May 31, 2026

Why 5BY.AI, an AI Conversation Memory Service, Studies Human Memory and Judgment

5BY.AI may look on the surface like a service that remembers AI conversations and helps you continue them.

But what 5BY needs to study in the long term is not just conversation data or feature lists.

The more fundamental subject of study is people.

What do people judge as important
Why do they forget important thoughts over time
Why do they重新 ponder problems they have already solved from scratch
Why do they repeat the same mistakes without recalling previous judgments
Which moments within long conversations do they want to revisit
What context is needed when starting a new AI conversation to continue thinking

Without answering these questions, no matter how many AI conversations are stored, it is difficult to actually help users resume their thinking.

What 5BY needs to study is not how software replaces human memory.

It is how to help people better discover, understand, and reuse their past thoughts and judgments.

Storing AI Conversations and Continuing Thinking Are Different

Storing AI conversation raw text is technically implementable.

You can put conversation titles, dates, questions, and answers into a database and provide search functionality.

But the mere fact that conversations are stored does not mean the user can return to their past thinking state.

What the user actually needs may be the following information.

What were they trying to solve at the time
What conditions did they consider important
What did they choose among multiple proposals
Why were certain options excluded
What triggered a change in judgment
What problems remained unsolved
Where should they continue from next

All this information is mixed within long conversation raw text.

But the user does not want to re-read the entire conversation; they want to quickly find where their thinking stopped and the criteria for their judgments.

So 5BY's problem is not a simple storage problem.

It is a problem about how people re-understand their past thoughts.

People Do Not Remember All Experiences with the Same Importance

People encounter vast amounts of information and conversations in a day.

But they do not remember all of it at the same level.

Emotionally strong moments, information used repeatedly, content connected to existing thoughts, and judgments that led to actual actions tend to remain longer.

Conversely, judgments that were important at the time but have no trigger to recall can quickly fade.

Similar problems occur with AI conversations.

Even if you made an important decision within a conversation, you may not immediately recall it the next day when starting a different conversation.

After a few days, you may remember that you decided something but forget why you judged it that way.

After more time passes, even finding which AI service the conversation was in becomes difficult.

Memory is not simply about whether data remains.

It is about whether you can recall and use it when needed.

Why Do Important Judgments Easily Disappear

Important judgments disappear because the judgment itself does not exist only as a short sentence.

A single decision contains multiple contexts.

The goal at the time
Available information
Time and cost constraints
Pros and cons of other options
Failures already experienced
Conditions that must be upheld
Values the user considered important

Over time, only the final conclusion may remain while this background fades first.

For example, the user may have decided in the past not to build a specific feature.

But months later, they may only remember the fact that they decided not to build it, without recalling the security risk, maintenance cost, or user confusion issues that were the reasons.

Then the same discussion gets repeated from scratch.

What 5BY must preserve is not a single conclusion but the coordinates of judgment that allow that conclusion to be re-understood.

Why Do People Repeat the Same Mistakes

The reason people repeat the same mistakes cannot be explained simply by insufficient memory.

It may be because previous experiences are not connected to the current situation.

Even if there was a similar problem in the past, the following reasons may prevent recognizing it as the same experience.

The expressions used have changed.
The project and people involved have changed.
The conversation happened in a different AI service.
Past records are buried under old conversations.
The cause of the failure at the time was not clearly organized.
Only the final result remains and the judgment process has disappeared.

Even when people have past experience, if they cannot connect it to the current problem, using that knowledge is difficult.

Therefore, to reduce repeated mistakes, rather than simply preserving past records, you must be able to discover past judgments related to the current problem.

This is also why Graph View and a thought connection structure are needed.

Why Good Thoughts Are Hard to Rediscover

When conversing with AI, sometimes unexpected good ideas emerge.

But at that moment, it is difficult to accurately judge the long-term value of the idea.

The user continues the conversation or moves on to other tasks.

When the idea is needed later, the following problems arise.

They do not remember which AI they conversed with.
They cannot remember the exact expression, making search difficult.
The conversation title differs from the actual content.
There are too many similar conversations.
The idea is embedded in the middle of a long answer.
It is not connected to content developed in other conversations.

Search is strong when you have words you remember.

But people often remember only meanings and feelings rather than sentences.

You may have the memory of 'I had a good thought similar to this before' but not know which words to search with.

5BY must provide reference points and relationships that allow users to return in such situations.

The More LLMs You Use, the More Easily Human Context Scatters

Users do not use only one AI.

They can check implementation methods with Copilot, review long materials with Gemini, and find evidence with Perplexity.

They can compare different analysis directions with Mistral or DeepSeek, and develop ideas and documents with ChatGPT or Claude.

They can also explore new perspectives or recent discussions with Grok.

Each AI offers different strengths and usage experiences.

