Reading Discussions Become a Newsletter

An era of assessing process alongside results

Reading Discussions Become a Newsletter

Hallym University Media School Colloquium · 2026.06.10

Reading discussions become
a newsletter

Writing becomes media:
designing major courses in the AI era

Presented by · Achmage (안창현) (Achmage)

Source · Slide 1

Today's question

In the AI era,
what are we
being assessed on?

The process behind the outcome — and where that process is recorded.

Source · Slide 2

Introduction

Even a reporter with 20 years of experience
struggled with
the AI exam

SK AX AI Competency Certification — KBS News 9 (2026.06.01)

Source · Slide 3

Introduction

SK's AI competency certification exam appeared on the nine o'clock news

KBS News 9 (2026.06.01) · The AI exam shaping recruitment and promotion: we took it ourselves

Source · Slide 4

Introduction

Nine years of experience: A; twenty years: B — even experienced KBS reporters struggled

Reporter Bang Jun-won

• Joined KBS in 2018 (45th intake) · nine years of experience: A

Reporter Kim Jun-beom

• Joined KBS in 2007 (33rd intake) · twenty years of experience: B

≠

The highest grade is S · Sources: KBS News 9 (2026.06.01); NamuWiki, KBS NEWS staff

Source · Slide 5

Introduction

Two exam stages: literacy, then bootcamp

STEP 1 · AI literacy

• From understanding generative AI to prompting and everyday applications: assessing literacy and execution in real work contexts. STEP 2 · AI bootcamp

Practice building RAG systems and AI web/apps, then design and develop an AI agent and receive certification through a technical assessment.

AI use: the process itself is assessed — KBS / SK AX

Source · Slide 6

Introduction

You can solve it without knowing how to code.
The assessment focuses on how well you
structure the problem and approach it
step by step.

— Yoon Seong-hwan, SK AX AI testing business team lead (translated quotation)

The exam examines the problem-solving process, not just the answer.

Source · Slide 7

In one sentence

Companies have begun assessing
the process of using AI
alongside the finished result.

Source · Slide 8

Introduction

New broadcast producers may soon face the same: three abilities assessed together

Even non-developers can solve problems through vibe coding — KBS News 9

01

Writing and planning

The ability to design relationships with readers.

02

Structuring work with AI

Breaking a problem down and directing AI step by step.

03

Executing through vibe coding

Demonstrating a dashboard or fact-checking app you have built.

Source · Slide 9

Academia and research

AI use and judgment in academic work

Academia is already looking at
the process

Nature · Frontiers · Wiley surveys (2025–2026)

Source · Slide 10

Academia and research

Of 4,946 researchers worldwide, only 1,043 had actually used AI in research

1,043

— had actually used AI in research (21%)

Even among reported uses, editing was 28% and drafting was below 8%
(Nature survey, as of February 2025)

Have you written a paper with AI? Most respondents had not, or did not intend to . Nature and Wiley surveys

Sources: Wiley, worldwide researcher AI-use survey (2025.02.04) · Nature

Source · Slide 11

Academia and research

For 66%, the barrier was not ability but a lack of training or guidance

44% needed AI training

22% did not know where to start

And most expected it
to become commonplace in manuscripts, grants, and peer review soon.

Most uses of AI were expected to become accepted within a few years — Nature

Source · Slide 12

Academia and research

What is acceptable? Editing: 90%; translation: 91%

Editing — human-written drafts refined by AI: 90% acceptance

Translation — AI used for a complete translation: 91% acceptance

However, 42% wanted even the translation prompts disclosed. Pressure to disclose the process has already begun.

Source: Nature survey (2025.05.14), around 5,000 respondents · Dr Bloggs scenario

Source · Slide 13

Academia and research

Yet a majority said AI should never be used for Methods, Results, or Discussion

The closer a section is to reporting results, Never appropriate the higher the rejection rate — Nature

Abstracts and introductions receive more tolerance.
AI supports productivity; judgment remains human.

Source · Slide 14

A key finding

People who had used AI
were more positive about its use.

“respondents’ self-reported experience with AI for writing or reviewing papers did correlate strongly with having favourable opinions of the scenarios.”

— Nature survey analysis (2025.05.14)

Attitudes differ through experience, beyond simple support or opposition — and experience accumulates through records.

Source · Slide 15

Hao et al. 2026 (Nature)

Papers that used AI as a tool were far ahead

3.02×

Number of papers published

4.84×

Citation count

−1.37 years

Time to promotion as a team leader

Analysis of 41.29 million papers, 1980–2025 · Natural-science papers using AI as a tool · Hao et al., Nature (January 2026)

Source · Slide 16

A paradox

Yet instead of opening new topics, AI encouraged
already popular topics
to be repeated.

Even new papers end up citing the first original: dissemination accelerates while thinking narrows.

Source · Slide 17

The key point

The question was not simply whether AI was used,
but whether researchers had their own original data.

