Page 109 of 129
Andrew Ng demonstrates error analysis on a deep researcher: columns are the pipeline stages, rows are 10 to 100 queries, you only look at the ones that went badly, and you mark each cell where something broke. The percentages don't have to sum to 100%. He says it takes three or four hours and saves weeks of going the wrong direction — and the fraction of people who actually do it is far below 100%.
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 15 minutes or so 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.
Andrew Ng walks a face-recognition door system through the entire project lifecycle, and the whole lecture has one thesis: speed. He gives teams a two-day deadline, on the reasoning that 'time spent preparing data should be commensurate with the time it takes to train the model once.' It closes on a line: my job is to build something that actually works, and that is not the same as building something that works on the test set.
CS230's second lecture derives embeddings through three case studies: day/night classification teaches you to use humans as a proxy for choosing resolution, trigger-word detection teaches you to manufacture a million training examples in three hours, and face verification walks you through designing your first loss function. The final step — from supervised triplets to self-supervised pairs — is why modern models can consume billions of unlabeled images.
Ask a model what a goose looks like to it and it draws a whole flock — because the labeled data tagged a flock as 'goose,' so it thinks the flock is the label. This lecture gives seven ways to open a CNN up, then says honestly: applied to transformers, even the frontier of this research only explains two layers.
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.
I tested 10 open-source PDF parsing tools on four scanned NTU graduate entrance exams. VLM-based tools—Firecrawl, MinerU 3.4, and Marker v2—overwhelmingly beat conventional OCR on formulas and code, but installation was the real barrier: MinerU's old package name creates dependency hell, Marker's first model download takes 10 minutes, and PaddleOCR needs a separate engine. In practice, use RapidOCR for screening and MinerU or Firecrawl for close inspection.
An elite-level engineer applied to over 300 jobs and failed every one, because the interview guides told him to stand his ground and have a backbone, and he read that as being hard-nosed. The interviewers' read: this is the 10x engineer, and I don't want him anywhere near my team. This lecture also gives the best criterion anyone has offered for vibe coding — technical debt.
The 2026-08-13 release added 363,246 lines and published the For You ranking weights for the first time: favorite 0.5, reply 5.0, report −234.0. But the weights are constants — the P(action) that actually decides order comes from a 2560-dim, 8-layer transformer.
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.
Salesforce's number from 20,000 deployments: 90% of the work on an agent happens after launch, the reverse of traditional software. Stripe merges 1,300 PRs a week with no human-written code, and credits the environment rather than the model.
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.
Standard RAG gives a wrong answer when it retrieves the wrong chunk, and nothing in the system will notice. Agentic RAG adds a self-check, at the cost of the evaluator paradox: the ceiling on self-correction is whatever the evaluating LLM can judge about relevance.
Three Legislative Yuan budget evaluations log 23 drone-caused airport closures in Taiwan between FY2019 and August 2023, and every row's response column repeats one sentence: on notification, go to the scene with the Aviation Police and investigate. Article 99-13(6) of the Civil Aviation Act gives an airport the verb 'enforce against', not 'stop or remove' — yet the FY2024 report says every airport has already procured handheld jammers.
Paragraphs 3 and 4 of Article 99-13 of the Civil Aviation Act say 'government agencies, schools or legal persons', but paragraph 7's proviso — the only sentence letting a site operator 'take appropriate measures to stop or remove' a drone — says only 'government agencies', and Taipower, CPC and the fabs are legal persons. Their remaining path is to have a municipality announce a restricted zone and 'enforce against' violators, which drops the penalty from an NT$300,000 minimum to an NT$300,000 maximum, while importing a jammer requires 'critical infrastructure provider' status that you learn you hold by receiving a letter.
Eight municipal police departments now run drone units, and the authority to fly is clear — Article 99-16(2) of the Civil Aviation Act exempts agencies performing statutory duties. But that is a flight-safety exemption: a Taipei City legal opinion says it 'may be unable to serve as the legal basis for conducting administrative investigation by drone', and the special-compulsory-measures chapter passed in July 2024 authorises GPS, IMSI-catchers and private-space imaging with no aerial-investigation provision at all.
'Drone' appears in 19 of Taiwan's central statutes and regulations; 'counter-drone' appears in exactly one, and that one is a procurement act. Six settings are authorised to 'stop or remove' an intruding drone and none of them says by what means — Article 99-13 of the Civil Aviation Act says only 'take appropriate measures', while the 31-item police instrument list includes mortars but no jammer, and Article 67(1) of the Telecommunications Management Act has no public-duty exception.