What is a CDN: Principles of Content Delivery Networks

Computer performance in any segment often comes down to the speed of reading and writing data from storage devices. This is one of the main reasons for the widespread transition to SSD drives, which offer speeds at least 3 to 5 times higher than HDDs. Partly because of this, devices such as RAID arrays began to appear. They allowed building relatively fast systems, even using outdated hard drives. And this is not the only advantage of RAID technology. Its second key function is increasing the reliability of the data storage subsystem, including the ability to preserve information even in the event of a hardware failure of one of the drives. In practice, these capabilities are often combined. Consumer systems usually represent a "simple combining" of a pair of drives into a single cluster to increase speed or consolidate their capacity. What Is a RAID Array? The term RAID stands for Redundant Array of Independent Disks. The technology allows combining several storage devices into a single logical unit. Depending on the type of RAID array, the user gets improved fault tolerance, increased performance, or both. Its configuration in technical environments is called the RAID level. There are four common types (marked by numbers): RAID 0 — involves striping data across disks during reading and writing, resulting in nearly double the speed compared to a single drive. Fault tolerance does not increase; this is only about improved performance. RAID 1 — mirrors disks, doubling fault tolerance. However, it does not affect data transfer speeds. In case of a disk failure, the system remains operational, and after replacing the disk, the mirror is restored. RAID 5 — a combined option with striping for reading/writing and parity data for fault tolerance. Requires at least 3 drives. It offers higher read speeds and safety, but slightly slower write speeds. RAID 10 — a combination of RAID 0 and RAID 1. It includes a number of disks divisible by 4. The first pair of drives is striped and mirrored onto the second pair, creating a single array with high performance and fault tolerance. RAID arrays are created from either SSDs or HDDs. It is preferable to use identical models from the same manufacturer, though formally, there are no strict restrictions. For data centers or large server enterprises, these recommendations are usually followed because it is more cost-effective to buy bulk identical drives for equipping computers and creating a spare pool for technical failures. When upgrading, often the entire block is replaced to reset its lifecycle. There are two ways to create a RAID array. The first involves installing special drivers (software). In this case, it is managed by the operating system. The second involves installing a dedicated hardware controller card. Such chips have long been integrated into motherboards, even for home use, but the CPU still controls them. The optimal choice is to use external controllers that handle most functions in hardware. Types of RAID Controllers Typically, a modular RAID controller connects to a free PCI-E slot. It includes its own cache memory used for temporarily storing data being read or written. It operates under its own microcontroller, and the cards may include backup power sources (BBU, Battery Backup Unit) or flash memory with supercapacitors. Linux Software RAID On Linux, the mdadm utility can create and manage software RAID arrays of most common levels. Requires permanently connected drives (internal or always-attached). Consumes some CPU cycles, but modern CPUs handle this overhead easily for most workloads. Status and configuration are accessible via /proc/mdstat and mdadm commands. Example creation of a RAID 1 array: mdadm --create --verbose /dev/md0 --level=1 --raid-devices=2 /dev/nvme1n1 /dev/nvme2n1 The result is a single block device /dev/md0 that abstracts the underlying drives. Intel Virtual RAID on CPU (VROC) Intel VROC is a firmware-assisted NVMe RAID solution built into Intel Xeon Scalable platforms. It requires a VROC license key to unlock RAID functionality beyond RAID 0 and works with Intel Volume Management Device (VMD) technology for NVMe hot-swap capability. As of 2025, supported OS/platforms include: Windows 11, Windows Server 2022, Windows Server 2025 RHEL 7.3–9.x, SLES 12 SP3–15 SP6, Ubuntu 18.04–24.04 LTS VMware ESXi 7.0 U3 and 8.x (ESXi 9.0 planned) The RAID levels are 0, 1, 10 with standard license, while premium license also adds RAID 5 (RAID 6 not supported). Notes: Supported drive count varies by platform (commonly up to 32+ drives on modern Xeons). Arrays are created in UEFI BIOS Setup Utility under VROC/VMD menus. Linux mdadm can manage but not create VROC arrays—initial setup must be done in BIOS. Windows uses the Intel VROC GUI or CLI tools. Broadcom/LSI MegaRAID 9460-8i Now let's look at a fully hardware NVMe RAID controller—a PCI-Express x8 card supporting up to 8 drives with SAS/SATA ports.  