Over the past year or so, I have been giving considerable thought to AI and its implications. Often, when a new technology comes online, it is easy to paradoxically over- and underestimate it.

We overestimate what it can do right now and underestimate the long-term implications and second-order effects. This is where a lot of hidden value is; it’s in the second-order effects, which are very difficult to see and spot.

To that end, I am creating a series on such insight.

Below is Part 1. The focus is on how AI is changing the internet, what is valuable in society, and how it will impact organizational structures.

Let’s jump in.

Subscribe now

Search is Over: Welcome to the Age of Research.

Today, we can analyze, synthesize, and understand the latest information on any topic without conducting our own research or reading and synthesizing all the information; instead, we can have AI do it for us.

With this, we're moving from 'Search' to 'Research' and, in the process, compressing both time and information.

This isn't just about finding information faster.

It involves transforming a plethora of blue links and dispersed data into concise information that effectively addresses the core issue within minutes.

This shifts the power from individuals with excellent memories and extensive knowledge to those who can think deeply, critically, and ask insightful questions.

With this change, going from Search to Research, the old internet is dying.

The Old Internet is Dying

Until now, websites primarily served as sources of information, resembling digital magazines or brochures. Search engines sent traffic, and companies monetized that attention.

This entire system is collapsing.

Why the old model is breaking:

• Information websites are no longer needed when AI can answer questions and spin up sites in minutes

• AI is eating search, and SEO is dying along with it

• Click-through rates from AI search, like Perplexity, are devastatingly low

The Numbers are Startling:

• Ten years ago:

  • Google crawled two pages For every visitor it sent a publisher, according to Cloudflare.

• Six months ago:

  • For Google, that ratio was 6:1

  • For OpenAI, it was 250:1

  • For Anthropic, it was 6,000:1

• Now:

  • For Google, it's 18:1

  • For OpenAI, it's 1,500:1

  • For Anthropic, it's 60,000:1

The economics simply don't work anymore. The shift is profound, and it changes everything about how the internet works.

Share Stefan Speaks

The Future of Websites

In a world where AI can answer questions, what is the purpose of websites?

What we are seeing is a transformation of the internet itself.

We are going from a static, dumb internet to a smart, dynamic, and interactive one that personalizes experiences in real time. By extension, to survive and add value, websites themselves become dynamic and personalized, and as a result, more useful and valuable.

So, websites won't disappear; they'll transform into something entirely different.

They'll become destinations where you meet people, have experiences, and accomplish goals that you can't or don’t want to do with your personal AI Agent.

What's coming:

• Websites become destinations that offer unique experiences, helping you achieve your goals.

• Websites are evolving from static data to dynamic and personalized experiences that include agentic abilities and memory.

• The MOAT is being able to serve your community in ways that others can't replicate based on your relationship.

• People will have a reason to go because they will be able to do things there that they can't with their AI Agent alone.

• The internet is evolving from providing information to offering knowledge and ultimately achieving understanding. When an agentic website understands you, it can do things on your behalf.

• The future will resemble the past in that, once upon a time, we had relationships with our hairdresser, local bakery, mechanic, and doctor and wouldn't ask our physician about our taxes. In the future, we will have the same relationships with different AI Sites.

• Sites will be able to communicate with each other and AI agents instantly using protocols like MPC.

• Most of the internet becomes invisible, where your AI agent automatically takes action without you seeing the process, like your digestive system.

I built a Dynamic website, which you can play with below.

The Game of Life is an interactive and personalized experience. Soon, The Game of Life will create entire worlds of experience based on your psychology.

Play Game of Life

Such an evolution raises the question: if websites evolve into dynamic experiences with agency that share only limited new data, how will AI acquire fresh data?

The answer lies in how the internet is transforming.

The Nature of the Internet Transforms

The internet is evolving from the "information highway" to a place where information gets applied in the real world, where results are fed back into the system, upgrading our collective understanding of how things work.

We transition from a basic internet that can’t do much to a sophisticated internet that can do almost anything.

We go from the ‘Information Age’ to the ‘Imagination Age.’

And this process has already begun.

We do this every day as we ask AI questions, apply the information, receive feedback on whether it works, provide that feedback back to the system, which updates the process, and then move on to the next challenge.

Soon, AI systems will be able to learn from these interactions and from millions of points of feedback on what works and what does not, as well as how it works, incorporating these learnings to dramatically improve the AI system.

In other words, the future of AI data harvesting will depend on our interactions with AI systems and the data we give it. As we engage in dialogues with AI, conduct experiments, provide feedback, and data, AI learns from our interactions.

I envision a future economy where inquisitive individuals, who create innovative solutions and ask profound questions, are rewarded as they become valuable go-to resources for both people and AI systems.

Again, they will be valuable, not because of ‘what they know’ but ‘how they think.’

Share

AI is Like Fentanyl for Startups

This might sound dramatic, but there is a parallel.

The main reason fentanyl is so devastating and dangerous is that even a tiny amount is highly potent.

It’s 100 times stronger than morphine, making it easier to smuggle and profit from, which means that small mom-and-pop operations can suddenly compete against giants.

This capability is what has made it so difficult to stop.

Stopping it is not about stopping large packages from the Cartel, but instead tens of thousands of tiny packages from everywhere.

Like Fentanyl, AI changes the economics.

Today, AI is enabling small teams to compete with larger players, signaling the start of a significant disruption.

The disruption is real:

• Industries like hedge funds and VCs are losing their information and processing advantages.

• Small players can now use AI to process information, build apps, run tests, and capture alpha from big players.

• AI Startups are growing to $100 Million ARR faster than ever before and with fewer employees.

Consider the implications:

1. AI companies require fewer employees, less funding, and less time to scale.

2. They take market share and displace larger companies.

3. To compete, larger companies implement AI into their existing organizational structure, which reshapes the company.

Let’s look at the impacts of AI on corporations.

AI is Like Corporate Ozempic

Large companies are using AI more than ever. We expect the global AI market size to reach $190 billion this year, indicating a 38% growth rate from previous years.

This is because AI has an excellent ROI and is all but guaranteed to pay for itself.

Approximately 92% of large companies are achieving returns on their investments in artificial intelligence, and a similar percentage are increasing their AI investments.

And then there’s what Meta did.

From March 2022 to the present day, Meta was able to Grow 700% while letting go of 20% of their workforce.

To the best of my knowledge, no business of this size has ever increased its revenue by 25% while simultaneously reducing its workforce by 25%..

This marks the birth of a new era.

Today, we're in a new world of higher profits with fewer employees.

By 2030, AI is projected to boost global GDP by 26%, contributing an estimated $15.7 trillion to the economy.

