documents. The RAG slides of NTU ADL Fall 2025 open with an LLM hallucinating about the lecturer herself, then split RAG into indexing, retrieval, and generation: sparse (TF-IDF, BM25) and dense (DPR, Contriever) retrieval, how dense retrievers are trained, pre- and post-retrieval techniques including pointwise and pairwise reranking. A closing roadmap organizes RAG, RETRO, FLARE, Search-R1, and others by what, how, and when to retrieve. For HW3, only the title is public: "Retriever & Reranker Training for RAG".
HW7 hands you two models fine-tuned from Mistral-7B-v0.1: shisa-gamma-7b-v1, strong in Japanese, and WizardMath-7B-V1.1, strong in math. You may only merge them at the parameter level (no further training, no MoE or ensembles), and the merged model has to answer 20 Japanese math questions written by a TA. Part 1 (60%) is tuning the method, weights, and density in mergekit, with simple and strong baselines at 50% and 75% accuracy. Part 2 (40%) is 8 multiple-choice paper questions. The spec, Colab, and Kaggle notebook are public, but JudgeBoi returned 502 on 2026-09-30 and the paper questions live on NTU COOL, so outside readers can only check accuracy inside the notebook.
CME295 Lecture 3 defines an LLM as a decoder-only next-token predictor, uses MoE to explain why a huge model only touches part of its weights per token, and spends most of its time on the knobs you can turn at generation time: greedy, beam search, top-k, top-p, temperature, guided decoding, plus three prompting techniques (few-shot, chain of thought, self-consistency). The 2026 edition folds this lecture into Lecture 2, and the prompting half disappears from the syllabus.
The first CS224U lecture of Spring 2023 asks "Which U.S. states border no U.S. states?" of every system from Chat-80 (1980) to text-davinci-001. The answers show that the progress is real. The lecture then questions whether that progress counts as understanding, using Levesque's "cheap tricks," models that invent links, and benchmarks that saturate within a year or two. That splits the course map in two: the first half teaches you to build systems with Transformers and retrieval-augmented in-context learning, and the second half teaches you to test them with harder benchmarks, behavioral evaluation, and causal explanation methods.
CME295 is a two-unit Stanford course with no homework; your grade is the midterm and the final, 50% each. The 2025 edition's nine lectures are fully public: videos, slides, and both exams with solutions. The 2026 edition rewrites the agent lecture around context compaction, harnesses, coding agents, and skills, and adds three full lectures on LLM systems, reinforcement learning, and Diffusion LLMs.
Lec 23 wires a transformer into a next-token predictor: tokenize, look up embeddings, stack L layers of masked attention, multiply back by the embedding table and apply softmax, and train with cross-entropy (that is, MLE). Pretraining supplies knowledge; to chat, a model also needs SFT, LoRA, RLHF, or DPO, and at inference time it leans on in-context learning, RAG, chain-of-thought, and tool calls. Lec 24 generalizes "invent a fake supervised task" to images: autoencoders, colorization, inpainting, rotation, jigsaw puzzles, clustering, and finally contrastive learning, SimCLR, and CLIP. Discussion 11 practices positional encodings, RoPE, causal masks, and the KV cache.
The entire Deep Research field has 80+ implementations, but the core structure is just three-stage roadmap × four components × three optimization methods. This article maps the full landscape: from Agentic Search to Full-stack AI Scientist, from query planning to answer generation, from workflow prompting to end-to-end RL.
An AI assistant platform running Opus 4.6 hit stream_stall (90s timeout) twice consecutively when generating docx. Root cause: skill instructions lacked one sentence — 'Write a script' — causing the model to output JS code inline instead of writing a file and executing with node. Claude.ai's official SKILL.md has that sentence, and the model consistently takes the safe path.
Gemma is Google's open-weights family paired with the closed-source Gemini flagship; Gemma 4 (E2B / E4B / 26B-MoE / 31B) released in April 2026 switched its license from Google Gemma Terms of Use to Apache 2.0 — the single most important change in this article.
Laguna is Poolside's agentic coding model family: XS 2.1 packs 33B-A3B into a 36GB Mac, while S 2.1 brings 118B-A8B with 1M context to 70.2% on Terminal-Bench 2.1 and 40.4% on DeepSWE, both open under OpenMDW-1.1.
Ant Group's Ling model family deep-dive: 2025→2026 evolution timeline, Ling/Ring/Ming three-series strategy, architecture journey from Ling 1.0 to Ling 3.0, Ling-3.0-flash-Fin finance model, and an Agent developer's selection guide
Muse Spark is Meta's closed-source agentic model line: version 1.3 combines a 1M-token context, multimodal inputs, and long-horizon tool loops. Standard pricing is $1.25/$4.25 per 1M input/output tokens, while Contributor drops to $0.10/$0.20 in exchange for training rights. It is not the next Llama; it is a separate product line built around models, APIs, and coding agents.
Nex-N2.5 is Nex AGI's open agentic model family: mini scores 82.9 on OSWorld-G at 35B-A3B, Pro tops Claude Opus 5 with 87.4 at 397B-A17B, and Max leads the whole official table on BrowseComp with 92.6 at 1.6T, all open under Apache-2.0.
Four AI security certs, four different bets: SecAI+ ($359) is CompTIA's mid-level expansion play, CAISP ($999 all-in) has the strongest hands-on labs and fullest OWASP LLM Top 10 coverage, GAIPS ($999/$9K) is the SANS gold-standard defender cert with CyberLive exams, and AAISM ($459+) is governance-layer but requires CISM or CISSP first. Under $400 → SecAI+. Want to actually hack and fix → CAISP. Company paying → GAIPS.
A close look at three open-source projects training LLMs from scratch in the Chinese community — baby-llama2-chinese (218M, 63.4B tokens), ChatLM-mini-Chinese (0.2B T5, 10.23M dialogues), and Steel-LLM (1.12B, 1T tokens, 8 months) — comparing corpus strategy, tokenizer decisions, and community ecosystem. Honest evaluation included: baby-llama2 scored a bottom-ranking 21 in MiniMind's side-by-side test; ChatLM has the strongest knowledge (62) but weak coding.
