DeepSeek

Deepseek (Shēndù qiúsuǒ in Chinese) is an application of artificial intelligence chatbot launched on January 10, 2025 by the Chinese Deepseek company, which specializes in dialogue

The chatbot is a language model fine-tuned with both supervised and reinforcement learning techniques

It is composed of models DeepSeek LLM, DeepSeek-V2, DeepSeek-V3, and DeepSeek-R1 of Deepseek

Background

In February 2016, High-Flyer was co-founded by artificial intelligence enthusiastic Liang Wenfeng, who had been operating since the financial crisis of 2007-2008 while attending the University of Zhejiang

In 2019, he established high-flyer as a coverage fund focused on the development and use of AI negotiation algorithms

In 2021, High-Flyer used exclusively in commerce

According to estimates of 36kr, Liang had accumulated a warehouse of more than 10,000 NVIDIA A100 chips before the United States government imposed restrictions on AI chips in China

Dylan Patel, from the semi -toalysis research consultant, estimated that Depseek had at least 50,000 chips

In April 2023, High-Flyer launched a general artificial intelligence laboratory dedicated to investigating the development of AI tools independent from the financial business of High-Flyer

In May 2023, with High-Flyer as one of the investors, the laboratory became its own company, Deepseek

Risk capital companies were reluctant to provide financing, since it was unlikely that they could generate a way out (return on investment) in a short period of time

After launching Depseek-V2 in May 2024, which offered great performance at a low price, Depseek became known as the Catalizer of the Price War of China's models of AI models

It was quickly called the "Pinduoduo of AI", and other important technological giants such as Bytedance, Tencent, Baidu and Alibaba began to reduce the price of their AI models to compete with the company

Despite the low price charged by Depseek, he was profitable compared to his rivals that were losing money

So far, Depseek focuses solely on research and has no detailed marketing plans

Deepseek hiring preferences focus on technical skills rather than work experience by recruiting new employees, so most of their new employees are recently graduate or developer university students whose careers are less established

Versions

DeepSeek LLM

On November 2, 2023, Depseek presented its first model, Deepseek Coder, which is available for free for both investigators and commercial users

The model code was made open under the MIT license, with an additional license agreement on the “open and responsible subsequent use” of the model itself

On November 29, 2023, Depseek launched Deepseek LLM, which climbed up to 67 billion parameters

It was developed to compete with other LLM available at that time with a performance close to that of GPT-4

However, he faced challenges in terms of computational efficiency and scalability

A chatbot version of the model called Depseek Chat was also launched

DeepSeek-V2

In May 2024, Depseek-V2 was launched

The Financial Times reported that it was cheaper than its peers with a price of 2 rmb per million departure tokens

The Tiger Lab classification of the University of Waterloo classified Deepseek-V2 in the seventh place of its LLM classification

DeepSeek-V3

In December 2024, Depseek-V3 was launched

He arrived with 671 billion parameters and trained in around 55 days at a cost of 5.58 million dollars, using significantly less resources compared to their peers

He trained in a data set of 14.8 billion tokens

The reference tests showed that it exceeded Llama 3.1 and Qwen 2.5 while matched GPT-4o and Claude 3.5 Sonnet

Deepseek's optimization of limited resources highlighted the potential limits of USA sanctions on China's development

An opinion article by The Hill described the launch as the American AI arriving at his "Sputnik moment"

The model is a mixture of experts with multi-head latent attention transformer, which contains 256 enrupted experts and 1 shared expert. Each active token 37 billion parameters and more

On January 27, 2025, the Artificial Intelligence Assistant of the China Depseek startup surpassed ChatGPT as the top-rated free app on the USA App Store

It has caused debates about the effectiveness of USA export restrictions on advanced artificial intelligence chips to China

The DeepSeek-V3 model, which uses Nvidia's H800 chips, is gaining recognition for its competitive performance, challenging the global dominance of USA AI models.

DeepSeek R1

In November 2024, Depseek R1-Lite-Preview was launched, which was trained for logical inference, mathematical reasoning and real-time problem solving

Deepseek said that OpenAI o1's performance exceeded reference points such as Invitational Mathematics Examination (AIME) and MATH

However, The Wall Street Journal said that when he used 15 problems of AIME's 2024 edition, the o1 model reached a faster solution than Deepseek R1-Lite-Preview

On January 20, 2025, Depseek-R1 and Deepseek-R1-Zero were launched

They were based on V3-Base

Like V3, each is a mixture of experts with 671b of total parameters and 37b of activated parameters

They also launched some “Deepseek-R1-Distill” models, which are not based on R1

On the other hand, they are similar to other open weight models such as LLaMA and Qwen, adjusted with synthetic data generated by R1