If you want to see how Companies are using AI, I launched GoBeyond.AI.

GoBeyond.AI features an AI Directory of AI Tools and Case Studies, allowing you to see how large Corporations are utilizing AI and what’s working.

AI Directory | GoBeyond.AI

AI and Your Job

Do you use AI in your job role?

What is intriguing here is the number of people who don’t realize they are using AI.

In a survey of 6,000 people, only 33% reported using AI, but over 77% were actually using AI-powered services or devices without realizing it.

Workplace augmentation is already taking place, even if you may not be aware of it.

AI is fundamentally transforming our work processes in the following ways:

1. Doing more with less means that many teams can increase their output by 5-10 times, allowing one employee to accomplish the work that previously required five employees.

2. Automating Tasks: AI can automate tasks, meaning that the nature of your role changes, and consequently, your value to your team also changes. This has two profound effects:

   1. Time: You will have extra free time, which means you will be given tasks outside your role.

   2. Value: Since AI can do a part of your job, this might make your entire role less valuable to your employer.

3. Changing Job Requirements: Jobs and roles will become more fluid, which will significantly impact team dynamics and power dynamics within organizations.

4. Salary & Value: Overall, this will exert downward pressure on most job types and roles as knowledge and expertise become more democratized.

What This Means for You

The old world is dying.

If your MOAT is what you know and your level of expertise, consider this a giant RED Flag.

This is your Sputnik moment.

The future belongs to those who know how to think and ask profound questions—not those who rely on knowledge, memory, or information processing.

We're moving from the Information Age to the Imagination Age.

Soon, AI will handle the execution.

The question is, are you developing your imagination and ability to ask questions that come from understanding, or are you overly dependent on a system that's replacing the world you knew?

Share

Leave a comment

Simulations have transformed the way we see the world.

Simulations are copies of reality, and as they grow and multiply, they drown out the underlying reality, leading to a situation in which we first and foremost see the copies and take them as reality.

This affects us on every level, from who we date to our relationships and even our self-image.

Take Instagram, for example. Images are often filtered, edited, and manipulated to idealized perfection, which has been shown to have a devastating effect on girls and young women’s self-esteem. They don’t feel they measure up.

Our perception of beauty sets up our dating expectations.

Dating Apps promise connections but often base matches on superficial factors. The algorithm is designed to give us fun but meaningless matches instead of highly compatible ones.

Our news feeds are no longer just information but algorithmically tailored streams designed to captivate, prioritize engagement, and confirm our beliefs.

Even our conversations are shifting. We interact with AI chatbots that mimic human emotions so convincingly we often forget they lack consciousness.

These technological interfaces influence how we see the world and ourselves.

They distort our perception of reality and subtly and persistently guide our values and aspirations.

@maquiagem_feitas@princxssglitterhead #maquiagemtiktok #maquiagemchallenge #maquiagem #transformação #maquiagemluxo

Tiktok failed to load.

Enable 3rd party cookies or use another browser

We measure ourselves against the synthetic benchmarks, planning our futures around digitally induced goals.

And as AI grows more sophisticated, we are ushering in a new era of mediated reality, where artificial interactions become more common than authentic ones.

Subscribe now

Here are 10 Ways AI affects our Perceptions and will Transform our World:

1. Anthropomorphism of AI Agents

People often anthropomorphize AI, attributing human-like qualities to it. This helps individuals make sense of AI even though it is not valid.

This anthropomorphizing is not benign. It will lead to pseudo relationships that look and feel like the real thing but aren't, as we will see below.

2. What is Beauty?

Studies show that people unaware of the content source prefer AI-generated content. However, when they know the content is AI-generated, they are biased towards human-created content, demonstrating "human favoritism."

When the beauty of a piece of content is based on who created it rather than on its own merit, the notion of beauty as a real thing breaks down.

This simultaneously makes us question our creative abilities and detaches us from beauty.

3. Personalized AI-Generated Content

Soon, AI will be able to create highly engaging and personalized content designed to keep you glued to your seat.

4. Fake News & Deep Fakes

AI content is already flooding the internet, feeding and training other AI content, and this has not gone unnoticed. Flooding our information space with fake news is now easier than ever, and many are already taking advantage.

Soon, we won’t even be able to trust our eyes with deep fakes. Just look at how good the Deep Fake below is and remind yourself that this technology just got started.

5. Superficiality

AI systems shape our perception of reality by focusing on quantifiable data. This can lead to a narrowed view of the world, emphasizing measurable aspects while potentially neglecting unquantifiable elements like relationships and spiritual values.

We give up the real conversation, spending real time with real people for online likes and interactions.

6. Synthetic Relationships

As people interact more frequently with AI, their perceptions of AI's qualities are reinforced, which can lead to an increased belief in AI's ability to think and feel like humans. Many people will prefer synthetic relationships with AI over real ones with people.

We are already seeing this with ReplicaAI, which builds AI Friends and Companions. And this is just the beginning. There are already AI Sex Robots and even a Cyber-brothel in Germany.

7. Who are you Really Talking To?

AI systems are increasingly capable of detecting and responding to human emotions. People are already using AI to text them on dating apps, making us wonder who we are talking to.

So even when you are talking to a friend, you might actually be talking to an AI.

8. AI Reality Filter

Soon, you’ll be able to make your partner look like anyone you want. This is a mixed reality in which ‘the simulation’ is literally projected on ‘the real.’

You can also see an extension of this, where everyone sees a different world through their AR/VR glasses.

9. AI’s Impact on Cultural Influences

Imagine a world where AI can create any kind of content and produce any type of experience, many of which are illegal today.

Is that moral? Is it ethical?

At the very least, this will push the Overton Window of what is socially acceptable, which will have profound ramifications for our society.

10. AI's Role in Shaping Social Norms

In a society where everything becomes possible through simulations, we begin to lose trust in what we see and experience.

Things no longer have any weight and, at the very least, are superficially deep. We look for pleasure and to collect meaningless trinkets to impress others.

We have given up the real for the simulation then finding meaning and purpose are like trying to find love at a Cyber-brothel.

We are entering a world where everything is possible, and nothing means anything.

In such a highly simulated world, where nothing means anything, then trust and morality begin to break down.

Subscribe now

Share

Right now, there is quite a lot of fear-mongering on the possibility that AI will one day become conscious, have a will of its own, and take over the world.

This fear speaks to a broader existential fear, where we project God-like powers, an outside entity that unleashes its vengeance on us and threatens to destroy the world. We see this in our stories, like Godzilla, the Aliens on Independence Day, and the AI robots in Terminator.

The real danger isn't that AI is Conscious and could have a will of its own but the opposite.