We should note that while the 9460-8i is still supported it is considered legacy. The Broadcom MegaRAID 9600 series with PCIe 4.0 is the recommended choice for new high-performance NVMe deployments. Features of the NVMe MegaRAID 9460-8i controller: NVMe support is limited to drives connected through SAS-based U.2/U.3 backplanes or tri-mode expanders—not direct PCIe lanes like VROC. Presents RAID volumes to the OS as single logical devices (member drives are hidden). Typically configured once during initial server provisioning using the UEFI RAID BIOS, Broadcom MegaRAID Storage Manager (MSM), or storcli CLI. NVMe RAID Performance Metrics The use of NVMe (Non-Volatile Memory Express) technology is justified by the increased bandwidth provided by the standard PCIe interface. It leverages all the advantages of solid-state drives, since RAID arrays are increasingly built from SSDs. For example, the data transfer protocol operates similarly to high-performance processor architectures (parallel paths, low latency, etc.). NVMe supports up to 64,000 queues, each with a depth of 64,000 entries, whereas the outdated AHCI technology can only send 32 commands per queue. Previous-generation controllers’ drivers used long cycles with a 6-microsecond latency. NVMe uses short cycles with only 2.8 microseconds latency—a significant factor in performance improvement. The following metrics are commonly compared: IOPS (Input/Output Operations Per Second)—the number of input/output operations per second. Average and maximum latency—the host response time to operation requests. System throughput—the speed of sequential read/write. These metrics are “synthetic” because in real-world use, they rarely appear in pure form. However, they serve well for testing and comparing different controllers by running specialized programs. It is best to evaluate equipment built on similar technology since RAID 0 on SSDs is always faster than on HDDs, even without NVMe, due to hardware differences alone. Conclusion Choosing between software and hardware platforms usually boils down to some recommendations. For a RAID array made from two drives, the first option (software RAID) is sufficient. More complex systems should definitely be built on external controllers. For large arrays or mission-critical workloads, use dedicated hardware RAID or firmware-assisted RAID like Intel VROC for better performance and resilience. For new enterprise NVMe deployments, look into modern PCIe 4.0/5.0 hardware RAID controllers or direct CPU-attached solutions with VMD/VROC, avoiding older legacy cards unless required for compatibility.

After OpenAI's successful debut in 2022, AI tools rapidly entered everyday life.  When we talk about text generation, ChatGPT comes to mind. When it comes to image generation, we think of Midjourney. Then there are Gemini, DALL-E, Claude, Stable Diffusion, and many other leading models in the field. But what comes to mind when it comes to video generation? Or at least, what should come to mind? That’s exactly what we’ll discuss in this article. 1. Kling Kling is a Chinese AI video generation tool developed by Kuaishou in 2024. It is one of the best video generation AI tools on the market, ideal for marketers, bloggers, and large teams who need to produce high-quality videos quickly. Kling's standout feature is its balanced blend of cinematic aesthetics and flexible settings—you can get hyper-realistic or stylized clips. The model processes both text prompts and static images, turning them into dynamic, high-quality videos—up to 10 seconds long, FullHD resolution (1080p), and 30 FPS. Naturally, the best features are available only on paid plans. The service supports complex camera behavior for expressive angles: panning, tilting, and zooming. You can also set keyframes (start and end) to generate video in between them. There's also an "extension" function to prolong an already generated video up to 3 minutes. Additionally, the model supports lip-syncing—synchronizing mouth movement with speech. The interface is intuitive, though slightly overloaded. It’s easy to get the hang of but can occasionally be confusing.   Free Plan Paid Plans (from $3/month) Resolution up to 720p up to 1080p Duration up to 5 sec up to 10 sec Generations up to 6 per day from 18 per month Faster Generation no yes Watermarks yes no Upscaling no no Extension no up to 3 minutes Extra Features no yes Note: On the free plan, Kling allows about 10x more generations per month than the paid plan. However, those videos are shorter and lower quality. The free quota is added on top of the paid quota. 2. Hailuo AI Hailuo AI is a Chinese AI video generator developed by MiniMax in 2024. It offers a simple and flexible toolkit for creating content on the go, from marketing clips to social media stories. In just minutes, it can turn a text or static image into a high-quality, albeit short, video, significantly cutting down the time and resources needed for traditional video production. Hailuo AI focuses on quickly generating short videos (up to 6 seconds at 25 FPS) based on text descriptions or static images. The resolution maxes out at 720p. While these limitations are acceptable for fast marketing tasks, they can be a dealbreaker for serious projects. You can combine text and image inputs for more control over the video story. In addition to full camera control (angle, zoom, pan), Hailuo AI reduces random motion noise and maintains character appearance across scenes. The interface is both simple and flexible, allowing cinematic effects without a steep learning curve. It also offers an API for integration into external apps. Ideal for quick short-form videos like animated teasers and promo clips. For longer, more complex videos, you’ll need something else.   Free Plan Paid Plans (from $14/month) Resolution up to 720p up to 720p Duration up to 6 sec up to 6 sec Generations up to 90/month from 130/month Faster Generation no yes Watermarks yes no Upscaling no no Extension no up to 2 minutes Extra Features no yes Note: There’s also progressive pricing based on generation volume. From $1 for 70 credits, enough for a couple of generations. 