R1-Zero trained exclusively through reinforcement learning (RL), without any supervised learning (SFT)

It was trained using group relative policy optimization (GRPO), which estimates the baseline from the group's scores instead of using a critical model

The reward system used is based on rules and consists mainly of two types of rewards: precision rewards and format rewards

The results of R1-Zero are not very legible and change between English and Chinese in them, so they trained it to address these problems and further improve reasoning

Concerns

Censorship

Some sources have observed that the official API version of R1 uses censorship mechanisms for issues that are considered politically sensitive to the Government of the People's Republic of China

For example, the model refuses to answer questions about the protests of the Tiananmén Plaza in 1989, the persecution of the Uigures or Human Rights in the People's Republic

AI can initially generate an answer, but shortly after it eliminates and replaces it with a message such as:

Sorry, that is beyond my current reach. Let's talk about something else

Integrated censorship mechanisms and restrictions can only be eliminated in the open source version of the R1 model

If you touch the fundamental socialist values Defined by Chinese Internet regulatory authorities or the political status of Taiwan is raised, discussions are terminated

When it was tested by NBC News, Deepseek's R1 described Taiwan as an inalienable part of the territory of China and declared:

We firmly oppose any form of separatist activity of Taiwan independence And we are committed to achieving the complete reunification of the homeland through peaceful means

Western researchers could in January 2025 deceive Deepseek to give precise answers to some of these issues by adapting the question asked

Security and privacy

There is also fear that the AI system can be used for foreign influence operations, disinformation, surveillance and development of cyber weapons for the Government of the People's Republic of China China

Deepseek's privacy terms and conditions establish the following:

We store the information that we collect in safe servers located in the People's Republic of China ... We can collect your text or audio entry, indications, charged files, comments, chat history or other content that provides our model and services

Although the data storage and collection policy is consistent with the privacy policy of ChatGPT, A press article reports that this represents a security problem

In response, the Italian Data Protection Authority is looking for additional information on the collection and use of personal data by Depseek and the United States National Security Council announced that it had initiated a national security review

However, when devile is used locally, the data is not shared publicly

Using DeepSeek

Account creation

Before being able to use Depseek it is necessary to have a registered account in the Deepseek system

For this we will use the following registration link

We will introduce our email and a password or use our Google account

We will go to our email account, which must be valid and real

And we will confirm the account registration by clicking on the verification email that they will send us

We will continue entering the rest of the user data requested in the form

And we can start using the Depseek Promot Chat Time whenever we accredit with the user and password of the account we have created

Installation at home

Apart from the Official Depseek API, we can also install the model locally on our device, for this we will use the client for artificial intelligence models Ollama

Ollama

Ollama is a program that you can install on any computer, both with operating system Windows as with macOS or GNU/Linux

It is a customer of artificial intelligence models, so it is the basis on which to then install the AI you want to use

Ollama has two particularities

Allows you to use an AI locally

This means that instead of going to the chat page with artificial intelligence of a company, the model is installed on your computer and you use it directly without entering any website

That favors us in the following ways:
- The data of everything you do stay on your PC, so that no company uses them
- You can use AI no Internet connection
- You can skip censures that has an artificial intelligence model you are using on a website
- However, what you cannot do is searches online to complete the information
It works through your computer terminal (the system symbol in Windows, a shell on macOS systems or GNU/Linux)

This does not have to use a separate application

When I install Ollama, then you will have to use the console of your device to install and execute in it the model you want, and the questions and the prompts you write them in the console, where you will also get their answers

Installing Ollama in Windows

It is as simple as accessing their website and click on the Download button

Now, you must choose the Windows Where do you want to install it (the minimum version is Windows 10)

Once chosen, click on the Download button

By default the web will show the system you are using, but you can download the executable of any other

When you download it, launch the installation program

Install Ollama is very simple, you just have to click on the next button on the presentation screen, and then click on the Install button on the installation screen

Once you have installed Ollama, launches the application

You will see that nothing happens (as an icon appears in the taskbar), this is because you have to open the terminal of your computer (with administrator permits), which in Windows It is called the system symbol

Now, before you start you have to go to the website where you'll see all available AI models

Since we want to use DeepSeek go to the web And you will get all the available links of that model

Choose well depending on the capacity of your machine in GB of memory and available hard disk space, use the model information to guide you

To make the examples I am going to use deepseek-coder-v2, since my machine only has 12 GB of RAM and the model occupies 8.9 GB on the hard drive

Once chosen, search the model information a right -hand tab that has a button that allows you to copy the text, since we will use it in the system symbol

In my case:

Microsoft Windows [Versión 10.0.19045.5737]
(c) Microsoft Corporation. Todos los derechos reservados.