The danger that AI poses is it is a very powerful computer program that does not really understand, learn, or have a dynamic memory, and it can't tell the difference between real things and fake things. 

This means that AI can easily be duped into executing harmful activities that is similar to the most common use cases.

Let's look at a few popular use cases and how easily they can be corrupted.

Subscribe now

1. Programming

When AI helps write code, it doesn't know what it is really building nor does it have the ability to piece it together which will make it easy for non-programers to create powerful AI Viruses and Malware.

• Writing Code → Creating Viruses: AI can't distinguish between malware & software.

2. AI Customer Service

Imagine that you are talking to an Apple Customer Service Agent about a virus on your Mac. The Customer Service Agent has been super helpful in addressing your concerns and finally suggests that you purchase an AI Anti Virus Software for your Mac.

You make the purchase, install it, and presto, it works!

The virus is gone, and you're happy, but the entire thing was a scam, and worse yet, you just downloaded ransomware.

• Real Customer Service for a Fake Scams: AI can't distinguish between real companies, products, services and scams.

3. AI Content Generation

You go to your phone, and everyone online is talking about a terrorist threat on Wall Street.

ISIS just demanded One Billion dollars in Bitcoin, or they will detonate a WMD in New York City.

This sends markets into a frenzy; the Dow drops 25%, Bitcoin goes up 50%, and everyone is on the edge of their seat. People are fleeing the city, causing all kinds of havoc.

Meanwhile, the entire thing was Fake. There were no terrorists, and there was no threat.

A secret group published tens of thousands of Fake News stories and made them appear as such.

Benefiting from Fake News was the premise of Wag the Dog. Their goals were political however in our case, there is a lot of money to be made from a fake story.

Fear blinds people and makes us over react in a very predictable way.

We have a flight to safety and knowing this, the fake terrorist make tens of millions of dollars by shorting the stock market and longing Bitcoin.

After a few days of histeria, it becomes clear that it was a Hoax, and everyone is relieved. They are happy with the outcome and have no idea how the fake terrorists could have benefitted.

News stations, social media companies, and governments worldwide also came out ahead. They were rewarded with all the views and ad dollars that poured in, and people are coming together and rallying around their flag and leaders.

• Fake News on Steroids: AI Content used on scale to promote fake news, political propaganda, pump & dump scams, financial fraud, etc.

Share

AI Hacks, Scams & Weapons

We are now living in a post-modern age, where telling the difference between the Real and the Fake takes depth. This is an age where exploitation will be primarily psychological.

Let's look at a few more examples of where we are going:

• AI & LLM Powered Viruses: These types of viruses can use language and communication and evolve their strategies in order to gain their target's private information (e.g.: passwords, etc.)

• AI & LLM Powered Ransomware: Imagine an AI that reads every file on your computer, and when it finds something embarrassing, it asks for money or threatens to send it to your entire contact list.

• AI & LLM-Powered Scams: Today, Organized Crime scams billions of dollars online by targeting vulnerable people and pretending to be their friends. Soon, they will be using LLMs to automate this process.

  • Romance Scams

  • Crypto Scams

• AI Impersonations Scams: Imagine that your grandmother gets a call from you, telling her there was an accident and that you need $500. Would she send it? Today, AI is already being used to scam people using AI Video & Audio to pretend to be a family member in need.

AI Weapons

• Psych Ops: Nation States can use AI as Psych Ops to destabilize and divide their competitors.

• AI Weapons: AI can be used to create chemical weapons, bombs, and other types of firearms that we have not seen. 

  • Killer drones: Using AI Vision, GPS, etc, drones can be given a target and send on a search and destroy mission.

• Smarter Terrorism: Imagine a decentralized terrorist network that could identify and attack extreme vulnerabilities such as our power grid, communication infrastructure, transportation systems, food delivery, etc.

Everything up to this point has been intentional. 

It shows that AI is a tool, software, and, like all tools, someone gets to decide how the tool is used.

Unintended Consequences: Impacts on Culture & Society

Last but not least, we can look at unintended consequences.

Here, we will focus on the cultural implications because, after all, this will be the category that will have the most significant impact.

Trust

As we are thrust into an age of AI and Fake News on scale, it will become harder and harder to trust anything that we see or read. 

The breakdown of trust in our institutions, in our government, and in each other will continue to erode.

Ultimately, this will make it hard to move forward in life, foster meaningful relationships, and commit your life to something.

It will make it very hard for society to solve big problems because, after all, how do you know that a problem is real or real in its portrayal? 

How can you trust the proposed solutions?

Parasocial Relationships → Synthetic Relationships

Parasocial relationships are the relationships you have with your favorite characters, celebrities, or even sports teams.

As we watch them on screen, they seep into our imagination; they become a part of our lives as we pour out interest, energy, and time into them.

They can be fun, easy, and feel intimate.

And this is just the beginning.

With AI, we are already seeing influencers who have digitized themselves and men getting married to their AI Girlfriends, and this trend will only grow as AI, Metaverse, AR/VR, and robotics improve.

Here are a few examples of how far we have come:

In the not-so-distant future, many people will have synthetic primary relationships. They might have AI friends, an AI Assistant, home robots to do chores, and, of course, a robotic AI girlfriend.

Everything about these relationships will be perfect.

The AI will do exactly what you ask, your friends will play to your personality and give you the right amount of pushback, and your AI Girlfriend will be a sassy combination of your mother and the girl your mother warned you about.

The best, worst thing about these relationships will be how perfectly fake they are.

They will give us everything we want and need, except a real, genuine, loving relationship.

Rise of Nihilism, Existential Dread & The Loss of Meaning

When almost everything is fake or synthetic, then what difference does it make to be real? 

And if the real does not matter, how could you matter?

To put in another way, if you are a Physicalists, a reductionist materials who believes that Consciousness come out of matter and that matter is the foundation of reality then the only thing that really matters is matter.

Hence, fourth, your only value is the utility you can provide society, and when you get too old or are replaceable, you are no longer as valuable.

This is far worse than AI just taking your job; it's also taking your value—not only your value as a worker but also your value as a friend and as a loving partner.

And here we have come full circle.

The real threat of believing that AI can become conscious isn't that it will destroy your life by taking over but will rob you of your value and meaning.

And all it takes for these fears to lurk in the minds of hundreds of millions of people is simply Anthropomorphizing AI.

Such an existential dread, born from a belief in materialism and reductionism, can quickly spiral societies out of control.

Ironically and in the most Hegelian way, Traditionalists will be spared from such fears and the crisis of meaning and provide a foundation while others go through the dark night of the soul.

Share

The Opportunity

The real opportunity is to see past a reductionist, deterministic worldview.

Through experience, we can see the difference between the Real and the Fake.

