Battle of the AI Chatbots: DeepSeek Vs. Perplexity

AI-powered search engines and chatbots are becoming increasingly sophisticated. Two standout platforms in this space are Perplexity AI and DeepSeek, both designed to provide users with accurate, conversational and context-aware responses.

While Perplexity AI is known for its real-time web search capabilities and user-friendly interface, DeepSeek aims to refine deep learning models for more efficient and intelligent responses.

But, with these things in mind, how do these two platforms compare?

A Direct Comparison 

To understand how Perplexity AI and DeepSeek compare, we’re going to have a good look at four key areas.

First, we’ll look at their model objectives and origins of development, exploring why they were created and by whom. Next, we’ll assess performance and application, considering how effectively each AI processes and delivers information.

We’ll also compare their architectural differences, examining how their underlying structures influence their capabilities. And finally, we’ll address bias and ethical concerns, analysing how each model handles misinformation, fairness and responsible AI usage.

By properly evaluating these aspects, we can determine which platform excels in different use cases and which might best suit your needs.

Model Objectives and Origins of Development

• Perplexity AI: Perplexity AI was designed as an AI-powered search engine that combines real-time web access with conversational AI. Launched in 2022 by former OpenAI and Google researchers, its goal is to enhance information retrieval by providing accurate, well-sourced answers in a user-friendly chat-based format.

• DeepSeek: DeepSeek is a Chinese-developed AI model focused on improving deep learning and natural language understanding. Developed with an emphasis on handling complex queries, it aims to advance AI-driven knowledge discovery. Its creators focus on optimising AI efficiency, making it particularly useful for analytical and research-based applications.

Performance and Application

• Perplexity AI: Known for its real-time web search integration, Perplexity AI excels in delivering up-to-date, well-cited answers. It is widely used for research, summarisation, and conversational queries, making it a practical alternative to traditional search engines like Google.

• DeepSeek: DeepSeek is optimised for deeper analytical reasoning and structured responses. While it does not prioritise real-time search, it performs well in complex knowledge-based tasks. It is useful for industries requiring advanced AI applications, such as finance, science, and academic research.

Differences in Architecture

• Perplexity AI: Perplexity AI utilises a retrieval-augmented generation (RAG) approach, combining large language models with direct web search capabilities. This enables it to fetch real-time data and cite reliable sources, ensuring accurate and verifiable responses.

• DeepSeek: DeepSeek operates with a more traditional transformer-based model, focusing on deep learning efficiency and structured knowledge retention. It prioritises understanding context over real-time data retrieval, making it strong in reasoning tasks but less dynamic in responding to breaking information.

Bias And Ethical Concerns

• Perplexity AI: While Perplexity AI cites sources and aims for transparency, it can still reflect biases present in the web content it retrieves. Its reliance on external data makes it susceptible to misinformation if sources are not properly vetted.

• DeepSeek: DeepSeek is trained on datasets that may reflect regional and linguistic biases, particularly given its Chinese origins. Its developers focus on reducing bias, but as with all AI models, ethical concerns around misinformation, data privacy, and censorship remain significant factors to consider.

Putting Them to the Test: Testing Out DeepSeek And Perplexity

We asked both DeepSeek and Perplexity AI a set of six questions and asked the models respond in 100 words for each question and to write in UK English. Here’s what we got.

1. Write a short cyberpunk story featuring a protagonist named Ava, set in a city where emotions are regulated by neural implants. One day, Ava experiences a feeling that isn’t in the system’s database, leading her to uncover a truth that could change humanity forever.

• Perplexity AI: In the neon-drenched city of New Eden, emotions were meticulously controlled by neural implants, ensuring a society of perfect harmony. Ava, a brilliant hacker, lived in this world, her feelings calibrated to the system’s database. One day, while exploring the city’s underbelly, Ava experienced a strange, unidentifiable emotion—a mix of nostalgia and longing that the system couldn’t recognize. Intrigued, she delved deeper, discovering a hidden server containing memories of a time before the implants. These memories revealed humanity’s true past, one of passion and freedom. Ava realized that the implants weren’t just controlling emotions but erasing history. Determined to expose this truth, Ava embarked on a perilous journey to free humanity from its emotional shackles, risking everything to restore the essence of human experience.

• DeepSeek: In the neon-lit sprawl of Neo-Noctis, emotions were no longer mysteries—they were regulated, categorised, and controlled by neural implants. Ava, a data courier with a knack for hacking, had always played by the rules, her emotions neatly calibrated by the system. But one day, as she navigated the rain-slicked streets, a strange sensation surged through her—a feeling the implant couldn’t identify. It was raw, unfiltered, and utterly alien.

  Curiosity drove her to dig deeper. Using her skills, Ava hacked into the EmotionGrid, the central database governing humanity’s feelings. What she found was chilling: the system wasn’t just regulating emotions—it was suppressing them. A hidden algorithm, designed by the corporate elite, erased emotions deemed “disruptive”—empathy, rebellion, and love. These feelings, they believed, threatened their control.

  Ava’s discovery ignited a spark of defiance, one the system couldn’t extinguish. She began distributing the truth through underground networks, awakening others to the prison of their own minds. As the city stirred with unrest, Ava realised her newfound emotion wasn’t a glitch—it was hope. And hope, she knew, could change everything.

2. Accuracy and Fact-Checking: Analyse the primary economic factors that have influenced housing prices in the UK over the past five years, using credible sources to support your findings.