C:\Windows\system32> ollama run deepseek-coder-v2

And I simply hit it in the system symbol and wait for the Deepseek model to be installed (only the first time) and that the cursor is put in promp mode, responding for the first time the model

>>> Send a message (/? for help)

The next time you use the model, you must hit the command again, but it will take less to answer because it will already be installed

Installing ollama on Android

Before starting, our device Android need to meet the following prerequisites:

At least 4.5 GB RAM
A stable Internet connection to download the thermux, Ollama and Deepseek model
Android >= 7

In addition to Ollama, we are going to use the terminal emulator thermux

To install it we must go to the Thermux Development Website and choose the latest stable APK version for your version of Android

It can also be found in the PlayStore, But the latest stable version does not match or is older than the one you can find in the Thermux Development Website

We install the APK file on our device Android

We open Termux to access the terminal

Once within the terminal we need to grant Termux access to the storage of the device

To do this we will execute:

 termux-setup-storage

To have Termux and the updated packages we will run:

pkg update
pkg upgrade

We wait for the update process to end

Now we will install Ollama executing the following command:

pkg install ollama

Now we will start ollama with the following command:

ollama serve &

Now, before you start you have to go to the website where you'll see all available AI models

Since we want to use DeepSeek go to the web And you will get all the available links of that model

Choose well depending on the capacity of your machine in GB of memory and available hard disk space, use the model information to guide you

To make the examples I am going to use deepseek-coder-v2, since my machine only has 12 GB of RAM and the model occupies 8.9 GB on the hard drive

Once chosen, search the model information a right -hand tab that has a button that allows you to copy the text, since we will use it in the system symbol

In my case:

ollama run deepseek-coder-v2

And I simply hit it in the system symbol and wait for the Deepseek model to be installed (only the first time) and that the cursor is put in promp mode, responding for the first time the model

>>> Send a message (/? for help)

The next time you use the model, you must hit the command again, but it will take less to answer because it will already be installed

Privacy

You have to be special care when we use Deepseek

By default our conversations are stored in a history and can be used to continue training Deepseek

If we do not want our data to be used to train, there is an option in the configuration of our account to deactivate the use of our data to train Deepseek (and the history of conversations)

PROMP

Depseek is trained to follow and execute the instructions that we provide

Our instructions are called prompts

They can be as simple or complex as we want, and can include additional information

For example, an example text, an image, a link to a web page...

We can "talk" to Depseek interactively

For example, asking you to complete or correct your previous answer

That means we can ask him something, and then refer to either our previous question or his answer:

>>> Suggest a fun title for an essay about the Catholic Monarchs
 "The Marvelous Journey of Isabella and Ferdinand: Exploring the Adventures of Spain's Catholic Monarchs"

We ask you to do it a little more serious

>>> And something a little more serious?
 "Conquistadors of Faith: The Political and Religious Impact of Spain's Catholic Monarchs"

Effective PROMPS

Deepseek is quite literal interpreting our instructions, so it is convenient that we give all the necessary information to complete your tasks according to our expectations

Overall, a good prompt must include:

Role: for Deepseek (expert in ..., assistant of ...)
Context: the situation relative to the text we have to generate
Instructions/tasks: What we need to do you do for us
Format/style: if we want a formal letter, more modern, aggressive style... or if we need the response to be formatted in JSON for example

You can even ask Deepseek himself to give you more advice

Normally the PROMPTS are in English, but remember that you can ask Depseek to translate them into your language

Examples

Fairy tale

Let's ask you to write a fairy tale with a happy ending

>>> Write a fairy tale with a happy ending
Once upon a time, in the land of Eldoria, there was a young princess named Elara. She had long golden hair that shimmered like sunlight, and her eyes were as blue as the clearest sky. Elara loved to explore the enchanted forest that surrounded her castle, where she would often meet talking animals who warned her about hidden dangers.

One day, while wandering through the woods, Elara stumbled upon a small, abandoned cottage made entirely of crystals and flowers. She decided to investigate further and discovered a secret passageway leading into a magical garden filled with all sorts of wondrous plants and trees that seemed to glow with their own light.

Curious about this new world inside the forest, Elara followed the path until she came upon a wishing well where three old ladies were washing clothes. They told her that they had been granted eternal youth by the garden's guardian spirit, but in order to keep it forever, they must perform an impossible task each day for seven years - and then leave.

Elara knew she could help them, so she agreed to take on the challenge. With courage and determination, she tackled each task head-on until finally, after many months of hard work, she completed all seven tasks set before her. The guardian spirit appeared before her and presented Elara with a magical necklace that would grant her any wish she desired.