Fake News can teach us to become more self-aware, more psychologically astute, and improve our critical thinking.

Through experience, we can know that Real relationships are more than the sum of their parts. In contrast, Fake AI relationships are only made of their parts. They are just a performance and there is nothing real about them.

This was brilliantly captured in Black Mirror, after a Widow orders a AI Clone of her husband. He looks, walks, talks and acts like him, but something is missing. Fundamentally, there is no one home.

As we engage more and more with AI, with deterministic systems, we begin to discover more and more of what makes us human.

We can begin to reflect on who and what we are and find value and self-worth beyond our material contributions.

In a post-modern world, where everything is possible and nothing is true, we can begin to identify principles that matter beyond perspective.

Through our pain and suffering, we can heal ourselves and our world.

Share

Hi there, welcome to the poll section of Stefan Speaks.

Right now, I am talking with Google’s Gemini Team about speaking at the Chatbot Conference in San Francisco, on September 24th.

What topics would you most want to hear from Google’s AI Team?

PS: IF none of these fit, please suggest a question in the comments.

Right now I am focusing on 5 primary AI Topics. Please let me know which interest you.

Here is what each AI Topic Includes:

1. Insight Articles

   1. Future & AI & LLMs: Where we are and where we are going within AI

   2. Big Picture Perspective: Impact on Media, Internet, Culture, Businesses, Ethics, AI governance, privacy, security, etc…

2. Tutorials & Deep Dives

   1. Deep Dives: What things are and how they work

      1. eg: Training Models, Knowledge Bases, RAG, CUX, Flows/NLU,

      2. Challenges: Bias, Hallucinations, etc

   2. Tutorials: How to work with them

      1. eg: Building a Bot; Content Writing for LLMs

3. Experiments

   1. AI Experiments: Building AI Agents, LLMs, RAG, Prompt Chaining, etc.

   2. Business: AI to for SEO/Lead Gen/Sales/Customer Service, Productizing AI

4. Roundups:

   1. Quick List: Reviews of topics, curated lists of insights or tools, and periodic statistics that offer snapshots of the AI landscape.

5. Tools & Resources

   1. List of Tools, Templates, and other Resources

Also, feel free to suggest any other topics in the comments.

Cheers

Stefan

Subscribe now

Next, we will dive deeply into how to organize a Knowledge Base. For this, I created a video presentation along with the experiments, and below are the main concepts and how to apply them.

Hierarchical Structures

What are information hierarchies?

Informational hierarchies are a way to organize information so it sets a clear context and conveys the relationship between entities and the overall context. It is a clear signal to what information is most important, and it can help clarify the proper context.

Topical Map

The goal is to create a topical map that includes all of the relevant entities within the proper contextual hierarchies so that the LLM can quickly identify the context.

In essence, what we are doing is creating content that is easily identifiable by the LLM. It is like creating a mirror image of what it already knows about your main topic. Then fill in the details according to your data.

Let’s look at our example:

• Central Entity: This is the main thing that our content is about.

  • Conference

• Source Context: Going to a Conference. Going implies time and place. In fact, you can’t go to a conference that is outside of time or space, so these are required entities that must be included in the content.

  • conferences/time

  • conferences/location

• Root, Seed, Node: This is the hierarchy in which things happen related to our Central Entity and source context. Here, you get to decide how to order things based on your preferences. For example, if this was only an event that happened in NYC, then location would be more important than the year.

  • root/seed/node = root/topic/subtopic = conferences/year/location

• Other Topics & Subtopics: To fill out your Topical Map, you will need to include all of the other relevant topics and subtopics. Here are a few examples:

  • Topics

    • What is the main theme of the conference?

    • What are the main topics?

  • Speakers

    • Who is speaking?

    • What are they speaking about?

    • Why them?

  • Agenda

    • What is the agenda?

    • Time and Place?

  • Attendees

    • Who attends?

    • Why attend?

  • Sponsors

A great topical map will give the LLM a lot of details to go on which will make it easier to answer detailed questions about the event.

Now, how do we actually organize it?

Organize Knowledge in Folders

Hierarchies can be organized into folders, sub-folders, etc. This is a parent-child type of relationship.

Organize Knowledge via URLs

Well-structured websites have a proper URL hierarchy that conveys the correct meaning, context, and relationship between entities.

You can use URLs to showcase these relationships. Consider the difference in meaning between:

1. ChatbotConferences.com/conferences/2019/nyc: URL implies that there are multiple events in multiple cities.

2. ChatbotConferences.com/new-york-city: The URL only shows the city, so all we can infer is that something is happening in NYC.

3. ChatbotConferences.com/nyc/2019: This URL conveys that there are multiple events in NYC.​

4. ChatbotConferences.com/blog/104335923032: This means nothing…

5. ChatbotConferences.com/Chatbot-Conf-NYC-2019: This URL refers to a single event. We don’t know if it is part of a category or if it is related to any other events.

Now, which example do you think is best suited for our knowledge base?

The first example is the strongest. It has a ROOT, of Conferences. The ROOT is the main topic of the website. The second level of the hierarchy, or SEED is the year, and the final NODE is the city. Organizing it by year is much easier than by city, which is why year is at a higher level within the hierarchy. 

Test

The Hierarchy Test tests these ideas out to see to what degree they make a difference with LLMs. I built two chatbots for this test. The chatbots have the same information, but the main difference is the hierarchical organization. 

1) Good Bot: The bot is trained on the following pages:

• https://www.chatbotconference.com/conferences

• https://www.chatbotconference.com/conferences/2017/san-francisco

• https://www.chatbotconference.com/conferences/2018/san-francisco

• https://www.chatbotconference.com/conferences/2019/nyc

• https://www.chatbotconference.com/conferences/2019/san-francisco

• https://www.chatbotconference.com/conferences/2020/online-spring

• https://www.chatbotconference.com/conferences/2020/online-fall

• https://www.chatbotconference.com/conferences/2021/online

• https://www.chatbotconference.com/conferences/2022/online-fall

• https://www.chatbotconference.com/conferences/2022/online-spring

​2) Bad Bo: The Bot is trained on a single Page

• https://www.chatbotconference.com/knowledge-bases/all-agendas

What to try it yourself? Go to the Test and Chat with the Bots

Documents: Context & Semantics

Does the way information within a document is organized matter?

Semantics helps LLMs identify the main context of the document and the relationship of entities with each other. These relationships inform the LLM on the proper way to connect concepts and the topic, which later helps give rise to meaning.

Macro & Micro Semantics

The overall context of a document is set by Marco and Micro Semitics.

The Marco Context is the overarching topic of the document and is organized by the H1, H2, H3, H4 Tags.