Source · Slide 18

Peer review

A reversal in just seven months

May 2025 · Nature

52% strongly opposed AI in peer review

December 2025 · Frontiers

55%

Already using AI in peer review (around 1,600 respondents worldwide)

Seven months

Source · Slide 19

Frontiers 2025: overall recommendations

Six academic recommendations, all pointing toward process

01

Transparency and responsibility

Disclose when, where, and how AI was used

02

AI literacy education

Structured education and accredited programs

03

Ethics and integrity

Human accountability, bias testing, and reproducibility audits

04

Equity and access

Ensure equitable access to AI skills

05

Community participation

Develop policies together and establish feedback mechanisms

06

Policy leadership

Contribute to global standards

Source · Slide 20

One recurring word

Validate the process
Teach the process
Assess the process

Source · Slide 21

Then

where do you store
that process?

In a GPT chat? Context resets whenever you open a new conversation.
In your head? You must reconstruct it from scratch whenever you need it.

Source · Slide 22

Journalism education

The human role that journalism demands

Journalism education
Where is it heading?

ACEJMC · Columbia · USC · Columbia Journalism Review

Source · Slide 23

ACEJMC (US journalism accreditation)

Core graduate competencies: the same four abilities

Find

Use appropriate tools to present images and information effectively

Evaluate

Apply critical thinking to research and evaluate information

Verify

Critically check accuracy, fairness, clarity, and basic numerical and statistical information in the work

Express

Write for the intended audience and purpose accurately and clearly

Source · Slide 24

Yet

all these competencies
depend on a habit
of keeping records.

Source · Slide 25

Columbia Journalism School (CJS 2030)

Learn. Build. Critique.

It emphasizes finding, collecting, and analyzing material, and explicitly calls on
individuals to build their own digital archives.

Data journalism is a working environment that retains material, sources, analysis, and judgment, beyond one-off prompts.

Source · Slide 26

USC AIMS × Columbia Journalism Review

Eighteen journalists were asked:
How do you use AI?

Source · Slide 27

Journalists' accounts · 01 / 04

AI is for productivity,
not for writing.

“As much as I believe that AI helps my productivity, I refuse to use it for writing.”

— Emilia David · VentureBeat

Hand writing over to AI and you lose both your relationship with readers and the skill itself.

Source · Slide 28

Journalists' accounts · 02 / 04

On its own, AI is
only a parlor trick.

“AI on its own is a parlor trick … useful when paired with properly structured data and someone who knows what they’re doing.”

— Zach Seward, head of AI at The New York Times

AI needs structured data and domain knowledge to work: the reason for personal knowledge management.

Source · Slide 29

Journalists' accounts · 03 / 04

Treat AI as hands,
not as a brain.

“I don’t fear AI if it’s used as a hand, not a brain.”

— Susie Cagle

The judging brain is human. Delegating it to AI may be faster, but becomes less human and less truthful.

Source · Slide 30

Journalists' accounts · 04 / 04

A smart assistant that can also produce nonsense:
I ask it to verify only after uploading my own draft and interview transcripts.

“I’ll upload sections of my book, along with interview transcripts … I would never ask it to write anything, though.”

— Nicholas Thompson · The Atlantic, CEO

AI begins to work when humans provide primary material and records.

Source · Slide 31

A shared structure

Journalists who use AI effectively follow the same pattern

01 · Human

Create primary material

• Create original data through reporting, recording, and writing: ▶

02 · AI

Assist

• Organize, verify, summarize, and translate (AI as hands): ▶

03 · Human

Make final judgments and publish

• If AI is the only source, do not publish: Ben Welsh(Reuters)·Sarah Cahlan(WaPo)·Khari Johnson(CalMatters) — CJR “How We’re Using AI”(2025.05.12)

Source · Slide 32

Conclusion

The better journalists use AI,
the more systematically they accumulate
their own primary material. = PKM

Without structured personal knowledge, AI remains a parlor trick.

Source · Slide 33

Northwestern University

Repeatable action and records

Agentic Investigation Challenge — an ongoing challenge at the time of the presentation

$5,000 prize · Deadline July 15 · Judging August 15 — an AI-assisted investigative journalism competition

Submit more than a single AI-assisted story:
submit the full record of how you used AI.

Source · Slide 34

What to submit

Four deliverables: documenting process alongside results

①

Agent Skills

Reusable AI workflow (directory)

②

Findings (PDF)

A findings report incorporating external material

③

Interaction Traces (JSON)

Complete logs of inputs, tool calls, outputs, and interventions

④

README.md

A Markdown index connecting the entire structure

Source: Agentic AI Investigative Journalism Challenge, Northwestern (ongoing in June 2026)

Source · Slide 35

① Agent Skills

A repeatable procedure that produces the same result
in nine out of ten runs

Knowledge assets that are structured and stored, beyond disposable prompts.

Source · Slide 36

③ Interaction Traces

Inputs, tool calls, outputs,
and the moments when I intervened to correct them:
the complete logs.

A workflow that disappears when the chat closes cannot do this. A separate system for storing and managing records is needed.

Source · Slide 37

The decisive point

Without files,
you cannot even submit.

GPT web chat → traces disappear ✗

Local records (PKM) → ready to submit ✓

Source · Slide 38

Records

A structure for records that retain the process

Ultimately, it comes back
to records

To assess a process, that process must be retained somewhere.