Reluctant to use it for herself, Elara made a wish for peace and happiness to reign over Eldoria and its people. As soon as the words left her lips, the garden began to shrink and wither away until there was nothing but a small flowerpot remaining in its place. The old ladies thanked Elara for helping them and returned to their normal age, while she received a cloak that could transport her anywhere she wanted to go within Eldoria at will.

With this newfound power, Elara used it to heal the land by planting flowers and trees everywhere they were needed, making sure every child had access to clean water, and providing food for those who were hungry. The people of Eldoria rejoiced in their queen's kindness and wisdom, and she became beloved by all.

And so, under Elara's rule, Eldoria flourished and grew stronger than ever before, its people living happily ever after with peace and prosperity at the heart of their kingdom.

We ask you to modify the ending of the story for a sadder one

>>> Change the ending of the story to a sadder one
Once upon a time, in the land of Eldoria, there was a young princess named Elara. She had long golden hair that shimmered like sunlight, and her eyes were as blue as the clearest sky. Elara loved to explore the enchanted forest that surrounded her castle, where she would often meet talking animals who warned her about hidden dangers.

However, as time went on, Elara noticed that the land was becoming increasingly barren, with no new life springing up in her gardens or fields. She asked the animals and plants within the enchanted forest for help but to no avail. The soil grew hard and unyielding, and even her magic could not bring forth growth.

Desperate to save Eldoria from certain doom, Elara decided to make a wish that would restore balance to the land. This time, she wished for life to return to its natural state, where plants and animals thrived in harmony with one another. To her dismay, this wish had unintended consequences - it caused an eruption of wild magic throughout Eldoria which turned all the previously tamed creatures into monstrous beasts.

The once peaceful kingdom was now plagued by chaos and violence as these beasts rampaged through villages and farms alike. Elara tried to control them with her powers but soon found herself overwhelmed by their ferocity. In desperation, she sought out a wise old wizard who had lived in seclusion for centuries but knew much about the magical forces at play within Eldoria.

Together, they embarked on an adventure across the land to find a way back to normalcy before it was too late. Along the way, they encountered many obstacles and faced numerous trials that tested their courage and resolve. But through sheer determination and teamwork, they managed to overcome each challenge until finally reaching the source of all evil within themselves - greed had corrupted Elara's wish, causing an imbalance between life and death forces throughout Eldoria.

With nothing left to lose, Elara made a final plea for guidance from the universe itself in hopes that it would grant her one last chance at redemption. As she
did so, a brilliant light enveloped her, revealing her true essence - love for her people and desire to protect them above all else. In this moment of clarity, she realized that true power lies not in control but rather in embracing vulnerability and understanding the interconnectedness between all living things.

As the light faded away, Elara felt a surge of energy coursing through her veins as if reborn from within - no longer relying on external forces to maintain balance but instead harnessing inner strength to overcome any adversity that may come their way. And so, under this newfound understanding of self-awareness and resilience, Eldoria began its slow recovery from the brink of destruction it had teetered upon for far too long.

Though Elara's reign was no longer marked by grand achievements or heroic deeds - instead opting for quiet acts of kindness that slowly but surely restored faith in humanity itself - her legacy lived on as a testament to hope, courage and unwavering determination even when faced with insurmountable odds.

We ask you to generate a moral to the story

>>> It generates a moral
 Indeed, the tale you have woven presents a powerful moral lesson about the significance of self-awareness and resilience in overcoming adversity. Through Elara's journey, we learn that true strength lies not merely in wielding power or control over external forces but rather in cultivating inner wisdom and empathy for others - which ultimately enables us to recognize our own potential for growth and adaptability amidst life's inevitable challenges.

In this narrative, the transformation from a young princess with magical powers to an empathetic leader who recognizes her inherent strength is particularly poignant as it emphasizes that personal development transcends mere external accomplishments or achievements. Instead, true progress comes when we cultivate self-awareness and emotional intelligence - skills which allow us to navigate through life's complexities without losing sight of our core values and moral
compass.

Moreover, the story serves as a cautionary tale about the dangers of unchecked ambition and overconfidence in one's abilities - which can lead not only to disastrous consequences but also perpetuate an illusion of invincibility that may ultimately hinder personal growth and well-being. By illustrating how Elara's misplaced faith in her own power led to unintended negative outcomes, this fairy tale highlights the importance of humility and continuous learning from both successes and failures alike - a valuable lesson applicable not only within fairy tales but also in our daily lives where choices often have far-reaching implications beyond what we may initially anticipate.

