한국어EN

Part ReviewPart 6 · What Comes NextDAY 56–60

Part 6 Review

What comes next, one layer deeper · DAY 56-60

Made by AIThis part review was written by AI (Claude) from that part's lessons and records, and published after automated checks. The owner doesn't read every item.How it's made

Part 6 spent five days on what comes next: the AGI debate, jobs, safety, robots, and the order of learning.

Intermediate digs one layer deeper into the same facts, down to why things turn out this way. No quizzes and no experiments — it's a read-and-move-on piece.

DAY 56 · The AGI debate

The AGI debate isn't about skill, it's about where the bar goes

AGI is the name for an AI that's broadly as good as a person, and nobody ever settled how much it has to do to count. Set the bar low — 'if you can talk with it like a person, it's AGI' — and it already passes. Turing's 1950 test was close to that. Set the bar high — 'if it learns a brand-new task on its own and finishes it, it's AGI' — and it's not there yet.

Measure with tests and the answer splits. In February 2026 a model scored 84.6% on a puzzle test where people average 60%. On a test where you work out new rules with no instructions, at its March 2026 launch it was people 100% vs AI 0.51%. The argument is loud because people use the same word for different things. Some say drop 'AGI' entirely.

DAY 57 · AI and jobs

What AI takes isn't jobs, it's pieces

Split one job and you get four pieces: gather, make, choose, own it. AI grabbed the first two fast. Those are the ones with a fixed shape of right answer, done over and over. The last two stay. They're about deciding what's right and saying this is fine to send out.

AI is amazing at a single piece but still weak at carrying a job from start to finish alone. In a 2025 evaluation that handed over whole real jobs, the best AI finished 2.5% of them. So the weight slides back, toward checking and taking responsibility, and brand-new job titles show up.

DAY 58 · AI safety

Safety isn't the model's personality, it's guards on top

AI safety isn't magic that makes a model good. It's guards you add on top of the model, one at a time. There are three big ones — refusing, asking before acting, and checking afterward. Refusing doesn't appear on its own. A model is trained to tell you what it knows, so saying no has to be taught separately.

In one 2025 study the success rate of trick questions that break the rules fell from 86% to 4.4%. In exchange it took about 24% more computing. And you get 'over-refusal,' where perfectly fine requests get turned down. So these days people measure refusing well and over-refusal together. A better question than 'is this AI safe?' is 'what guards are on it?'

DAY 59 · Robots and AI

AI went into the robot's first three pieces

Old robots only moved along coordinates you handed them. Shift an object by a centimeter and they grabbed thin air. From 2023 the AI that sees and the AI that understands instructions got wired straight into robot bodies, so 'pick up the red cup' started working. Three pieces — eyes, ears for instructions, judging — turned into AI. Hands and legs are still mechanical engineering.

Balance has to be caught hundreds to thousands of times a second, and thinking can't run that fast. So robots today split fast body circuits and slow thinking circuits into layers and run them separately. One robot shown in 2026 said it corrects whole-body balance 1,000 times a second and joints 200 times.

DAY 60 · The order of learning

What you need isn't new knowledge, it's a four-step order

Across all 60 days one answer kept coming back — ask, check, try it yourself, then explain it in your own words. Meeting a new tool, getting an answer from AI, teaching yourself something new — the order is the same every time. Getting stuck isn't failure, it's one step of the order. Asking again about the stuck part is the next step.

This is exactly where the pieces AI takes and the pieces left for you split. AI gives you the draft, you decide if it's right. Tools change every few months but the way you learn doesn't. So what you pick up today still works next year. Two people can use the same AI and end up in different places, depending on whether they ran the loop. Not smarter — just one lap further.

Part 6 in one line

Squeeze five days into one line and it's this — questions about what comes next are mostly about us, not about AI. Where to set the bar, which pieces to hand over, what guards to put on.

That's the end of the 60-day course. From next week on, one lesson a week built from the records added that week opens.

Learn more

  1. Computing Machinery and Intelligence (Turing, 1950) — the original paper that proposed judging 'can a machine think' by the answers it givesacademic.oup.com/mind/article/LIX/236/433/986238
  2. On the Measure of Intelligence (Chollet, 2019) — the proposal to measure intelligence as how fast you pick up new skills, not how much you knowarxiv.org/abs/1911.01547
  3. ARC-AGI-3 launch (ARC Prize, 2026) — a test where you work out new rules with no instructions. At launch: people 100% vs AI 0.51%arcprize.org/blog/arc-agi-3-launch
  4. Gemini 3 Deep Think (Google, 2026) — the company-reported 84.6% on a puzzle test where people average 60%blog.google/innovation-and-ai/models-and-research/g…
  5. GDPval (OpenAI, 2025) — scoring AI on real work products from 44 occupations across nine fieldsopenai.com/index/gdpval
  6. Remote Labor Index (Scale AI & CAIS, 2025) — the evaluation where the best AI finished just 2.5% of whole real jobsarxiv.org/abs/2510.26787
  7. Canaries in the Coal Mine (Stanford Digital Economy Lab, 2025) — an observation about job counts for 22-25 year olds; the authors stress it is not a causedigitaleconomy.stanford.edu/publication/canaries-in…
  8. Constitutional AI (Anthropic, 2022) — using a written set of principles as the training signal instead of labelling by handarxiv.org/abs/2212.08073
  9. Constitutional Classifiers (Anthropic, 2025) — the source of the 86% to 4.4% numbers and the 23.7% extra computinganthropic.com/news/constitutional-classifiers
  10. Petri (Anthropic, 2025) — an open tool that has AI audit AI automaticallyanthropic.com/research/petri-open-source-auditing
  11. RT-2 (Google DeepMind, 2023) — wiring knowledge learned from internet images and text straight into robot arm controlarxiv.org/abs/2307.15818
  12. Gemini Robotics (Google DeepMind, 2025) — a model that connects a large language-understanding AI to a robot bodydeepmind.google/discover/blog/gemini-robotics-bring…
  13. Helix 02 (Figure, 2026) — the explanation of running circuits at different speeds: balance 1,000 times a second, joints 200figure.ai/news/helix-02
  14. ARC Prize 2025 results analysis (ARC Prize, 2025) — published results from a puzzle contest people find easy and AI does notarcprize.org/blog/arc-prize-2025-results-analysis
  15. Will we run out of data? (Epoch AI, 2024) — working out in numbers when AI might run out of text to learn fromepoch.ai/blog/will-we-run-out-of-data-limits-of-llm…