OpenELM uses layer-wise scaling to shift parameters toward layers near the output; with 1.08B parameters and 1.5T tokens it beats OLMo 1.2B (+2.36% on the LLM360 average) despite OLMo training on 3T tokens. MiniCPM trains multimodal small models from scratch with a three-stage unfreezing recipe (Resampler first, vision encoder next, everything unfrozen last); MiniCPM-V 4.5 reaches sub-30B SOTA on VideoMME with only 8B parameters, and 4-bit quantization squeezes fp16's 16–17GB memory footprint down to about 5GB for phones.
A tour of Karpathy's three teaching repos: nanoGPT (2022, ~300 lines each to reproduce GPT-2 124M), llm.c (pure C/CUDA training), and nanochat (2025-10, one speedrun.sh from tokenizer to WebUI). $100 and 4 hours on 8×H100 buys a chatty model; GPT-2-grade capability is now down to about 2 hours and $48.
LitGPT (Lightning AI, ~13,600 stars, Apache 2.0) rewrites 20+ mainstream LLMs — Llama 3, Qwen2.5, Phi 4 — from scratch as single-file, no-abstraction implementations, with a full pretrain / finetune / evaluate / serve CLI. TinyLlama (1.1B parameters, 3T tokens) was trained on this codebase. This post breaks down how it differs from MiniMind, how to actually use it, and where it stops.
MiniMind is an open-source project for training LLMs from scratch: a 64M Dense model and a 198M-A64M MoE model that run the entire chain — Pretrain → SFT → LoRA → DPO → PPO/GRPO/CISPO → Agentic RL — in ~2 hours on a single RTX 3090 at roughly 3 RMB (~$0.40). Every core algorithm is implemented natively in PyTorch with no high-level wrappers.
Switzerland's Apertus (8B/70B, 15T tokens, 1,000+ languages) filters opt-outs and personal data before training to satisfy the EU AI Act; Japan's LLM-jp consortium shipped LLM-jp-4 (12T tokens) in April 2026, claiming wins over GPT-4o and Qwen3-8B on standard benchmarks. Both prove that from-scratch training outside the English sphere is a data-governance problem, not a technical one — plus a note on RWKV-7 as the non-Transformer alternative.
"Open-source LLM" is a spectrum: weights-only (Llama), weights plus data (most fully open projects), or data order, intermediate checkpoints, and training logs all released (LLM360 K2, OLMo 3's model flow). This piece unpacks the two projects that pushed transparency furthest: OLMo 3 shipped the first fully open 32B thinking model in November 2025, and K2 is the first 65B-class model whose checkpoints even include optimizer states.
The hands-on installment of the series: rent an RTX 3090 on RunPod (Secure Cloud $0.5/hr, Community Cloud $0.22/hr), follow the MiniMind README through pretrain (~1.21h) + SFT (~1.10h), spend roughly $0.55–1.50 USD total, and chat with your own 64M model trained from scratch in the terminal.
Open-source projects have pushed the cost of training an LLM from scratch absurdly low: MiniMind runs the full PreTrain-to-RL pipeline for about $0.4 (2 hours on a single RTX 3090), while at the other end OLMo 3 and LLM360 K2 publish everything — data, code, and stage-by-stage checkpoints of 65B models. This series walks the whole project spectrum from $0.4 to 65B in 11 articles, and flags where the map is biased.
Training from scratch only makes sense in three cases: you want to learn how training works, you have 10B+ clean tokens no open model has seen, or you need a fully transparent training process for research. Otherwise fine-tuning or RAG is almost always cheaper. This post collapses the series' main routes into one cost ladder and a decision tree.
YuLan-Mini is a 2.4B open-source model from Renmin University's AI Box lab, trained on 48 A800 GPUs with only 1.08T tokens — scoring 37.8 on MATH-500 and 64.0 on HumanEval, beating Qwen2/Qwen2.5 peer models trained on 7T–18T tokens at math and code. What's public isn't a slogan: per-phase data mixes, pre-annealing optimizer states, and even W&B logs of the ablation studies.
Thinking Machines Lab (founded 2025 by Mira Murati, $2B seed at a $12B valuation) released Inkling in July 2026 under Apache 2.0 (975B total / 41B active params, 1M context, native multimodality, controllable thinking effort) plus a smaller Inkling-Small (276B / 12B), paired with the Tinker fine-tuning platform—turning customizability itself into the product.
looplane now has static ModelRole/ModelRoute candidates, opt-in aliases such as --model @cheap, cross-provider fallback, and a no-tool reviewer lane that runs after verification. Role inheritance/override rules and automatic summarizer, parser, or scout routing remain open.
AI Gateway is the control plane for AI calls: one layer for logs, analytics, cache, rate limits, retry/fallback, BYOK, and Unified Billing. In Workers, use env.AI.run(..., { gateway }); with external SDKs, change the baseURL or provider-native endpoint.
Looplane's native lane is controlled by AgentRunner: prepare a workspace, request a model turn, execute tool calls, append observations, and enter verification only when the model stops calling tools. Step, wall-time, repetition, token, and cancellation guards can terminate the run independently of the model. Protocol translation belongs to the next article.
GPUtw.ai makes sense as a short-rental GPU learning tool: start with Jupyter, Ollama, or ComfyUI, then try LoRA/QLoRA on a small model. It is not a large foundation-model training platform, and the first run should verify deployment, billing, and data retention with a small budget.
The same model can cost 2-5× more depending on the channel. Direct API is simplest, aggregators (OpenRouter) are most flexible, cloud platforms (Bedrock/Vertex) suit enterprises. This post compares actual August 2026 prices across six channels with a decision tree.