But the user's thinking does not follow service boundaries.

For the user, it is one project and one concern, but the actual records are divided across multiple chat windows.

Problem definition remains in one AI.
Material research remains in another AI.
The final judgment is made in yet another AI.
The actual implementation process remains in a separate tool.
The reason for the user's choice may not be organized anywhere.

What 5BY must solve is not the problem of copying multiple AIs' answers into one place.

It is the problem of making it possible to understand how one person's thinking continued even across multiple services.

What Technology-Centric Services Easily Miss

When designing a product with a technology-centric approach, you tend to focus on easily measurable items.

Number of stored conversations
Number of generated summaries
Number of search results
Number of graph nodes
Auto-classification accuracy
Number of user clicks

These numbers are necessary for operating a product.

But high numbers do not mean the user is better at continuing their thinking.

You may have stored many conversations but never use them again.

Many summaries may have been generated but make it harder to know what is important.

Many nodes may be displayed in the graph but the user may not know where to start.

Automation may have increased but the user may find it harder to distinguish their own judgments from AI's proposals.

Therefore, 5BY must study not only the quantity of features but what actual cognitive help is provided to people.

Core Questions 5BY Must Study

That 5BY studies people should not end with abstract philosophy.

It must address concrete questions directly connected to product design.

When Do People Judge a Conversation as Important

A long conversation is not necessarily important.

A single short question can change the core direction of a project.

We must examine at what moment users feel 'I need to revisit this conversation.'

What Do Users Remember and What Do They Forget

Users remember conclusions but may forget the rationale.

They remember problems but may not find the conversation where they were solved.

Depending on the form of memory, the design of search, Anchor, and Graph View must also change.

How Much Context Is Needed to Restart

If context is too short, important conditions are missing.

Conversely, if too long, the new conversation becomes complex and the user has difficulty grasping the core.

Handoff must contain enough context to start the next task.

Do Users Trust Automation or Direct Selection More

Having AI automatically select all important conversations can be convenient.

But it may differ from the user's actual judgment.

Conversely, making users select everything increases usage burden.

We must study where to automate and where to require explicit user selection.

How Do People Revise Past Judgments

A decision that was right in the past is not guaranteed to be right now.

5BY must not simply authorize old judgments as authority but help users re-judge by comparing past and current conditions.

Anchor Must Be a Feature That Respects Human Choice

Anchor is not a simple bookmark.

It is a reference point where the user personally selects a flow worth reusing from their conversation so far.

Anchor is needed because AI cannot perfectly judge importance on behalf of the user.

AI may judge long answers or repeated topics as important.

But the moments important to the user may be completely different.

The moment a long-considered problem was first expressed precisely
The moment a wrong assumption was discovered
The moment one option among several was abandoned
The moment a principle to uphold going forward was set
An idea not yet complete but with felt potential
A question they definitely want to continue in the next conversation

Anchor must reflect these user judgments in the product structure.

Saved Shows Conversations People Actually Want to Revisit

Saved is a selection left when the user judges that a specific question and answer are worth revisiting.

Examining Saved records helps understand what kinds of information people repeatedly need.

For example.

Answers that explained complex concepts in an easy-to-understand way
Warnings that prevented important errors
Procedures directly applicable to actual work
Sentences where the user felt they expressed their thoughts accurately
Starting points for new ideas
Criteria to repeatedly check in subsequent judgments

5BY should not view Saved merely as a count of favorites.

It should be seen as a clue to understanding what people judge as useful and important.

Handoff Must Supplement Human Working Memory

When starting a new AI conversation, the context of the previous conversation does not automatically carry over.

The user has to re-explain the previous situation.

But even people do not accurately remember all conditions from previous conversations.

They may omit some conditions or confuse recent decisions with older ones.

Handoff is a feature that conveys the context needed when the user starts a new conversation.

A good Handoff is not a simple conversation summary.

It must be able to distinguish the following content.

Current goals
Already confirmed facts
Decisions the user has finalized
Conditions to maintain
Excluded options
Unresolved problems
The task to request from the next AI

Handoff should not be a feature that replaces human memory but one that supplements the working context people easily miss.

Graph View Must Let Users Explore the Structure of Memory

People's thoughts are not completely separated like folders.

One idea can connect to multiple projects, and past failures can influence new judgments.

Graph View is a feature that shows these relationships.

But simply displaying a lot of information in a graph is not enough.

Users must be able to answer the following questions.

Where did this thought start
After which judgment did the direction change
By what criteria do different conversations connect
Which past experience influenced the current judgment
Which flow is worth continuing now

The success of Graph View should not be judged by node count but by whether the user can better understand their thinking and choose the next action.

Studying People Does Not Mean Collecting All Actions

The expression 'studying people' can be dangerous if used incorrectly.