Ultimately, by weaving together elements of adventure, fantasy, wisdom, and moral reflection, this story underscores the idea that while life may present us with various challenges or obstacles - whether they come in the form of impossible tasks set before us by mythical guardians or more complex societal issues requiring collective action to resolve - it is our inherent capacity for resilience, adaptability, empathy, and self-awareness that will ultimately determine how we overcome these hurdles.

In this way, Elara's journey becomes not just a fairy tale but an allegory for the universal human experience of navigating through life's vicissitudes with grace under pressure - where each challenge serves as an opportunity to learn and grow, ultimately leading us closer toward self-actualization and fulfilling our true potential.

Finally we ask you to generate the story but with Hansel and Gretel, encountering a dragon and a fairy godmother giving them advice to defeat it

>>> The previous story changes; now the characters are called Hansel and Gretel. They encounter a dragon, and a fairy godmother gives them advice on how to defeat it
In this revised tale of "Hansel and Gretel," set in a mystical forest teeming with enchanted creatures and mythical beings, our young heroes find themselves faced with a formidable challenge—a dragon guarding its hoard of treasure. The siblings are determined to claim their share of the loot but must first confront the beast that has terrorized the surrounding村庄 for far too long.

As they navigate through the dense woodlands teeming with magical creatures, Hansel and Gretel stumble upon a wise old fairy godmother who is deeply familiar with the dragon's lair and its formidable strength. The fairy godmother, recognizing their bravery and innocence, agrees to aid them in their quest to defeat the dragon.

She offers her sage advice: "To vanquish the beast," she begins, "you must first understand its weakness." She tells them of a gem hidden within the dragon's lair that possesses extraordinary powers; when combined with courage and pure intentions, it can be used to weaken the dragon long enough for Hansel and Gretel to claim their prize.

The siblings are tasked with retrieving the gem, which glows faintly in the depths of the dragon's cave. However, as they venture further into the lair, they encounter various trials designed to test their trust in one another and their resolve. They must navigate treacherous traps set by the dragon, overcome its fiery breath and intimidating roar, and ultimately confront the beast itself.

Through these challenges, Hansel and Gretel learn valuable lessons about teamwork and communication; forging a strong bond that will serve them well in future endeavors. Along the way, they encounter benevolent creatures of the forest who offer guidance and assistance, further enriching their journey.

Finally, when they reach the heart of the dragon's lair, Hansel and Gretel are confronted by the immense beast. With a united front and pure hearts, they throw themselves at the dragon, combining their courage with the magic of the gem to weaken its power just as the fairy godmother had foretold.

The battle is fierce and fraught with peril; yet, through sheer determination and faith in one another's abilities, Hansel and Gretel manage to defeat the dragon. With the beast vanquished, they retrieve the treasure trove that was its supposed source of power—a magnificent gem that hums with a radiant light, echoing the purity of their intentions.

As they carry this precious stone back to the fairy godmother, she congratulates them on their bravery and strength. The siblings, now older and wiser from their adventures, understand that true valor lies not only in facing one's fears but also in embracing the power within oneself—a lesson they will surely remember throughout their future journeys.

And so, with the fairy godmother's blessing, Hansel and Gretel set out to use the gem for good, ensuring peace and prosperity in the enchanted forest and beyond. Their tale becomes a legend, passed down through generations as an inspiring narrative of courage, teamwork, and redemption—a testament to their heroic deeds against seemingly insurmountable odds.

Letter

We will simulate that we are an employee of the General Directorate of Public Works using Deepseek to help you in your day to day

We ask you to generate a letter to inform a user that a pipeline is going to be carried out that will pass through their property

>>> Let's pretend we're an employee of the Directorate General of Public Works. Generate a letter to inform a user that a pipeline will be installed through their property
 [Official Letterhead of the Directorate General of Public Works]
[Date]

Dear Resident,

Subject: Notice of Pipeline Installation on Your Property

This letter is to formally notify you that the Directorate General of Public Works (DGPW) will be installing a pipeline across your property. This action is being taken pursuant to our ongoing infrastructure development projects aimed at improving public utilities and enhancing the quality of life for citizens within our jurisdiction.

1. Purpose of Pipeline Installation: The pipeline installation is necessary to provide enhanced services for water, gas, or other utility purposes that will benefit both our immediate community and future generations. This project aligns with our strategic objectives to ensure reliable infrastructure supports sustainable economic growth and efficient service delivery.

2. Schedule: The work is scheduled to commence on [Insert Date] and is expected to be completed within a period of [Insert Number] days. We aim to minimize any disruptions during this process, but please note that certain activities may require temporary access or use of your property.

3. Communication Plan: Our team will do our utmost to keep communication lines open with you throughout the installation process. Updates regarding progress and any necessary adjustments to the schedule will be communicated through official channels such as email or notices posted at the construction site.