Allen AI's OLMo is the only language model family that fully publishes weights, training data (Dolma, 9.3T tokens), training code, all intermediate checkpoints, and evaluation tools. OLMo 3's 32B Think model hits 96.1% on MATH — and you can use OlmoTrace to trace any output back to the exact training data that produced it.
Models don't read words — they read tokens. A Chinese character is typically 1-2 tokens; an English word is 1-3. The context window is the token limit per request. Inference is using a model; training is teaching one. What you do every day is inference.
Data changes often and you need citations → RAG. Need consistent style or want to run on a small device → fine-tuning. In practice, many production systems use both: fine-tune a small model that speaks your domain language, then use RAG to supply up-to-date facts.
Scaling laws show that loss decreases predictably with more parameters, data, and compute — following power-law relationships. The Chinchilla paper's key finding: most models were too large and undertrained. Given the same compute budget, training a smaller model on more data produces better results. This reshaped the entire industry's training strategy.
You don't need to become a researcher to understand AI models systematically. This series starts from what you can see (tokens, context windows) and works up to self-hosting open-source models — 18 articles covering everything you need to choose models, read benchmarks, and estimate costs.
Models charge by tokens, not characters. The BPE algorithm starts from individual bytes and repeatedly merges the most frequent adjacent pair to build a vocabulary. English 'understanding' might be 1-2 tokens, but Chinese '理解' could take 2-3 — same meaning, higher cost.
Computational biology PhD turned UW-Madison professor Sebastian Raschka launched Ahead of AI on Substack in 2022, publishing monthly deep dives into LLM papers and architectures. Four years later: 200K+ subscribers, zero sponsorships, and a book-newsletter flywheel that proves low frequency and high depth can win in a crowded AI newsletter market.
LLMs break unified diffs on bookkeeping: wrong hunk counts, hallucinated context lines. The five reference projects split into two camps — simplify the diff grammar (Codex drops line numbers), or drop diffs entirely (Claude Code/Pi/OpenCode exact replace); OMP goes further by binding read state into the format via hash anchors. looplane took the minimal-intervention path: keep the guarded apply_patch, add a zero-fuzzy replace_text, and its qwen3:4b eval went from stable failure to 5/5.
Unsloth is the fastest, most VRAM-efficient local LLM fine-tuning tool — 2× training speed and 70% less VRAM. In 2026 it added a Desktop app that bundles inference, training, image/video generation, web search, and agent integration into a complete local AI workstation.
Apple Foundation Models (AFM) is Apple's closed-ecosystem AI family. It evolved from a 3B dense model with LoRA adapters in 2024 into five models in 2026. AFM 3 Core Advanced runs a 20B IFP sparse architecture on phones while activating only 1–4B parameters; Cloud Pro runs on Google Cloud NVIDIA GPUs and is refined through Gemini distillation. There is no public API price or third-party benchmark, and access is limited to Apple's Foundation Models framework.
In 2023, LLMs took over the machine-learning conference agenda. NeurIPS received more than 12,000 submissions; both Outstanding Papers addressed large models, while runner-up DPO became a practical alternative to RLHF within two years. DreamFusion opened the text-to-3D field, ICML spotlighted LLM watermarking and learning-rate adaptation, and the Mamba preprint emerged as the first serious architectural challenger to the Transformer.
2023 was the first year in which LLMs comprehensively rewrote the AI research agenda. DPO received a NeurIPS Outstanding Paper Runner-Up award, ReAct became an ICLR Oral, and hallucination grew from a marginal term into a major track at every conference. Meanwhile, 3D Gaussian Splatting swept through computer vision after its SIGGRAPH debut, Mamba emerged at the end of the year to challenge the Transformer attention monopoly, and publication volume for traditional NLP pipelines began a clear decline.
Stanford Marin pre-registers a paloma macro-loss of 2.04 with a 5-rung Scaling Ladder at 1% cost, then trains 535B-A23B on 11×GB200 in public with live W&B telemetry — 847 training buckets already show the most teachable frontier run.
You cannot take model vendors' self-reported scores at face value. This guide covers the most important independent evaluation platforms, domain benchmarks, adoption indicators, and official sources in 2026: what each measures, how to read it, where it is biased, and which figures matter for different use cases.
Claude is Anthropic's closed-source LLM family, known for Constitutional AI training, agent capabilities, and coding performance. In July 2026, Opus 5 scored 96% on SWE-bench Verified to claim the coding crown, while Fable 5 led general capability at 83% on LiveBench. Four tiers (Fable / Opus / Sonnet / Haiku) span $1–$10, making this the only family in the series with zero open weights.
Cohere is the only family that ships generation, retrieval, reranking, and multilingual as distinct products. Command A runs 256K context on two GPUs at 111B, Embed v4 does mixed image-text retrieval, Rerank v4 handles 32K semi-structured data, and Aya covers 101 languages — a four-piece stack built for RAG. This post breaks down each pillar's positioning, licensing, and selection guide.
DeepSeek used MLA and MoE innovations to drive inference costs to an industry low. V4 Flash activates only 13B parameters while approaching frontier-model quality and ranks first by OpenRouter usage. This guide traces V1 through V4, the R1 reasoning branch, and how to choose each version.
Gemini is Google DeepMind's native multimodal LLM family, famed for a 1M-token context window and native video/speech input plus scientific reasoning. 3.1 Pro tops GPQA Diamond 94.1% and ARC-AGI-2 77.1% to claim science-reasoning dual crowns, at $2/$12—1/6 of Claude. 3.7 Flash delivers near-Pro agent capability for $0.75/$3.75.
GLM is Zhipu AI (Z.ai)'s open LLM family from Tsinghua's KEG Lab. GLM-5.3 (2026/08) lifts coding +50% over the previous generation, hits 84.5% on CyberGym ahead of Anthropic Mythos 5 and OpenAI GPT-5.6 Sol, and scores 60 on the Artificial Analysis Intelligence Index tied with Kimi K3 for open-source #1. The only frontier open model trained entirely on Huawei Ascend.