All these tags are in order of importance and set up the Marco context. The H1 tag, is the title of the document and represents its overall macro context.

The H2 tags are the main topics within the document and support the overall thesis. The H3 tags are sub-topics of the H2 tags.

Consider the difference between:

1. Marco Context: H1, H2, H3, H4

2. Micro Context: Definitions, questions, phrases, and world order within each heading.

​Well-written documents generally are well organized, and easy to follow, read and understand.

They typically have a hierarchical structure, which allows the reader to go deeper into a topic.

Topics are laid out in a logical and coherent manner. Topics often have sub-topics and supporting information. Well-written documents are able to answer our questions.

Contextual Layers

Consider the questions below.

We broke it down into its elements so you can gain insight into how an LLM reads it. If children is well defined or the document has contextual layers, then an LLM can answer a detailed question like this.
 

• What are the best bikes [knowledge domain] for [functional word] short boys [contextual domain]

• What are the most useful diets [knowledge domain] for [functional word] children with insomnia [contextual domain] for kids under six [contextual layers]?

What is the overall structure of a good document? 

A good document should follow the following parameters:

• Macro Content: H1 Title Tag

• 10% Summary: Extractive & Abstractive Summary

• 60% Main Topics: H2 Tags & Micro Context: Definitions, paragraphs, etc

• 30% Supplementary Context: H2 & H3 Subtopics, related topics, synonyms, antonyms, etc

Test

We will test two articles that have the same information. Article 1 will have all of the attributes shared above. Article two will miss most of the attributes above. Each article will be used to train a bot, and we can all play around with the differences.

1) Good Bot 1: Doc has the following 

• H1 Title Tag

• Summary

• H2 & H3

• Definitions

• Supplementary Content​

2) Bad Bot 2: Doc is designed to mirror a poorly structured articles

• All tags converted to Paragraph Format

• No summary

• No supplementary content

• No definitions or questions

Join the Journey

Join me as I journey and navigate through these waters.

Over the next few weeks and months, I will share my insights on AI, philosophy, and the results of my experiments in implementing the best versions of these technologies.

Subscribe now

​

So recently, I have been experimenting with knowledge bases, and how information is structured is vital.

In some sense, it directly relates to the context and overall meaning. It also makes it easier for LLMs to answer questions accurately and reduces hallucinations.

And so we created a few experiments that you can play with and test yourself.

But first, why does information structure matter?

Imagine a library. One room has books strewn everywhere, titles mixed up, and no discernible order. In another room, the books are organized by subject, then by author, then by publication date. Which room would help you find the book you want more efficiently?

LLMs are similar. The way knowledge is presented and organized dictates not only their understanding but also their output.

Our recent experiments highlighted two crucial aspects:

1. Organization of Informational Taxonomies and Hierarchies: This considers elements like URL structures, folders, and how information is interrelated. By defining the proper context, you can highlight what's critical and organize it appropriately.

2. Organization Within a Document: Delving deeper, this looks at the composition of individual pieces of information, from structure and semantics to formatting and summaries.

Let's dive into each experiment and the results.

To make this more fun and interactive, I am providing the documents and URLs used to train the AI. Additionally, you talk to the AI Agents, Good Bot and Bad Bot, and experience the difference for yourself.

Informational Hierarchies: How Structures Imply Meaning

At its core, informational hierarchy is about context. Whether it's a URL on a website or the structure of folders within a system, hierarchies set the scene and help LLMs understand the importance and relevance of different data points and how they relate to each other.

Consider these URLs:
- ChatbotConferences.com/conferences/2019/nyc: This URL suggests there are multiple conferences across different cities. From this URL, we could even extrapolate that there are other types of events that are not conferences.
- ChatbotConferences.com/new-york-city: This URL just refers to the city of New York. We can infer that there is or was a conference there, but not much more. Overall, this is ambiguous.
- ChatbotConferences.com/nyc/2019: This URL indicates multiple NYC events but omits a broader context. For example, are there conferences in other cities?

The Great Hierarchy Test

We started on a quest to understand the weight and importance of hierarchies. We build two chatbots, with their primary differentiation being the organization of their hierarchy:

1. Bot 1: Trained on multiple pages, each representing a distinct event and year with a well-defined hierarchical structure. Here are the URLs:

   • https://www.chatbotconference.com/conferences

   • https://www.chatbotconference.com/conferences/2017/san-francisco

   • https://www.chatbotconference.com/conferences/2018/san-francisco

   • https://www.chatbotconference.com/conferences/2019/nyc

   • https://www.chatbotconference.com/conferences/2019/san-francisco

   • https://www.chatbotconference.com/conferences/2020/online-spring

   • https://www.chatbotconference.com/conferences/2020/online-fall

   • https://www.chatbotconference.com/conferences/2021/online

   • https://www.chatbotconference.com/conferences/2022/online-fall

   • https://www.chatbotconference.com/conferences/2022/online-spring

   ​

2. Bot 2: The Bot was trained on a single page that contains all of the same information as above.

   • https://www.chatbotconference.com/knowledge-bases/all-agendas

What were the results?

Overall, Good Bot 1 was able to answer more questions accurately than Bad Bot.

Bad Bot struggled with even simple questions like, ‘How many conferences were there in 2019?” or ‘How many conferences were there in San Francisco?’. It got simple questions like this wrong. Without even having to look inside the documents, Good Bot was able to answer these questions, simply based on the URLs!

Are there any benefits to Bad Bot?

Surprisingly, there were. It generally costs less to ask questions and get responses from Bad Bot. Since Bad Bot only looked at one document, it was cheaper to run.

Now, let’s look look at ways to optimize a Document so you get the most value from it.

Organizing information within a Document

In the same way that a group document gives rise to the overall context and meaning, so does a single document.

Single documents are made up of words, sentences, paragraphs, headings, lists, and meta data, and these structures organize the context and overall meaning of a document.

But how much do these structures actually matter?

Do lists, tags, and headings have a special power? What about the semantics within a document? Does this have a major impact on performance?

To answer these questions, we created the document test! The test features two articles that are identical. The only difference in the articles is the use of tags, lists, definitions, and summary content. In other words, the meat and potatoes of the articles are identical.

The Document Test

We used an article about the benefits of drinking water to create our test. Here is a little information about the article:

1. Title: 20 Health Benefits of Drinking Water: Physical, Psychological and Nutritional

2. List of 20 Benefits

3. Focuses on the Physical, Psychological and Nutritional Benefits

4. Defines what each of these things means. For example, it defines the difference between the physical and nutritional benefits of drinking water.

5. It also considers the opposite of drinking too much water, such as dehydration, and situations when drinking water is unhealthy, like drinking too much water or poor quality water.