Source · Slide 39

Putting it together

Three paths point to one destination

Companies

SK AX

Assess the process of structuring and solving problems alongside outcomes

Academia

Nature · Frontiers

Move beyond whether AI was used toward disclosure of prompts and workflows in practice

Journalism

Northwestern

Require records and verification of the entire reporting and working process throughout the work

One common thread: attention to process.

Source · Slide 40

The evolution of AI use

Five stages of AI: from prompts to harnesses

Prompts → CoT → MCP → Skills → Harnesses & AI OS

Another account of the same trend: convergence on AI that knows me

Source: The Better Newsletter, issue 220

Source · Slide 41

National standard

The national AI strategy committee adopted Markdown as a standard document format

National AI Strategy Committee press release (2026.03.05)

Markdown (.MD)
A document format that uses simple symbols for headings, paragraphs, and contents instead of complex formatting, making it easy for both humans and AI to read.

The stated concern is that extensive formatting in existing HWP documents makes their structure difficult for AI to recognize.

Source · Slide 42

The form of a record

The same file: clean to human eyes, broken for AI

Human view (Hangul .hwp)

AI view (opened in a text editor)

=

Source · Slide 43

The critical issue

The issue is the form of the record, not simply whether one exists

Common objections	The underlying problem
But I organized everything in Hangul or Word.	A closed container: AI cannot read it
But I wrote it in a notebook with a pen.	It cannot be searched, reproduced, or connected
But it is all there when I log in to GPT.	Separate chat windows do not connect with one another

Having a record and having a record that AI can read are very different things.

Source · Slide 44

In the classroom: Odyssey Seminar 3, spring 2026

I interviewed 16 students.
All said they used AI
and kept records.

Source · Slide 45

Three shared findings from the interviews

I know, but I do not act

01

Little more than temporary storage

Notebooks, text editors, and messages to oneself in KakaoTalk. Experience exists, but there is no process for converting it into knowledge.

02

Existing only in one's head

There is plenty of observation and reporting, but the process is not retained:
portfolios are reconstructed from memory.

03

Awareness ≠ Action

Everyone agreed on the need.
They simply did not know where or how to begin.

Source · Slide 46

The key point

Memory fades,
but records accumulate:
this is a question of design, not willpower.

Source · Slide 47

TL;DR · Markdown is all you need

The problem is not a lack of records.
The records are not in Markdown,
so AI cannot read them.

Source · Slide 48

How to store it

A workspace separating collection, understanding, and action

How should .md files
be stored?

CMDS LLM Wiki · CMDS Vault · Achmage OS — a three-part architecture

Source · Slide 49

A three-part architecture

The key is separating collection, understanding, and action

Collection

CMDS LLM Wiki

• **AI automatically collects quality material, reducing the effort of searching:** understanding

CMDS Vault

• Only what I have read, understood, and written myself (no AI): Execution

Achmage OS

A system that shortens and standardizes repeated production.

Manage collection, understanding, and execution separately

Source · Slide 50

What the three spaces do

A place to collect, a place to understand, a place to create

Three vaults, simply: store / collect / execute

• Collection ≠ Understanding ≠ Execution: distinct roles LLM Wiki collects; Vault supports understanding; Achmage OS produces.

Source · Slide 51

A daily workflow

Collect material → Judge understanding → Accumulate knowledge → Act

Separate collection, understanding, and execution to keep knowledge organized and AI execution stable

Source · Slide 52

A student's personal newsroom

Research desk · Editorial and reading notes · Production desk

Spend the most time on editorial and reading notes (Vault)

LLM Wiki

Collect and supply material to the editor

Vault

Read, review, and digest: the largest time investment

Achmage OS

Less repetitive production, more attention to content

Source · Slide 53

The key point

Beyond tools, ask:
What knowledge am I collecting,
for what purpose, and why?

Source · Slide 54

Ten principles

Ten principles for sustained reading and writing

Beyond tools:
principles of thinking
Keeping records is a practice

Ten commandments: principles for people who keep records

Source · Slide 55

Keeping records is a practice · 1–5

1

Write even a single line yourself: one line of my own over a thousand written by AI.

2

Even outside your main interests, save what you read closely for more than ten seconds .

3

Think in terms of #hashtags rather than folder classifications.

4

Prioritize the density of notes over their number — 2,400 AI notes < 24 notes I wrote myself.

5

Use your own words, clearly enough to explain in ten seconds in an elevator.

Source · Slide 56

Keeping records is a practice · 6–10

6

There is no later — take the note now and move it into Markdown.

7

Keeping records is a habit — like exercise, even three lines every day.

8

Explain to others: share and teach — that is how you become confident that you understand.

9

Divide the work between AI and me — AI collects and processes; I read, review, and analyze.

10

Note-taking is a process toward note-making — keep reading until a thought emerges.

Source · Slide 57

Reading Discussions Become a Newsletter

Hallym University Media School Colloquium · 2026.06.10

Reading discussions become
a newsletter

Writing becomes media. Keeping records is a practice.

Presented by Achmage (안창현) · Thank you

Source · Slide 58