GPT is OpenAI's LLM family, from 117M parameters in 2018 to the three-tier GPT-5.6 Sol/Terra/Luna lineup in 2026, serving 1B+ users and 2M enterprise customers. GPT-5.6 Sol leads LiveBench 81.1%, Terminal-Bench 2.1 88.8%, and Artificial Analysis Coding Agent Index 80 across multiple agentic benchmarks, while OpenAI's first open-weight model GPT-OSS ships under Apache 2.0.
Grok is xAI's LLM family: founded July 2023, opened with a 314B MoE under Apache 2.0 in March 2024, and two and a half years later spans Grok 4.6 (500K, $2/$6, four reasoning levels), Grok 4 Fast (2M), Imagine for image/video, and Grok Build for terminal coding — its moat is distribution (X / grok.com / Tesla / Bedrock), not single-model supremacy. This post traces Grok 1→4.6, sub-line positioning, pricing, and licensing traps.
Kimi is Moonshot AI's LLM family, born from ultra-long context. Kimi K3 (2026/07) is the world's first open 3T-class model—2.8T params, 104B active, 1M context, scoring 60 on the Artificial Analysis Intelligence Index tied with GLM-5.3 for open-source #1. Its Kimi Delta Attention brings a 2.5× scaling efficiency gain.
Llama is Meta's open-source LLM family, with the largest enterprise deployment footprint and the most mature ecosystem. Llama 4 Scout (10M context) and Maverick (17B active / 400B total MoE) are the current open multimodal benchmarks, but Meta pivoted to closed-source Muse Spark in April 2026—Llama 4 is likely the last major open Llama, and its license is not truly open (Llama 4 Community License, separate license required above 700M MAU).
Mistral is Europe's most successful AI startup, cutting through the market with a 'smaller, faster, cheaper' strategy and European data-sovereignty positioning. Mistral Large 3 is Europe's strongest commercial LLM, Small 4 is the 24B efficiency king, and Medium 3.5 is the open Modified-MIT model optimized for agentic coding. Its moat is not technical scale but the 'European compliance' card.
Qwen is the most-downloaded model family on HuggingFace, spanning sizes from 0.8B to 2.4T. In August 2026, Alibaba open-sourced a Max-tier flagship for the first time (Qwen3.8-2.4T-A95B) — but swapped the customary Apache 2.0 license for custom terms. Meanwhile the other new release, Qwen3.8-27B, runs native vision on laptop-class hardware and is the only one shipping under Apache 2.0. This post traces the family from 2023 through generation 3.8, explains how the open line and the commercial line split apart, and helps you pick the right model at each tier.
In 2026, AI models span seven major categories and more than 20 subcategories. This introduction to the AI Model Families series maps use cases to models and models to families, with current rankings and selection advice for each use case.
US-stock LLM agents have attracted nearly 100,000 GitHub stars, yet no Taiwan-stock project has even passed 10. I consolidated three side projects into a Taiwan-stock research agent where every conclusion must first survive a backtest; this article explains why.
Numbers are the easiest part of an LLM report to hallucinate. I therefore put every trusted number into a SHA-256-addressed evidence manifest and let the LLM cite only {{fact.id}} placeholders. If it writes a bare number, the entire output is discarded and replaced with a deterministic template.
Baseten puts custom-model packaging, GPU deployment, inference engines, autoscaling, and release workflows on one platform. Its value is not another OpenAI API, but retaining runtime control while operating less GPU orchestration.
Cerebras can dramatically accelerate generation on supported models, but agent latency still depends on prefill, tool I/O, model quality, and platform compatibility.
CS124 Winter 2026 opens by mapping a ten-week path from tokenization and classification to retrieval, speech, networks, and LLMs, while PA0 establishes the Jupyter environment used throughout the quarter.
Week 6 uses public neural-network slides for weighted sums, nonlinearities, loss, and backpropagation, then a public LLM/Transformer deck labeled 2025 for decoder-only architecture without treating it as the 2026 live transcript.
Week 7's public path is PA6a: implement causal self-attention, train a small Shakespeare Transformer, sample text, and compute perplexity; the live speech lecture remains an explicit source gap.
Week 9 builds movie recommendations with item-item collaborative filtering, then packages recommendation, web search, databases, and memory as agent tools under API-budget and team constraints.
Fall 2025 opens with computational thinking: reliability comes from decomposing retrieval, formal representation, verification, and generation into testable algorithms, not from one heroic prompt.
Lecture 3 does not turn its survey of modern LLMs into a single best recipe. It finds a conservative consensus—pre-norm, RMSNorm, no biases, SwiGLU, and RoPE—plus a small set of deviations justified by inference cost or stability.
Lecture 4 studies two kinds of sparsity: linear/recurrent attention reduces sequence-length cost, while MoE activates only part of a model for each token. Both turn saved FLOPs into routing, balancing, communication, and kernel problems.
Lecture 14 moves raw documents through language, quality, and safety filtering; exact and near deduplication; and source mixing. Each stage reshapes model behavior, while synthetic instruction and agent trajectories extend the pipeline into executable environments.
Lecture 13 traces training sources through Common Crawl, Wikipedia, GitHub, arXiv, books, and open datasets. Technically accessible is not the same as licensed, and raw data is not training data; provenance must precede cleaning and mixing.
CS336's first lecture does not treat building a language model from scratch as reenacting every old technique. It separates mechanics, mindset, and intuitions, then uses BPE to show how raw bytes become trainable tokens.
Lecture 2 reduces model training to tensors, FLOPs, bytes, and time: use einops to track dimensions, arithmetic intensity and roofline analysis to identify bottlenecks, then trade compute for memory with gradient accumulation and activation checkpointing.
Lecture 9 begins with log-log linear relationships between data and error, then uses scaling laws to compare architectures, optimizers, batches, and model-data allocations. The Chinchilla dispute shows how fitting methods, observed ranges, and deployment objectives change the answer.