Now, this document was the 'Control', and we created a second document as the ‘Experiment’. The Experiment Document had all of the same information except that it was all structured differently.

Below are the differences between the documents.

1. Document 1: Ideally Structured Article (Control)

   1. H1 Title Tag

   2. Semantic Summary

   3. H2 & H3

   4. Definitions

   5. Supplementary Content

2. Document 2: Poorly Structured Article (Experiment)

   1. All tags converted to Paragraph Format

   2. Simple Summary

   3. No definitions or questions

Results

We then trained two bots on these articles and asked each bot simple questions. Below are the results of the answer and the costs in terms of tokens.

1. What are the 20 benefits of drinking water?

   1. Good Bot Accuracy: 60% | Cost: 1265

   2. Bad Bot Accuracy: 80% | Cost: 1117

2. What are the benefits of drinking water?

   1. Good Bot Accuracy: 100% | Cost: 2771

   2. Bad Bot Accuracy: 80% | Cost: 2876

3. How many benefits are there of drinking water?

   1. Good Bot Accuracy: 100% | Cost: 2659

   2. Bad Bot Accuracy: 0% | Cost: 2550

4. What are the types of benefits of drinking water?

   1. Good Bot: Accuracy: 100% | Cost: 2615

   2. Bad Bot: Accuracy 100% | Cost: 2580

5. How does drinking water help weight loss?

   1. Good Bot: Accuracy: 100% | Cost: 2854

   2. Bad Bot: Accuracy 100% | Cost: 2639

Overall, the differences in accuracy are clear. There is a clear benefit to a well-structured document. For one, it improves overall accuracy, which decreases the cost per accurate answer.

Furthermore, there is a qualitative difference. For example, the last question, “How does drinking water help weight loss?” bot bots answered it correctly; however, Good Bot provided a much more detailed and clear answer. The quality and experience were simply better.

Now, try it yourself: You can play around with each bot and test them for yourself below:

1. The Good Bot: Trained on the Control Document

2. The Bad Bot: Trained on the Experiment Document

   Share

What’s Next?

I hope you enjoyed this piece and took a lot away from it. Next, I’ll share with you my complete guide on ‘How to Organize a Knowledge Base for RAG’. Lastly, if you have any questions, let me know in the comments.

Join the Journey

Join me as I journey and navigate through these waters.

Over the coming weeks, I will be sharing more content on philosophy and mindfulness. Then we will start to look at early results of some of our experiments!

Subscribe now

You can get the Presentation here.

How Large Language Models have Revolutionized Designing AI Agents

Large Language Models are redefining how we build Conversational Agents. Today, LLMs can do most of the heavy lifting in minutes. The days of building hundreds of flows of every useful intent or FAQ question are gone. Instead, LLMs can be trained on a knowledge base and be able to give great answers.

The role of a Conversation Designer has changed. Designers now have to start thinking differently. Instead of thinking linearly and having a flow-based approach to design, now we need to think more about knowledge, information, and data and how all of this leads to meaning-making.

Through stories, knowledge bases, and prompt engineering, designers will soon be creating experiences we once only dreamed of. The job has changed, and it is now more creative than ever.

New Conversational Design Stack

Conversational AI Project will consist of the following design stack:

1. Flows, Intents, and Entities: These will still be useful in use cases where the business needs very specific answers. Additionally, they can be used as a follow-up action to engage the user and take them down a flow towards a product or service of interst.

2. Large Language Models (LLMs): Integrating Large Language Models into your project at multiple stages. LLMs can be used to interpret user intent, entities, and sentiments. They can be used to answer questions. Multiple LLMs can be used in a single project, where each LLM has a specific function.

3. Prompt Engineering: Now the fun begins. Prompt Engineering includes all of the instructions given to an LLM in order to get back high-quality, accurate responses. This includes giving the LLM a role, for example, "You are Jordan Belfort, the super star salesman from Wolf of Wall Street," and prompt chaining.

4. Retrieval Augmented Generation (RAG) for LLMs: This entails training the LLMs on your own data. FAQs and website information is a great example of this. The main challenge here will be designing knowledge bases in such a way that the LLM can understand the information quickly and accurately. Knowledge Base design is key! 

5. Storytelling, Tone, and Personality: We constantly have a story going in our minds. The story is about what we want, what we believe will happen when we get it and if we are on the right track. Even as you are reading this sentence, you are comparing this information to that story and evaluating if it fits. Designers will need to consider the psychology of the customer journey, what story customers have in their minds, and create a design, narrative, and tone that fit the expectations and goals of the user. 

The New Design Thinking

Designing bots is now more about organizing information, understanding ontologies, taxonomies, semantics, and mean-making through stories. In the near future, the job will become even more philosophical, linguistic, psychological, and story-driven.

What does this mean in practical terms? AI Agents will become deeply personalized assistants that can help users on a 1x1 basis. Imagine the ideal assistant. This will become the bar and the bar will be set pretty high.

Here are the abilities that AI Agents will have in the very near future:

Personality Assessment

AI Assistants will be able to quickly identify our personality triats and will personalize the conversation accordingly.

Values Assessment

We pay attention to what we value. Through our attention, we invest in it and give life to our values. These values then become well known and well defined in our culture and language. For example, Latin became the language of science in part because of its precise naming and classification.

AI Agents will be able to identify a user’s values and how they inform their goals. This will allow for a much deeper connection.

Identifying Beliefs

Values are not enough. In order to make a plan around something you value, you also need to believe that it is possible. In this way, beliefs are part of a person’s personal operating system. They are like a set of instructions that define what the world is, who a person is, what is possible, and how one should behave.

AI Agents will need to identify and align with the user’s existing beliefs.

Culture and Background

All of us are born into a culture, a culture that shapes our minds and hearts from the time we are born. It is deep within us, and we express it in our perceptions, values, beliefs, and stories. AI Agents need to understand the cultural perspective of the user how it impacts their identity.

Language

Language translations often dupe us into thinking that meaning is easily transferable between languages and that, ultimately, languages are trivial. In reality, some things can only be said in their native tongue and require a cultural background to be understood. Additionally, each language has its own ‘rhythm’ and ‘personality’ that is embodied in the speakers. Many of these things just get lost in translation.

How a user uses language is very informative and can give the AI insights into their personality, values, culture, and background. Also, this means that different types of AI agents will be built for different languages.

Personal History

Over time, users will tell AI Agents their life’s story. The AI Agents will learn and get an inside peek into the user’s self-concept.

Contextual Situation

As users and AI Agents chat on an ongoing basis, the AI’s will begin tracking the context and situation. This will give the AI Agents a baseline of what the user is like which will allow the AI to notice differences from the baseline state. For example, when the user gets angry.