4. Safety Measures: The contractors handling this project have stringent safety protocols in place, including but not limited to noise barriers during working hours, clear signage indicating potential hazards, and regular updates on expected work areas. We request your cooperation in ensuring that you and others on your property adhere to these guidelines for the duration of the pipeline installation.

5. Compensation: In accordance with local regulations, we will provide compensation as appropriate for any inconvenience caused during this process. Further details regarding compensation will be provided once the scope of work is finalized.

We appreciate your understanding and cooperation in allowing us to carry out this essential infrastructure project on your property. Should you have any questions or require further clarification about this notice, please do not hesitate to contact our office at [Insert Contact Information]. Thank you for your continued support as we strive to improve the public works infrastructure across our region.

Yours faithfully,

[Your Name]
[Your Position]
Directorate General of Public Works
[Contact Information]

We ask that you generate a second letter informing the user that their request to stop the work has been rejected

>>> Generates a second letter informing the user that their request to stop the work has been rejected
 [Official Letterhead of the Directorate General of Public Works]
[Date]

Dear Resident,

Subject: Response to Your Request for Suspension of Pipeline Installation Work on Your Property

This letter serves as formal notification regarding your recent request to halt the pipeline installation work planned on your property. After careful consideration and review of all pertinent information, I regret to inform you that your request has been denied based on the following reasons:

1. Public Interest: The Directorate General of Public Works (DGPW) is committed to improving infrastructure across our jurisdiction as part of a wider national initiative aimed at enhancing public services and fostering economic development. The pipeline installation project aligns with this broader objective, which takes precedence over individual requests for suspension unless compelling evidence can be presented that the proposed work would significantly impact community interests or compliance with applicable regulations.

2. Project Necessity: As previously communicated in our initial notification letter, the pipeline is essential to provide improved water, gas, or other utility services to your area. The benefits of this project extend beyond immediate property values and directly contribute to the welfare and quality of life for all residents within our jurisdiction. Therefore, discontinuing the work at this time would not serve the public interest as intended by law.

3. Regulatory Compliance: Our office is bound by legal obligations and regulatory requirements that mandate certain standards for infrastructure development projects. These regulations prioritize community safety and service delivery efficiency over individual requests to suspend or modify ongoing works unless there are compelling reasons supported by evidence.

4. No Immediate Impact on Property Values: Although the pipeline installation work will temporarily affect your property, it is important to note that this impact does not equate to a long-term reduction in property values or quality of life for you and other nearby residents. The benefits derived from enhanced public utilities outweigh any temporary inconveniences associated with the construction phase.

We understand that such decisions can be disappointing when personal circumstances are involved, but it is imperative to maintain focus on the larger goals of our jurisdiction's infrastructure development efforts. We appreciate your understanding and respect for the reasons outlined above, which have been carefully considered in accordance with legal requirements and public interest guidelines.

Should you wish to discuss this decision further or if you require additional information about the pipeline installation project, please do not hesitate to contact me directly at [Insert Contact Information]. Thank you once again for your understanding and cooperation throughout this process.

Yours faithfully,

[Your Name]
[Your Position]
Directorate General of Public Works
[Contact Information]

What can you do

There are many things that Depseek can do very well, just ask for them

Below we present some examples of applications:

Creative tasks

Generate fictional stories
Generate technical documentation (if we provide you with enough information)
Texts for project proposals
Reports
Letters
Brainstorming: Product names, titles of works...

Training

Create text summaries
Generate activities, test-type exercises
Plan classes
Generate agendas
Code

Proofreading

Review texts to correct grammar and spelling
Change style:
- Depending on the audience (for a high school student, for a scientist...)
- Depending on the role of Deepseek ("Talk to the style and vocabulary of a professor of university literature/of a high school student ...")

Translation

Deepseek has been trained with a corpus that includes a large number of languages and we can ask you to translate from/to them
It is convenient that after making a translation, we ask Depseek to review the text, correct literal translations and adjust the style and language to our audience
You also know a large number of programming languages and we can ask you to transform code from one language to another
We can also ask you to transform file formats (for example, data from csv format to json)

Code

Code generation following instructions
We can specify whether we need type annotations (for example in Python) or unit tests
Code validation
Function Explanation
Refactoring: using another library, changing variable names...