Lecture 11 reads public recipes from MiniCPM, DeepSeek, Qwen, and Llama 3: hold most architectural ratios fixed, sweep learning rate and batch at small scale, then choose model/data allocation with IsoFLOPs. μP helps, but normalization, optimizers, and weight decay can break transfer.
Fireworks AI puts open-weight model evaluation, dedicated GPU deployments, and LoRA customization behind one API surface. Serverless fits low-volume starts, On-demand fits sustained traffic and custom models, while reserved capacity adds enterprise capacity guarantees.
LiteLLM is not a model provider. It is a Python SDK and self-hosted proxy that normalizes 100+ LLM APIs, then centralizes routing, fallbacks, virtual keys, budgets, and observability at the gateway layer.
Promptfoo combines prompts, providers, test cases, and assertions in YAML to produce repeatable local and CI evaluation matrices, with red teaming against the same targets. It lowers the testing barrier but does not remove output variance, LLM-judge bias, or hosted data-flow concerns.
Chapter 17 runs from next-token loss through Transformers, KV caches, MoE, and SFT, connecting an LLM's objective and architecture to its inference costs.
Together AI puts serverless APIs for open-weight models, dedicated GPU endpoints, batch inference, and fine-tuning on one platform, letting teams validate per token before moving to reserved deployment when traffic or customization justifies it.
Vertex AI is more than the Gemini API: it puts access to 200+ models, training, evaluation, deployment, and governance under one Google Cloud control plane. Since April 2026, its products and roadmap have moved into Gemini Enterprise Agent Platform, while the Vertex AI API, documentation paths, and many resource names remain in active use.
Units 15–18 place NLP models inside perception, reasoning, tool, and environment loops; the question shifts from next-token prediction to allocating inference compute and validating multi-step action.
CS288 moves from n-grams to RAG, reasoning, and agents through 18 public slide units and three assignments; Berkeley-only recordings make this an A3 materials route, not a public video course.
Units 08–12 turn a base model into an interactive system: pre-training establishes capability, post-training shapes behavior, and generation plus evaluation determine how outputs are used.
Units 13–14 connect models to external knowledge; A3 requires data collection, QA annotation, indexing, and ablations under CPU and latency constraints.
Chroma's controlled study shows that even when it fits, a full context degrades performance. Coding agent vendors have landed on seven different responses: compact, hand off, prune, defer loading, isolate, train it into the model, or change the unit of work. Amp removed /compact outright, Atlassian argues summarization should be a last resort, and Cursor's A/B test measured a 46.9% token reduction. The three real disagreements come down to what each team is measuring.
Crawl4AI handles retrieval after a URL is known: use JsonCssExtractionStrategy for stable DOMs, and switch to LLMExtractionStrategy only when extraction needs semantic judgment or must tolerate irregular layouts.
MIT 6.S191's 2026 edition publishes nine lecture videos, slides, three software labs, and solutions, making it an A3 self-study course. The supplied path still depends on Google/Colab, Comet, and OpenRouter for Lab 3, while unaffiliated learners do not receive MIT credit, project feedback, or API credits.
CS124 is the first course in Stanford's NLP branch. Its textbook is Jurafsky's own Speech and Language Processing, free online, and all nine assignment repos are public. But a banner sits on the course homepage: it will not be taught at all in AY 2026–27. And the chapter numbers the syllabus points at no longer match the August 2026 textbook.
CS224N has kept every course website since 2000 online. In Winter 2019, Transformers were lecture 14, taught by a guest. In Winter 2026 they are lecture 5, and every lecture after that assumes you already know them. The machine translation assignment is gone; assignment 3 now has you code a decoder-only Transformer from scratch, with pytest suites that run on your laptop.
CS224V only became Agentic AI in the 2026–2027 catalog, and the rename changed nothing underneath: the course still translates natural language into formal semantics and constrains agents with SMT solvers and knowledge graphs instead of wiring frameworks together. Seven of the eleven mandatory readings come out of the instructor's own lab. Every slide deck is public, and the course site says outright that they are deliberately incomplete.
CS25 is Stanford's 1-unit seminar where attendance is the only homework and anyone can audit. Of the nine talks in the Spring 2026 season, the three worth your time are Albert Gu on the inductive biases of SSMs vs Transformers, Charles Frye on serving inference across thousands of GPUs, and Victoria Lin on what native multimodality still hasn't solved.
Of the seventeen regular CS336 lectures, only nine are executable Python programs; the other eight are PDF slide decks — and the split falls exactly along the two instructors. Assignment 1's handout carries eight 'Low-Resource Tips' for finishing it on a laptop. Assignments 2 through 5 carry none. The course page lists the hourly price of a B200; the handouts list how many B200 hours each problem needs.
The AI SDK splits an AI message into a parts array — text, reasoning, source-url, tool-* — each an independent typed fragment (introduced in v5, unchanged since). That data structure dictates how modern AI conversation UIs are written: render by switching on part.type, handing each fragment to its component. This post unpacks the design logic of the parts model, useChat's streaming behavior, and how it became the foundation for component libraries like AI Elements.
LLM Application Design is the hottest new interview topic in 2025-2026. Key focus areas: RAG pipeline chunking/retrieval/reranking design, agent tool-use and planning loops, context window management strategies, guardrails and safety design, and LLM application evaluation methods. Interviewers especially value whether you've hit real-world pitfalls.
The dividing line in LLM interviews is whether you've actually used these things. High-frequency topics: BPE tokenization logic and multilingual challenges, pretraining objectives (CLM vs MLM), three levels of fine-tuning (full/LoRA/prompt tuning), RLHF workflow and failure modes, prompting as engineering practice, and the difficulty of LLM evaluation with current methods.
CS329A is built around the generation–verification gap: models can produce the right answer but can't tell which one it is. The conclusion the course draws about itself matters more — today's methods make models more consistent, not smarter. Nine lectures are public, out of twenty.