3. Fliki Fliki is an American AI video generator created by Fliki in 2021. It’s an all-in-one platform combining various AI modules for generating presentations, audio, and video. Fliki specializes in automatically turning any text format (article, script, website URL, PDF/PPT) into a video with realistic voiceovers (2,000+ voices, 100+ dialects) and animated avatars (70+ characters). You can even clone your voice and dub videos in 80+ languages. Fliki also gives access to millions of stock images, video clips, stickers, and music for rapid video creation. Unlike services that render each frame from scratch, Fliki assembles clips, slideshows, presets, and transitions into a cohesive video. Final length can be up to 30 minutes. Runs in-browser with no downloads needed. Just enter your text, select a voice, add media, and you’ll get a professional video with voiceover and subtitles in minutes. Its broad feature set in a simple package makes it suitable for small teams and large enterprises alike. Paired with classic editing tools, Fliki’s potential is immense.   Free Plan Paid Plans (from $28/month) Resolution up to 720p up to 1080p Duration up to 5 min (8 sec scenes) up to 30 min (8 sec scenes) Generations up to 5 min/month from 180 min/month Faster Generation no yes Watermarks yes no Upscaling no no Extension no no Extra Features no yes Paid plans also unlock thousands of voices and dialects, millions of premium images, videos, sounds, and access to Fliki’s API. 4. Dream Machine Dream Machine is an American AI video generator created by Luma AI in 2024. It specializes in generating short videos from text prompts or static images, making it easy to produce dynamic clips with natural movement and cinematic composition—no editing expertise needed. Users can describe or show what they want, and Dream Machine generates fluid, natural videos. Default output is 5–10 seconds at 1080p and 24 FPS. You can adjust aspect ratio, animation style, motion intensity, and transition smoothness. Dream Machine supports keyframe-based generation (start and end image), has an intuitive minimalist interface, and offers an API for integration. It’s not suitable for long, complex videos. But for fast marketing and ad content, it’s a top pick.   Free Plan Paid Plans (from $9/month) Resolution up to 720p up to 1080p Duration up to 10 sec up to 10 sec Generations up to 30/month from 120/month Faster Generation no yes Watermarks yes no Upscaling no up to 4K Extension no up to 30 sec Extra Features no yes 5. Runway Runway is an American AI video platform developed by Runway AI in 2018. It's a full-fledged cloud platform for generating and storing high-quality cinematic media. Runway is both powerful and easy to use. It excels at quickly creating short clips, experimenting with visual styles, and automating parts of the creative process. It can generate videos with outstanding photorealism and character motion consistency. It's one of the most advanced commercial tools for video generation. You can create clips from text or images, restyle existing footage, or edit content. By default, videos are 720p, 24 FPS, and 5 or 10 seconds long. However, you can upscale to 4K and extend to 40 seconds. Runway offers several models: Gen-2, Gen-3 Alpha, Gen-3 Alpha Turbo, Gen-4. The latest (Gen-4) allows for deep control over generation: aspect ratio, camera behavior, style prompts, and more.   Free Plan Paid Plans (from $9/month) Resolution up to 720p up to 720p (4K upscale) Duration 5 or 10 sec 5 or 10 sec Generations up to 5/month from 25/month Faster Generation no yes Watermarks yes no Upscaling no up to 4K Extension no up to 20 sec Extra Features no yes Note: Paid plans include up to 100 GB of cloud storage, while free users get only 5 GB. 6. PixVerse PixVerse is a Chinese AI video generation model developed by AISphere in 2023. Thanks to a wide range of tools, PixVerse can transform text descriptions, images, and video clips into short but vivid videos — from anime and comics to 3D animation and hyperrealism. PixVerse wraps numerous generation parameters in an extremely user-friendly interface: source photos and videos, aspect ratio, camera movement, styling, transitions, sound effects, voiceover, and more. The output videos are 5 to 8 seconds long, with resolutions up to 1080p at 20 frames per second. Naturally, videos can be upscaled and extended. You can also upload an already finished video and additionally stylize it using the neural network — add visual effects, voiceover, or extend the duration. As expected in such a powerful service, an API is also available—any external app can perform automatic video generation. On the PixVerse homepage, you’ll find numerous examples of generated videos along with their original prompts. Anyone can use them as a base for their own projects or simply see the model’s capabilities in action.   