It does not mean monitoring all user actions and conversations or collecting as much personal data as possible.

What 5BY must study is not methods for analyzing and evaluating individuals but the common memory and context problems people experience.

The following principles are needed.

Users must personally choose what to preserve.
Unnecessary raw text and personal information should not be collected.
User thoughts should not be arbitrarily evaluated or graded.
Designs intended to manipulate human behavior should be avoided.
The basis and limitations of automation judgments must be clear.
Users must be able to control their records and connections.

5BY's human research should not be research to obtain more data but research to help users better continue their thinking with less information.

5BY's No Raw Storage Principle Also Stems from Human Research

Storing all AI conversation raw text may increase the possibility of finding needed content eventually.

But storing entire raw text carries other costs.

Sensitive personal information may remain unnecessarily.
Important thoughts and casual conversations mix.
As records grow, the cost of re-reading increases.
It becomes hard to know what the user chose as important.
The stored records themselves can become a new burden.

5BY's 'no raw storage, meaning only' direction is not simply a technical choice to reduce storage space.

It stems from the question of whether what people actually need again is a copy of every sentence or coordinates that can restore judgments and context.

Studying People Changes Feature Priorities Too

When thinking centered on human behavior and memory, product priorities also change.

Rather than adding more features, the following problems may become more important.

Do users easily understand when to leave an Anchor
Is the difference between Saved and Anchor distinguishable by action
Is Handoff neither too long nor too short
Do users get lost in Graph View
Can old judgments and recent decisions be distinguished
Even when moving to another AI, do accounts and context not mix
Does information the user did not select get arbitrarily preserved
Does reviewing records naturally lead to new work

Answers to these questions cannot be obtained by reading code alone.

You must observe where actual users stop, what they misunderstand, and at which moments they feel value.

User Mistakes Are Also Important Evidence for Product Research

When users use a feature incorrectly, you should not simply conclude they did not read the explanation.

The product's concepts and screens may not match human expectations.

For example, if users repeatedly confuse Anchor and Saved, the following possibilities should be examined.

The purposes of the two features do not appear sufficiently different.
The names differ but the actual actions feel similar.
The moment when the user should choose a feature is unclear.
The difference between the two features is not revealed in the result screen.

A product that studies people does not blame only the user for mistakes.

It must find from what mismatch between expectation and design the mistake originated.

5BY Must Not Judge on Behalf of People

A service that helps with human memory and judgment may be tempted to draw conclusions on behalf of the user.

It could automatically select important conversations, recommend next actions based on past records, and infer user tendencies.

Some automation can be convenient.

But 5BY must not weaken the user's decision-making authority.

What is important must be ultimately selected by the user.
Past records should not become evidence that forces current decisions.
AI recommendations and user-confirmed judgments must be distinguished.
Users must be able to modify and delete their records.
When automatic connections are wrong, they must be easily correctable.

5BY should aim not for a system that judges better than people but for a system that helps people better continue their own judgments.

The Research 5BY Needs Must Connect Multiple Fields

How people think, remember, and judge is difficult to explain with a single field.

Fields 5BY should examine in the long term may include the following areas.

Human memory and forgetting
Decision-making and judgment errors
Problem-solving processes
Metacognition and self-reflection
Knowledge management and organizational memory
Human-AI collaboration methods
Interface and user experience
Privacy and data control
Provenance and trust of records
Collective intelligence and knowledge transfer

Simply applying concepts from these fields to the product is not enough.

You must also observe what problems actual 5BY users experience as they move across multiple AIs and which features provide help.

5BY's Success Criterion Is Not Storage Volume

5BY's success should not be judged solely by the number of stored conversations or generated coordinates.

More important questions are as follows.

Did the user rediscover important past judgments
Did the time spent repeating explanations from scratch decrease
Could they continue thinking even when moving to another AI
Did they connect old ideas to current problems
Did they receive help to avoid repeating past failures
Did they come to understand how their thinking developed
Could they pass important know-how to the next person

If these changes actually occur, 5BY can go beyond a simple AI conversation storage tool and become a service that helps continuity of thinking.

Our Company's Real Research Subject Is People

5BY needs good code and stable systems.

Technology to respond to changes in AI services, connect conversations across multiple LLMs, and safely manage user records is also important.

But technology is not the goal.

Technology is a means to help people not lose their thoughts and continue to better judgments.

The question 5BY must keep asking is not how many more features to build.

Why do people forget important thoughts
What cues are needed to recall them
How much past context must be shown to understand
When should AI help and when should people choose directly
How can records become not a burden but the starting point for the next thought

If these questions are missed, 5BY can become a service that stores many conversations but struggles to become one that connects people's thoughts.

5BY's real research subject is not software.

It is how people think, remember, forget, choose, and judge again.