Stanford CS rests on CS103, CS107, CS109, CS111, and CS161; CS221 names three of those plus CS106B as preparation. This guide combines official prerequisites with an explicitly editorial reading order and marks public-material and offering risks.
Google PMLE, AWS AIF-C01 and AIP-C01, Microsoft AI-103 and AI-500, and NVIDIA NCP-GENL all test how to make a GenAI application fast, cheap, and reliable — and they form a three-rung ladder: AIF-C01 asks whether you know cost scales with tokens, AIP-C01 and the two Microsoft exams ask whether you can instrument and control it, NCP-GENL asks whether you can change the model and the hardware. Three different altitudes. NVIDIA works at the kernel and quantization layer (Model Optimization 17% + GPU Acceleration 14% = 31%, the heaviest single cost/latency block in the whole series), AWS and Microsoft at the application layer (three caching tiers, token caps, chargeback), and Google at the MLOps layer (CPU/GPU/TPU evaluation, data vs model parallelism, scaling serving backends by throughput). The shared core is eight levers, but each lever becomes a different question at each altitude. This post deliberately carries no prices and no hardware specs — that is the part of this topic that rots fastest.
Pure image understanding has flattened out — four frontier models all clear 80% on MMMU-Pro within 3 points of each other. The real differentiation is video, long-document OCR, and realtime speech, each with a different leader. But the most useful lesson from assembling these rankings is that two credible sources named different Video-MME leaders more than 10 points apart — and that July and August each turned the field over again.
NCA-GENL is usually what a job posting means by 'NVIDIA Generative AI / LLM certification.' But the official blueprint diverges sharply from the name — Core Machine Learning and AI Knowledge 30%, Software Development 24%, Experimentation 22%, Data Analysis 14%, Trustworthy AI 10% — with LLM and RAG content scattered at bullet level rather than forming a domain, alongside spaCy, NumPy, Keras, and cross validation. The other thing to know first: NVIDIA's official preparation courses all cost money ($30–$500), making it the only vendor in this series without a free official learning path. Official specs: $125, 1 hour, 50–60 items, two-year validity, English only, pass/fail with no score reported.
NCP-GENL is NVIDIA's professional-level LLM credential — $200, 120 minutes, 60–70 items. What separates it from every other GenAI exam is where the weight sits: Model Optimization 17% plus GPU Acceleration 14% is 31% on quantization, distillation, pruning, distributed parallelism, and CUDA profiling — not on calling APIs. Two things first: the Register button says Coming soon, so you cannot sit it yet; and two description cells in the official weight table are corrupted — Fine-Tuning is described with OpenUSD data-interchange text and Model Optimization with deployment text. I verified both verbatim; the correct descriptions are in the official PDF.
The Workers AI catalog currently holds 84 models. For general chat pick glm-4.7-flash ($0.06 / $0.40 per M, 131K context), for vision pick gemma-4-26b-a4b-it ($0.10 / $0.30, 256K), for cheap high-volume steps pick granite-4.0-h-micro ($0.017 / $0.112), and for embeddings pick qwen3-embedding-0.6b or bge-m3 (both $0.012 per M). This post is updated on a schedule.
Week 1 of CS146S is 'build Claude Code in 200 lines' plus a dissection of production system prompts. The agent loop really is that small. The course slides close with four things Claude does underneath, one of them being `<system-reminder>` tags scattered everywhere to stop the model drifting — which appears in no official documentation.
The third reason Go can't be learned with supervision is the interesting one: the ground truth itself is ill-defined — the strongest human doesn't play their best moves every day, and even their best move isn't optimal. The last 20 minutes map RLHF fully back onto RL: the agent is the model being fine-tuned, the action is the next token, an episode is one full generation, and the reward is extremely sparse.
CS230's first lecture is a course overview, but Andrew Ng spends most of it on three things: why scaling works, when prompting stops being enough, and why he thinks 'don't learn to code' is one of the worst pieces of career advice ever given.
Chroma tested 18 frontier models and all of them degrade as input grows — as a cliff, not a slope. Memory failures are usually retrieval failures in disguise. And the real cost of KV cache is bandwidth, not storage: every generated token reads the whole cache.
In November 2025 three frontier labs jointly broke all 12 previously proposed prompt-injection defenses. EchoLeak's payload passed Microsoft's own dedicated classifier. So the goal is not blocking every attack — it is surviving the ones that land, and that is harness work.
The line between workflow and agent is who decides the steps — the developer at design time, or the model at run time. By that definition most LLM systems in production today are workflows. Plus a usable test for choosing between RAG and an agent.
MCP governs agent-to-tool, A2A governs agent-to-agent, Skills govern reusable knowledge. The test is whether the data changes: if it changes between calls you need MCP; if it's stable enough to write down, a skill file is simpler and has no runtime that can fail on its own.
Microsoft, OpenAI, Salesforce, Stripe and three others independently say the same thing: reliability comes from the engineering around the model. And 'give the deterministic parts back to code' has been shipped as a product four separate times — Agent Script, Procedures, runtime, blueprints.
Firecrawl's open-source Rust conversion library turns 14 office formats (including legacy .doc / .ppt / .xls) into GFM at a 4.7ms median — 109× faster than Docling under the same timing basis. The trade-off: it does no OCR at all.
The most common mistake in feeding documents to an LLM isn't picking the wrong tool — it's picking the wrong layer. Structure already in the file goes to the conversion layer (milliseconds); text without structure goes to extraction; only inferred structure needs parsing. anydoc's 4.7ms against Docling's 513.6ms is a 109× gap, and most people jump straight to the most expensive layer.
The single most-cited meta-analysis on ChatGPT in education (g = 0.867, ~500k views) was retracted by Nature in April 2026. But the positive finding was not overturned — the issue is that it measures performance while the AI is available. Bastani's PNAS RCT measured something else: +48% accuracy during practice with GPT-4, then 17% below never-users once access was removed.