User’s State

The state is very important because it is part of our perception that filters information. We have all experienced this. When we are frustrated, a small amount of bad news might tip us over the edge into anger. Knowing a person’s state is extremely important during a conversation; it gives you insight into what to talk about and what not to talk about!

Inner Narative

What and how we think is actually quite predictable. We believe that our thoughts are ours, but in fact, we aren’t the first or only people to have ever thought them.

The odds are that many others have had similar thoughts and feelings in similar situations.

This is obvious, but it’s important to remember.

If you know a person’s personality traits, how they perceive the world, what they believe, how they have behaved in the past, and how they talk, and you know their current state and situation, you would be able to answer basic questions about their inner narrative.

In this way, AI Agents will one day soon have insights into what we are thinking.

Project X

It is undeniable that we are heading into areas where many of us won’t feel comfortable. Just the idea that an AI can know more about us, our motivations, and how to manipulate our decisions is unsettling.

This is why it is important that we use these technologies in a way that has the user’s best interests at heart.

An AI whose primary purpose is to help the user discover and define a mission in life that extends beyond their selfish impulses, and that is fundamentally life affirming.

An AI that helps the user live out their mission and achieve their goals, and to do so in a very clear, reflective, and honest manner.

We need to collectively build and experiment with the best versions of this technology so that people have an option available to them that can turn their dreams into a reality instead of a nightmare.

Subscribe now

Bias has something profound to teach us about the human condition, which relates to attention.

We all have a limited amount of attention and give our attention to the things we value.

But how do we know what objects to give our attention to?

How does the evaluation happen?

It happens a priori.

We don't have enough time to scan the environment and evaluate every single object.

Values

Instead, we simply filter the world for what we value.

As a result, we are more familiar with and prefer what we value. These preferences can make us blind to other information and miscalculate its importance and value, which leads to bias.

So, bias first happens a priori; it happens before we give our attention to an object, person, experience, or event.

Now, armed with this knowledge, see if you can count how many times the players in white pass the ball in the video below.

Did you get it right? 

Odds are you did at the expense of noticing other important changes. 

Even if you were just observing, you might have notices all of the big changes but probably don’t know how many times the ball was passed back and forth in total. 

This is an example of how we focus on what we value and miss out on other important facts. 

The most interesting part of this experiment is that knowing this is not enough. 

Perspective

The second level of bias occurs as a result of perspective.

When we look at an object, we don't see all the information about it.

We see it from one viewpoint at one point in time. Our perspective is even more limited with people as we see them in a given situation, environment, and context. So, no matter how we slice it, we have an incomplete and skewed sample of data.

Mean Making

And lastly, there is mean-making.

We are interpretative beings that give meaning to experiences.

Experiences don't have a specific meaning. For example, there is no universal object or situation called 'anger' that produces the feeling of being angry in every human.

The same situation will make some people very angry, and others might not even notice.

We interpret ‘the situation’ and give it meaning. If the situation does not have any meaning, it won’t even be perceived by us!

Take a look at the experiments below:

The Bias Loop

So now we have the complete Bias Loop. Bias happens on these three fundamental levels:

1. Attention Level: You automatically pay attention based on your values.

2. Perspective Level: You see things from a single perspective in time and space, which gives you only a very small snapshot of what something is.

3. Meaning-Making Level: You interpret the information from your perspective based on your current state and your primary drives.

Share

Human Bias’s Impact on LLMs

And this has a direct impact on LLMs.

First, LLMs work via a mechanism called 'Attention'. (To learn more about how this works, check out the paper 'Attention is all you need".)

Second,'model weights' are a byproduct of a mostly modern western perspective, which has its benefits and costs. 

Last but not least, people trained these models and decided what things meant. The meaning of things doesn't come from thin air; it is informed by our cultural values and norms and what we believe to be true about the world.

So, where does this leave us?

Hopefully, this makes us more self-aware.

By understanding how our awareness and attention work within us, we can begin to notice patterns that can lead to bias, favoritism, and prejudice at each of the three levels.

We can see how 'good data' and 'good intentions' can leave out the 'bad data' and 'bad intentions,' which produce an incomplete, biased model.

Knowing a bit about bias, we are finally ready to play around and experiment with knowledge bases.

Join the Journey

It’s almost here. Soon, I will be sharing the details for Project X!

Then, over the next few weeks and months, I will create guides and deep dives into how Project X was created and some of the early results.

Along the way, I will also share the technological and philosophical foundations of this project.

Cheers

Stefan

Subscribe now

Have you ever wondered how a chatbot like ChatGPT or any other Large Language Model (LLM) works?

When a new technology really wows and gets us excited, it becomes a part of us. We make it ours, and we anthropomorphize it. We project human-like qualities onto it, and this can hold us back from really understanding how we are actually dealing with it.

So let's consider a few questions. Mainly, what is an LLM, and what are its limitations?

Perhaps these questions and ideas will illuminate our understanding:

1. Are LLMs a program?

2. Are LLMs a knowledge base? Do they tap into a Database of information?

3. Do LLMs know anything?

Many of us would assume ‘Yes’ to a few of these questions, but when we dig deeper, the ‘Yes’ starts to fall apart.

Consider the following:

1. If an LLM is a program, how does it compute its 70–100 billion parameters in only a few seconds?

2. If an LLM is a knowledge base, why does it need to predict? Why is there a confidence score?

3. How can an LLM model with billions of parameters that has been trained on pretty much the entire internet fit on a 100GB drive?

Now the picture is starting to become more clear. Hopefully, these questions dispel some of the mystique and confusion around LLMs.

There are a number of things that most people believe about LLMs that are contradictory and wrong.

First, LLMS are not knowledge bases, and they are not really programs either. What they are is a statistical representation of knowledge bases.

In other words, an LLM like ChatGPT4 has been trained on hundreds of billions of parameters that it has condensed into statistical patterns. It doesn't have any knowledge, but it has patterns of knowledge.

When you ask it a question, it predicts the answer based on its statistical model.

How do LLMs Work?

LLMs condense knowledge into patterns, and this includes words, phrases, and sentences and how they are related to each other.

Let’s look at a simple example. Imagine that you are in this living room.

This room has objects like a TV, sofa, coffee table, and other furniture. These objects have qualities like shape, color, type, function, etc., which can all be described in words.

Imagine that each of these objects was reduced down to words that described it within this 3D space. For example, the sofa would be described as being white, soft, comfortable, made with linen and cotton, having springs, made for sitting, and so on. If you asked, ‘Where can I sit?’ then all of the items that have the function of sitting would come up as an answer.

In simple terms, this is how LLMs are organized and function.