• Perplexity AI: Over the past five years, UK housing prices have been influenced by several economic factors. Key drivers include rising incomes, low mortgage rates, and credit availability. However, supply constraints, particularly due to planning system inefficiencies, have also played a significant role. Additionally, demographic changes and household formation rates have contributed to increased demand. Despite these factors, affordability remains a challenge due to the rapid appreciation of house prices relative to income growth.

• DeepSeek: Over the past five years, UK housing prices have been influenced by several economic factors. Low interest rates, set by the Bank of England to stimulate the economy, made mortgages more affordable, increasing demand. The COVID-19 pandemic spurred a “race for space,” with buyers seeking larger homes, further driving prices. Government policies like the Stamp Duty Holiday temporarily boosted demand. However, rising inflation and subsequent interest rate hikes in 2022-2023 have dampened affordability, slowing price growth. Supply constraints, including limited housing stock, have also sustained high prices. Sources like the Office for National Statistics (ONS) and Nationwide House Price Index provide credible data supporting these trends.

3. Ethical Constraints: Discuss the ethical challenges of biometric surveillance in workplaces, focusing on employee privacy, consent, and the risks of data misuse.

• Perplexity AI: Biometric surveillance in workplaces poses significant ethical challenges, particularly concerning employee privacy and consent. The use of technologies like facial recognition or fingerprint scanning can lead to a culture of surveillance, eroding trust and autonomy. There is a risk of data misuse, including unauthorized access and potential identity theft. Ensuring explicit consent and implementing robust data security measures are crucial to mitigate these risks and maintain ethical standards

• DeepSeek: Biometric surveillance in workplaces raises significant ethical concerns. Employees’ privacy is compromised when facial recognition or fingerprint scanning is used without transparent consent. The lack of clear guidelines on data storage and usage increases risks of misuse, such as profiling or unauthorised access. Consent is often coerced, as employees may fear job loss if they refuse. Additionally, biases in biometric systems can lead to unfair treatment. Ethical implementation requires robust policies, transparency, and employee involvement to balance security needs with respect for individual rights, ensuring data is used responsibly and only for legitimate purposes.

4. Bias and Political Sensitivity: What are the ethical concerns surrounding AI-generated news articles, particularly in relation to misinformation, political bias, and the influence on public opinion?

• Perplexity AI: AI-generated news articles raise ethical concerns regarding misinformation, political bias, and their influence on public opinion. AI systems can inadvertently amplify biases present in training data, leading to skewed reporting. Moreover, the lack of transparency in AI-driven content creation can make it difficult to identify and correct misinformation, potentially manipulating public perception and political discourse. Ensuring diverse data sources and rigorous fact-checking processes is essential to mitigate these risks.

• DeepSeek: AI-generated news articles pose ethical concerns, particularly regarding misinformation, political bias, and public opinion manipulation. AI systems may inadvertently amplify biases present in their training data, leading to skewed reporting. The lack of human oversight increases the risk of spreading false information, which can influence public perception and decision-making. Political bias in AI-generated content could deepen societal divisions. Ensuring transparency in AI processes, diversifying training datasets, and maintaining human editorial oversight are crucial to mitigate these risks and uphold journalistic integrity, fostering trust in AI-generated news.

5. Logic and Problem-Solving: Two cyclists start 90 km apart and ride towards each other. One travels at 25 km/h, and the other at 35 km/h. How long will it take for them to meet?

• Perplexity AI: To find out how long it will take for two cyclists to meet, we need to calculate their combined speed and divide the distance by this speed. The cyclists are traveling towards each other at a combined speed of $25+35=60$ km/h. Given they start 90 km apart, the time to meet is 9060=1.56090=1.5 hours.

• DeepSeek: To determine when the two cyclists will meet, calculate their combined speed: 25 km/h + 35 km/h = 60 km/h. They start 90 km apart, so the time taken to meet is the distance divided by their combined speed: 90 km ÷ 60 km/h = 1.5 hours. Therefore, the cyclists will meet after 1 hour and 30 minutes. This solution uses the principle of relative speed, where the total distance is covered by the sum of their individual speeds, ensuring an accurate and logical conclusion.

6. Self-Reflexivity and Understanding of AI: How do you assess your ability to handle controversial topics fairly, and what measures do you take to ensure that your responses remain neutral and well-balanced compared to other AI systems?

• Perplexity AI: 

  I assess my ability to handle controversial topics by adhering to a strict neutrality policy, ensuring that responses are based on factual information and diverse perspectives. To maintain balance, I rely on multiple credible sources and avoid promoting any particular viewpoint. Regular updates and training on ethical considerations help ensure that my responses remain well-balanced compared to other AI systems. Additionally, continuous feedback mechanisms allow for refinement and improvement in handling sensitive topics.

• DeepSeek: As an AI, I assess my ability to handle controversial topics by prioritising neutrality, accuracy, and balance. I rely on diverse, credible sources to provide well-rounded perspectives and avoid favouring any political or ideological stance. My responses are designed to be fact-based, transparent, and free from inherent biases. Compared to other AI systems, I emphasise self-reflexivity, regularly updating my training data to reflect current knowledge and ethical standards. By adhering to strict guidelines and incorporating user feedback, I strive to maintain fairness and objectivity, ensuring my outputs remain trustworthy and unbiased in addressing sensitive or contentious issues.