From MarkItDown (175k stars, MIT) to curl_cffi (6k stars), a survey of 34 open-source tools for feeding data to AI. Categorized along five axes: whole-site crawling, AI browser agents, document conversion, smart extraction, and anti-detection infrastructure. The key to selection isn't which tool is best — it's scenario matching.
From the CLI tool kin3o to the CVPR 2026 paper OmniLottie — a survey of open-source approaches for converting text and images into Lottie animations, with performance benchmarks and selection guidance.
MUSE-Autoskill (2026) introduces a five-stage skill lifecycle framework. Self-created skills achieve 60.35% (+7.16%) on SkillsBench overall, and an impressive 87.94% on tasks where skill generation succeeds — surpassing the human-authored skill ceiling. This post synthesizes six arXiv papers to map the full landscape of skill evolution research.
Even with temperature=0, LLM outputs can still fluctuate by up to 15% in practice. To rigorously compare agent changes, you need a frozen golden set, at least 3 runs per query averaged out, LLM-as-judge blind evaluation (pairwise preference flip rate reaches 35%), and paired statistical tests -- not just running each version once and going by feel.
The industry has converged on using OpenTelemetry GenAI semantic conventions to turn every LLM call and tool call into a span. Detecting the three major failure modes then splits into three tracks: faithfulness + semantic entropy for hallucinations, framework-level symbolic guardrails for tool misuse, and max steps + action hash deduplication for infinite loops — all wired into a Final / Trajectory / Single-step three-layer evaluation framework.
Agent decision-making under resource constraints is bounded rationality reborn: Rational Metareasoning uses VOC rewards to save 20-37% of tokens, BATS proves that adding budget without budget awareness is futile, FrugalGPT cascades cut costs by up to 98%, and Speculative Actions reduce latency by 20%. The three constraints ultimately converge into a single Pareto curve, and the overarching trend is moving from humans tuning knobs to models making resource-rational decisions on their own.
Three seemingly distinct agent security problems — tool output injection, trust boundaries, malicious agents — share the same root cause: LLMs flatten instructions and data into a single token stream, making them architecturally unable to distinguish between the two. Understand this through-line and you can trace every attack from EchoLeak (CVE-2025-32711, zero-click) to the Morris II AI worm, and see why 'making the model behave' doesn't work — only architectural constraints (six design patterns, CaMeL) do.
Traditional RAG is a fixed pipeline of 'retrieve then answer.' Agentic RAG splits retrieval into three decision layers: when to retrieve (FLARE uses token probabilities; Adaptive-RAG uses a complexity classifier), what to retrieve (HyDE / RAG-Fusion / decomposition / Step-back), and how to fuse (RRF k=60 then cross-encoder rerank then compression -- Anthropic measured a -67% failure rate reduction). Key counter-intuitive insight: unnecessary retrieval hurts quality -- 'deciding not to retrieve' is a first-class capability.
Automatic prompt optimization (APO) has evolved from APE/OPRO to GEPA: replacing sparse rewards with linguistic reflection, winning over GRPO by ~6pp with 4-35x fewer rollouts. Meanwhile, tool descriptions are the overlooked prompt -- small wording changes can shift tool selection rates by 10x, and Anthropic's experiments show Claude self-rewriting tool descriptions outperforms human experts. These two lines are converging: eval-driven automatic optimization is eating hand-tuned prompts.
Inferring another's beliefs/goals/intentions from observed behavior is called Machine Theory of Mind. Three lineages: symbolic BDI, Bayesian inverse planning, and deep learning ToMnet. The biggest controversy in the LLM era is that GPT-4 still trails humans by >10 points on ToMBench — are high scores genuine reasoning or statistical shortcuts?
At 99% accuracy per step over 100 steps, the error-free completion rate drops to just 36% -- error compounding is a structural problem, not something prompt tuning can fix. Distributed systems' supervisor trees, bulkheads, circuit breakers, sagas, and durable execution can be mapped almost one-to-one into agent orchestration. But LLMs introduce a failure class that traditional systems never had -- semantic errors that don't crash -- which require Inspector agents (recovering 96.4%) and redundancy voting (MAKER: one million steps with zero errors) to address.
Cosine similarity and relevance systematically diverge across an entire class of scenarios: negation (most IR models score at or below random on NevIR), exact identifiers, numeric thresholds, and logical combinations (SoTA models achieve recall@100 < 20 on LIMIT) -- some of these hit the theoretical ceiling of the single-vector paradigm, and switching to a larger model will not help. Recommended remedy order: hybrid BM25 -> reranker (Anthropic measured -67%) -> upstream metadata routing -> domain fine-tuning / multi-vector.
As tools scale up, selection accuracy doesn't degrade gracefully — it collapses: 4 to 51 tools drops from 43% to 2%, 10 to 100+ drops from 78% to 13.62%. The root fix is to stop stuffing everything in at once — Anthropic's Tool Search Tool uses defer loading plus retrieval to cut 85% of tokens, pushing Opus 4.5 accuracy from 79.5% to 88.1%. Description quality has conditional payoff: negligible in simple scenarios, but correctness jumps from 44% to 50% in multi-tool chaining.
Traditional Chinese RAG retrieval failures are a three-layer stack: embedding granularity defects (BGE/GTE from 0.1B to 7B all mis-rank on simple queries like 'fried chicken'), Simplified Chinese / English corpus dominance causing local vocabulary drift ('premium', 'exclusion clause' alignment is unreliable), and MTEB Chinese benchmarks being Simplified Chinese making model selection signals misleading. The fix is architectural: OpenCC normalization -> hybrid + jieba segmentation -> reranker -> local fine-tuning last -- and the prerequisite for all of it is building a Traditional Chinese eval set first.
arXiv does not perform peer review, and roughly 2% of submissions are rejected. Quality judgment relies on external signals: top venue acceptance > institution + open-source reproduction > citation quality. Includes a 20-item practical checklist and a 2026 toolbox (PWC has shut down).