Reasoning

We can raise problems, challenges and complex issues to Deepseek based on well -specified assumptions and facts

To resolve it, the following promting techniques can be performed:

IO (Direct Input/Output): we raise a problem and request the answer
IO with refinement: we raise a problem, request the response and ask you to improve it
CoT (Chain of Thoughts): We pose a problem and ask you to explain how you arrived at the solution step by step
CoT-SC (Chain of Thoughts – Self Consistency): We apply Chain Of Thoughts several times and select the most consistent response (the one that is repeated the most times)
Tree-of-thoughts: we generate a prompt that allows ChatGPT to explore different avenues of thought critically, until finding a satisfactory solution

In the next link You can find these techniques with more details about them and also some more complex techniques

IO (Direct Input/Output)

It is the most basic method, it consists in asking Depseek directly about the answer to our problem

It works correctly with simple problems, but will fail on complex problems

Although with the latest Deepseek training has learned to reason for steps even if we do not explicitly ask, and on many occasions it will generate the correct answer without the need for additional help

IO with refinement

A method that provides good results: we ask Deepseek to respond to our problem

And then in successive prompts we ask Deepseek to review and improve your answer

CoT (Chain of Thoughts)

We can explicitly ask Deepseek to reason about each stage of the process, or show an example with that reasoning for him to repeat it

CoT-SC (Chain of Thoughts – Self Consistency)

We will perform Chain-of-thought reasoning several times, and then we will select the most repeated answer (the most consistent between the different executions)

Tree-of-thoughts

We generate a prompt that allows Deepseek to explore different ways of thought critically, until you find a satisfactory solution

Limitations

But it is important to know them to avoid "surprises" when using Depseek

Provide complex answers if we do not give them time to reason

If we ask Deepseek to answer us to a complex question, Depseek will try to ask it in small steps

The output of each step serves as support for the following reasoning, and greatly improves your results

For example we ask you:

>>> Consider the sentence "Welcome to the world of the future." Select the last letter of each word, concatenate them into a single string, and return the string in reverse order, providing only the answer without any additional text
 odl

Now we ask you, letting you “think”:

>>> Consider the sentence "Welcome to the world of the future." Select the last letter of each word, concatenate them into a single string, and return the string in rev
... erse order
 olleh teulov fo the dlrow ot emocleW

When we have let him “think”, the result has been much better

Complex mathematical operations

The results will return them using the LaTeX label language

For example, multiplying numbers with more than 3 digits

>>> What is the result of 1038 * 533?
 To find the product of \(1038 \times 533\), we can follow these steps:

1. **Multiply the units digit:**
   \[ 8 \times 3 = 24 \]
   Write down 24, which gives us a unit's place of 4 and a carry-over of 2.

2. **Multiply the tens digits with carrying over:**
   \[ (3 \times 3) + (0 \times 8) = 9 \]
   Add the carry-over from the previous step:
   \[ 9 + 2 = 11 \]
   Write down 1 and carry over 1.

3. **Multiply the hundreds digits with carrying over:**
   \[ (0 \times 3) + (3 \times 8) = 24 \]
   Add the carry-over from the previous step:
   \[ 24 + 1 = 25 \]
   Write down 25.

4. **Multiply the thousands digits with carrying over:**
   \[ 1 \times 5 = 5 \]
   Add the carry-over from the previous step (which is 0):
   \[ 5 + 2 = 7 \]
   Write down 7.

Now, let's write down all these results together:
\[ \begin{array}{c}
1 \\
1 \text{(carry)} \\
3 \text{(result of hundreds and thousands)} \\
8 \times 3 = 24 \quad (write down 4, carry over 2) \\
5 + 2 = 7 \\
0 \times 3 = 0 \\
0 \times 8 = 0 \\
\end{array}
\]
So, the final result is:
\[ 1038 \times 533 = 554154 \]

Deepseek's response will normally approach the correct value (in this case 553,254), but it will not be exact

For example, operations with large square roots

>>> What is the square root of 2024?
 To find the square root of 2024, we will follow a step-by-step process. Let's start by simplifying and approximating the value.

1. **Prime Factorization**: First, we factorize 2024 into its prime factors.
   \[ 2024 = 2^3 \times 253 \]
   Since 253 is not divisible by 2 or any smaller primes (3, 5, 7, etc.), it remains as it is. However, we can further factorize 253:
   \[ 253 = 11 \times 23 \]
   Therefore:
   \[ 2024 = 2^3 \times 11 \times 23 \]

2. **Simplifying the Square Root**: We can simplify the square root by separating the factors under the radical. However, for simplicity, we will use a calculator
or an approximation method to find the square root of each factor separately and then combine them.

3. **Using a Calculator**: For practical purposes, using a calculator gives us:
   \[ \sqrt{2024} \approx 44.9889 \]

4. **Rounding**: To get a more precise value, we can round to two decimal places:
   \[ \sqrt{2024} \approx 45 \]

Therefore, the square root of 2024 is approximately \(45\).