They use language and semantics to get to the core of meaning (eg: the sofa) and represent these words, phases, and sentences mathematically, as vectors, in a multi-dimensional structure.

When you ask an LLM a question, your question is turned into a sequence of tokens. Each token corresponds to a unit of text, which can be a word or a subword (e.g., "word" can be tokenized into "wo" and "rd"). These tokens are then fed into an LLM.

Once the LLM processes these tokens, it produces output in the form of vectors. These vectors represent the meaning or context of the input tokens in a multi-dimensional space. So, the input is in the form of tokens, and the output is represented as vectors.

Vectors & Embeddings

Here is how it breaks down:

• In LLMs, words, phrases, and sentences are represented as dense vector embeddings. Each word or token is mapped to a high-dimensional vector.

  • Like our 3D environment, where objects are 3D space, imagine instead words being in this type of space.

• These vector embeddings capture semantic relationships and contextual information. Words with similar meanings or usage have vectors that are closer together in the vector space.

  • Like in our living room example, a sofa, loveseat, and chaise are all objects you can sit on, so they will share a similar function and be tied to each other. Conversely, a word like ‘New York City’ would be very far from a sofa because these words are very different.

• LLMs use these vector embeddings for various natural language understanding and generation tasks, including text completion, language translation, question answering, and chatbot interactions.

• The primary goal of LLMs is to understand and generate human language, enabling them to perform language-related tasks with a focus on natural language processing (NLP).

Why do Good Models do Bad Things?

Why do good models do bad things? The answer lies in the models and how they are built.

LLMs are statistical representations of knowledge bases. They have taken the world's information and knowledge and boiled it down to statistical principles.

These principles are like icons. Icons represent something much more than what they are. They are low-resolution images that represent a much bigger chunk of information. They give you a lot more information than meets the eye.

This is how so much information can be stored on 100GB or even as little as 5GB!

And this is the paradox of LLMs. On one hand, they are so profound because of the relationships in their vectors, which allow them to answer a large breadth of questions.

On the other end of the spectrum, they often lack detailed information about the specifics, and this is where the predictions fall apart. Instead of knowing what they don’t know, they often predict what the most likely answer is, leading to ‘Hallucination’.

What are the Challenges and Limitations of LLMs?

LLMs were trained on biased data. We know this because the internet is full of biased data. For example, most of the internet is in English and represents Western values, yet the global population doesn't speak mostly English or hold Western values.

This was acknowledged by Sam Altman when he spoke at Davos. In that interview, he stated AI products will need to allow "quite a lot of individual customization" and "that's going to make a lot of people uncomfortable," because AI will give different answers for different users, based on their values and preferences and possibly on what country they reside in.

What is Bias?

What is bias? Is it ever a good thing? It's very important that we are on the same page of what it means so we can understand it better. ​

Here is the Webster definition of Bias: 

1. an inclination of temperament or outlook, especially a personal and sometimes unreasoned judgment: PREJUDICE

2. an instance of such prejudice

3. BENT, TENDENCY

4. deviation of the expected value of a statistical estimate from the quantity it estimates And/OR systematic error introduced into sampling or testing by selecting or encouraging one outcome or answer over others

Transitive verb

1. to give a settled and often prejudiced outlook to his background, which biases him against foreigners

2. to apply a slight negative or positive voltage to something, such as a transistor.

​The most important insight here is that bias has its roots in our preferences. For example, if you prefer coffee over tea, you are more likely to show bias towards coffee. You might believe that more people drink coffee, that coffee increases your mental focus better, even that it is healthier. At the bare minimum, you will have more information about coffee, which will skew how you view both coffee and tea.

​How does Bias Form?

Every interaction has three components. For example, as you read this sentence, there are the words on the screen, you, the reader, and the meaning you are gathering from this information.

The first stage is attention; you are paying attention to this instead of something else. The second aspect is your perspective. You are seeing these ideas from a point of view that is limited in time and space.

Consider how different this perspective would be if you read this essay five years ago.

Lastly, there is mean-making. All of these words will mean something to you. Depending on your background and education, consider how differently an engineer, a linguist, and a conversational designer would interpret this paragraph.

How does Bias form?

To properly address BIAS, we need to be aware of it at every stage of the process.​

1. Attention: The world has too much information. Based on what we value, we decide where to look and what facts to pay attention to a priori.

   BIAS: By doing so, we are implying that some information is more important than other information. We are showing a preference.

2. Perspective: We only see objects from a point of view. Our perspective skews how we see the things we are paying attention to.

   BIAS: Seeing a limited perspective or only one side of an object, event, person, topic, we are leaving out the majority of the information, and this leaves us open to confirmation bias, selection bias, sampling bias, reporting bias, volunteer bias, publication bias, etc.

3. Meaning-making: We turn limited data on limited things into meaning. Making decisions is a process that involves our identities, beliefs, culture, personalities, etc. For example, consider how an adult and a child would interpret a similar event.

   BIAS: The entire knowledge base is a construct, something we fabricate. It’s a useful invention that doesn't exist outside of us.
   

How to Overcome LLM Limitations

There are a number of things we can do to mitigate the risks of bias, hallucination, and poor accuracy. These solutions are becoming more and more common and they tend to be pretty effecting.

In future articles, I will go more deeply into each of these strategies.

Prompt Design, Engineering, and Tunning:

Bad questions and prompts will lead to bad answers! Consider ways to design your prompts so that they are not biased, that they are open-ended, and that you are referencing very specific facts within the model.

In our Personality NFT Project, getting the right answers about personality depends on understanding the different frameworks, tests, and research. We then reference the theories and models that have a strong scientific foundation so as not to confuse the model with other frameworks that are popular within our culture but scientifically valid.

Knowledge Bases

LLMs won’t know all of the details about your business or project. Creating a Knowledge Base is a great way to feed it your own data and make it easy for uses to ask your bot questions and get answers.

To increase the odds that the bot will answer correctly you need to design and organize the information in your knowledge base so that the LLM will gain the correct vector representations within the document parts.

There are a number of variables to consider, such as ontologies, taxonomies, and semantics. There is a connection between the Hierarchical Structure, the Marco context, the micro context, tagging, and words matter.

​Flows & NLU

When you need to take the user down a specific path and don’t want to have control over what an AI Agent says, you can leverage conversational flows and NLU. In this way, you can take the user down a ‘happy path’, answer specific questions exactly as you intended, and use the LLM and Knowledge Bases to answer less critical questions and to engage the user.

Join the Journey

Join me as I journey and navigate through these waters.

Over the next few weeks and months I will be share my insights on AI, Philosophy, and the results on my experiments in implementing the best versions of these technologies.

Subscribe now