The hard part of LLM agents is not building function calling, skills, code interpreter, and document tools individually -- it is assembling them into a system that selects the right tool, writes code when needed, decomposes tasks, verifies results, and resists prompt injection. This post organizes the key papers into six engineering decisions: function calling reliability, tool/skill selection, code-as-action, multi-step planning, skill systems, and safety plus document generation.
Reading papers is two problems stacked together: methodology (Keshav's three-pass method, 5-10 min / 1 hour / 4-5 hours) determines how to read, and tools (arXiv HTML, alphaXiv, NotebookLM, Connected Papers, Zotero) shorten the time for each pass. AI lowers the barrier to understanding; judging correctness always stays with the human.
Rewriting tool descriptions from soft suggestions to hard rules (whitelist + consequence explanation) eliminated the LLM's incorrect tool selection; adding skip_signal=True fixed vector store double-indexing.
For side projects, toy demos, and RAG prototypes, nobody wants to swipe a credit card on day one. This is a verified roundup of 40+ LLM inference providers still operating as of 2026/05, tiered by whether free resources auto-replenish or are one-time grants. Each entry notes credit-card requirements, supported models, paid starting prices, and catches. Chinese-origin providers including Zhipu GLM (permanently free), Doubao (2M tokens/day), Kimi, DashScope, and the Ollama local option are all included.
PageIndex skips chunking, embedding, and vector storage entirely. Instead it relies on LLM reasoning over a tree-structured table of contents the LLM itself wrote, reporting 98.7% on FinanceBench in its own vendor-run evaluation. It solves a different problem than vector RAG — finding the right section in a well-structured long document.
Groq Console is the developer portal for Groq's in-house LPU chip, offering an OpenAI-compatible API, Playground, and free tier credits. Its selling point is running open-source models like Llama, Qwen, and DeepSeek at the fastest tokens/second on the market.
For running Traditional Chinese LLM workloads on Cloudflare Workers AI, the Gemma family follows instructions more reliably than same-tier Llama models. gemma-3-12b-it was marked deprecated on 2026-05-30; the current equivalent is gemma-4-26b-a4b-it: 256K context, Vision, Function calling, at $0.10 / $0.30 per M tokens.
AEO/GEO tools aren't a single category — they span three distinct layers: the input layer (is your website ready for AI to read), the traffic layer (how much are AI bots actually crawling), and the output layer (how is your brand mentioned in AI answers). This post maps out all three layers, from open-source self-hosted options to commercial SaaS.
Encyclopedia of Agentic Coding Patterns catalogues 190 patterns to help you make the right software decisions in the age of AI-written code — and the book itself is autonomously written and maintained by an AI agent.
Autoreason replaces the traditional critique-and-revise loop with a competitive multi-version evaluation mechanism (A/B/AB + blind Borda count), solving three structural problems in LLM self-refinement: prompt bias, scope creep, and lack of restraint.
Comparing the NVIDIA DGX Spark, Apple Mac Studio M4 Ultra, ASUS Ascent GX10, MSI AI Edge, and more — helping you find the right local inference hardware.
The NVIDIA DGX Spark is powered by the GB10 Grace Blackwell Superchip, 128 GB of unified memory, and delivers 1 petaFLOP of FP4 compute — starting at around $3,999 USD. It lets developers run 200B-parameter models locally and fine-tune 70B models, making it the most accessible NVIDIA AI development platform available today.
2026 Q1 saw a full-blown open-source model explosion: on the LLM front, GLM-5, Kimi K2.5, and Qwen3.5 caught up with closed-source models; Embedding and Reranker are dominated by Qwen3 and BGE; speech has Voxtral TTS and Whisper V3; image has FLUX.2; and video has Wan 2.2 rivaling Sora. This is the complete navigation map.
OpenClaw's failover runs in two stages: rotate auth profiles within the provider, then fall back to another model. But what really governs behavior is who chose the model — a model you picked yourself with /model is strict, and its failure is reported rather than answered by some other model.
OpenClaw's hard requirement for a model is tool use plus a large enough context — onboarding only auto-suggests a local model when it confirms tool support and at least a 16K context window. The easier thing to get wrong is that provider, model, and agent runtime are three separate layers: an `openai/*` ref does not mean Codex.
Langfuse is currently the most mature open-source LLM Observability platform. This post covers four core capabilities — Tracing, Prompt Management, Evaluation, and Datasets — showing you how to use them in real projects.
An AI agent is not a black box — it is built from three layers: what it knows (Context), how it thinks (Cognition), and what it can do (Action). Understanding these three layers is the key to grasping why agents are sometimes brilliant and sometimes go off the rails, and how to design a truly effective agent system.
Ollama wraps llama.cpp in a Docker-style CLI + REST API, letting you run LLMs locally with a single command. This post covers core concepts, installation, API, hardware requirements, Modelfile customization, and what this tool is — and isn't — good for.
Good prompts aren't written in one go — they're iterated into existence. Start with the simplest prompt, test with real cases, classify error types, and make targeted fixes. This article covers the three-part System Prompt structure, reasoning framework selection, few-shot optimization, token budget management, and six common mistakes.
The attacks RAG systems face go beyond the technical level — Prompt Injection and Jailbreak are real threats. Both inputs and outputs need independent protection layers.
Search found the right documents, but the LLM's answers are still poor — often the problem lies in prompt design. System prompt structure, context formatting, and instruction placement all affect output quality.
RAG and Fine-tuning solve different problems. RAG gives the model new knowledge; Fine-tuning changes the model's behavior and style. In most cases you use both, not pick one.
A dynamically composable RAG pipeline built on Cloudflare Workers AI (gemma-3-12b-it + bge-m3): 14 base steps + 6 LangGraph-specific nodes, with three strategy graphs (Baseline / Agentic / Plan-Execute) selected at runtime.