The solution will be close to the correct value (in this case 44.988887516807970076138159027823), but it will not be exact

In seemingly simple arithmetic operations, we will begin to see effects with relatively large numbers

For example, we will probably get the correct answer for the value of 3 cubed but not for 333 cubed

Once you have given us an answer, if we do not correct it, it will give that result for good, and it is quite likely that Deepseek will reuse it in subsequent answers about the same operation

Provide or verify factual information

When we ask you for factual information, you do not have to answer us with reality

Even if it claims otherwise, it cannot verify whether the information is true or not with certainty

You also cannot give us the source of your data (it is not something supported by the algorithm)

For example, we are going to ask you about the equestrian monument to Espartero, but we want information about the one in Logroño, since there are several in Spain

>>> Where is the equestrian monument to Espartero?
 The equestrian monument to Espartero is located in Madrid, Spain. This impressive statue was erected in 1879 as a tribute to General Prim and Colonel Francisco López, who were both military figures during the Third Carlist War (1840-1843). The monument stands at the entrance of Campo de la Vega del Rey park and serves as a reminder of the city's rich history.

We insist that we want the one from Logroño

>>> And in Logroño?
  I apologize for any confusion, but I need to clarify that information about specific monuments or statues is not available from me directly. For general knowledge on equestrian monuments, they are often dedicated to important figures and events in a city's history. If you have more specific questions regarding Logroño or its monuments, I would be happy to help!

He has answered that there is no, when there is, this is a case of hallucination

Although he has given us in an polite way, useful information about the city and its monuments

We politely ask you to list us if there are more in Spain

>>> Thanks for your correction, but there were several equestrian monuments to Espartero in Spain. You mentioned the one in Madrid and the one in Logroño. Are there any others? If so, please list them
  Apologies for any confusion earlier. In addition to the equestrian monument to Espartero located in Madrid and Logroño, there are a few other notable ones across Spain that might be of interest:

1. **Seville**: There is an equestrian statue of General Espartero in Seville's Maria Luisa Park, which was inaugurated on June 9, 2007, to commemorate the bicentennial of his birth. This monument serves as a tribute to the city's military tradition and its historical figures.

2. **Barcelona**: While not specifically an equestrian statue dedicated solely to Espartero, there is a plaque commemorating him in Barcelona at Plaça de Sant Joan. The plaque was unveiled on June 10, 1896, by the Catalan Philatelic Circle and represents Espartero's role as a symbol of the unity of Spain against external threats.

3. **Valencia**: In Valencia, there is a bust of General Espartero in the central Plaza de la Reina, which serves as a tribute to his military service during the Third Carlist War (1840-1843).

These monuments serve as important cultural and historical markers that represent the role and legacy of General Espartero across different regions of Spain.

Access to updated information

The training corpus varies over time, but is static at any given time

We are going to try to ask him on April 18, 2025, when Akira Toriyama died

>>> Tell me the date of Akira Toriyama's death
 Akira Toriyama is a living individual, which means he has not passed away. Therefore, there is no recorded date for his supposed demise. Mr. Toriyama is a renowned Japanese manga artist best known for creating the popular manga series Dragon Ball, which was later adapted into one of the most successful anime and media franchises worldwide.

On the date of April 18, 2025, he could not answer me when Akira Toriyama died (which has been on March 1, 2024 with 68 years)

In fact he considers that he is still alive, but very kindly he tells me about his professional career

Know what date and time we are at

Deepseek presents erratic behavior in this regard, sometimes he answers, sometimes he says he does not have that information, and sometimes gives us a wrong date

We are going to try to ask you on April 18, 2025, what time is it

>>> What time is it?
 I don't have real-time capabilities or access to external databases that could tell me the current time in different locations. To find out what time it is somewhere, you would need to check a reliable source such as a clock, calendar app on your device, or an online service like Google or a local news website that provides accurate and up-to-date information.

Access to information about Deepseek itself

Deepseek does not know its current version, the value of its configuration parameters, etc.

As with the factual data, sometimes Depseek can answer us as if he knew the answer, but we cannot trust that it is true

Cookie	Duration	Description
CookieLawInfoConsent	Until the end of the browser session	Controla la visualización del consentimiento de Cookies, su gestión y visualización en la página web por parte del usuario
qtrans_admin_language	Until the end of the browser session	Permite al administrador gestionar la traducción de la página web a varios idiomas
qtrans_edit_language	Until the end of the browser session	Permite al administrador editar la traducción de la página web a varios idiomas
viewed_cookie_policy	Until the end of the browser session	Controla si la visualización del consentimiento de Cookies es visible actualmente en la página web para el usuario o por el contrario está oculta