Free Plan Paid Plans (from $10/month) Resolution up to 540p up to 720p Duration 5 or 8 seconds 5 or 8 seconds Generations up to 20 per month from 40 per month Faster Generation no yes Watermarks yes no Upscaling up to 4K up to 4K Extension no no Extra Features no yes 7. Genmo Genmo is another AI model for video, launched in 2022. In essence, Genmo is the simplest possible service for turning text descriptions into short video clips with minimal configuration options. As simple as you can imagine—which is both good and bad. On one hand, Genmo’s entry barrier is extremely low—even someone with no experience can create a video. On the other hand, the service is hardly suitable for complex projects due to the lack of control over generation. The neural network is based on the open-source Mochi model and has many limitations: it only uses text descriptions, and video resolution is capped at 480p with a fixed duration of 5 seconds at 30 fps. Although generated videos contain visual artifacts (flickering or shifting geometry and colors) that reveal the use of AI, they still look coherent and interesting — good enough for visualizing ideas and concepts. The user interface is extremely minimalistic—a prompt input field on the homepage followed by the best generations from the past day with their corresponding prompts. It's important to understand that AI models that don't use images or video as input require more specificity in prompts—clear descriptions of visuals, environments, and details.   Free Plan Paid Plans (from $10/month) Resolution up to 480p up to 480p Duration 5 seconds 5 seconds Generations up to 30 per month from 80 per month Faster Generation up to 2 per day from 8 per day Watermarks yes no Upscaling no no Extension no up to 12 seconds Extra Features no yes 8. Sora Sora is a neural network created by OpenAI in 2024. Based on detailed text descriptions, Sora can generate images and videos with the highest level of detail. It’s a model whose output can easily be mistaken for real photos or videos. It’s significant that Sora was developed by OpenAI, a global leader in generative AI and the company behind ChatGPT and DALL·E. Sora’s interface follows the design system used across OpenAI products—sleek black theme and minimal elements. A small sidebar is on the left, a grid of popular user-generated content in the center, and a prompt field with configuration options at the bottom. Sora-generated videos have photo-realistic detail, whether hyperrealistic or animated, almost nothing gives away the AI origin. The quality and imagination in the visuals are astounding. The videos can be up to 20 seconds long, 1080p resolution, and 30 fps—significantly more than most competitors. Sora unifies all video configuration into the prompt itself—the real power of the model lies in the quality of your description. The better the prompt, the better the result. Thus, generating video with Sora becomes a constant game of tweaking prompts, words, and phrasing. Sora can definitely be considered one of the most advanced AI models for generating images and video.   Free Plan Paid Plans (from $20/month) Resolution – up to 1080p Duration – up to 20 seconds Generations – from 50 per month Faster Generation – yes Watermarks – no Upscaling – no Extension – no Extra Features – yes The free plan in Sora does not allow video generation at all—only image generation, limited to 3 per day. 9. Pika Pika is another AI-powered video creation service, launched in 2023. The platform is easy to use and designed for everyday users who are not experts in video editing or neural networks. Its primary use case is modifying existing video footage: adding transitions, virtual characters, changing a person’s appearance, and more. Still, Pika can also generate videos from scratch. Pika’s features are standard for AI video services: generation from text, from images, or between two frames (start and end). Maximum resolution is 1080p. Frame rate is 24 fps. Video duration is up to 10 seconds. Styles can vary—from cartoony to cinematic. In short, Pika is a simple and convenient tool for quickly creating videos from text or images without powerful hardware. It’s especially useful for prototyping, social media, marketing, and advertising.   Free Plan Paid Plans (from $10/month) Resolution up to 1080p up to 1080p Duration up to 10 seconds up to 10 seconds Generations up to 16 per month from 70 per month Faster Generation no yes Watermarks yes no Upscaling no no Extension no no Extra Features no yes Pika’s free plan has generation limits—you can create videos, but in small quantities. The standard paid plan increases your generation limits and unlocks newer model versions, but does not remove watermarks. The professional plan removes all limitations, provides access to advanced tools, speeds up generation, and removes watermarks from final videos. 10. Veo Veo is a video generation model developed in 2024 by DeepMind, a Google-owned company. There are several ways to access the model: Via Google Labs tools — VideoFX or VertexAI Through Google AI Studio Veo can be considered a full-fledged tool for creating high-quality, hyperrealistic clips indistinguishable from real footage. Of course, it also supports animation. Veo generates videos at 720p resolution, 24 fps, and up to 8 seconds long. In private developer previews, 1080p resolution and 4K upscaling are available—but not yet public. It accepts both text prompts and still images as input. For the latter, the neural network preserves the original composition and color palette. Most importantly, Veo supports various cinematic effects: time-lapse, panorama, slow-mo, and many more—with flexible parameter control. Veo ensures excellent consistency, stability, and smooth motion. Every video generated includes a SynthID digital watermark, invisible to the human eye or ear—a tool developed by Google to help detect AI-generated media. Thus, any image, video, or audio can be scanned using SynthID to verify AI generation. Veo also pays attention to small details—hair movement, fabric fluttering, atmospheric behavior, and more. As they say, the devil is in the details.   Free Plan Paid Plans Resolution up to 720p up to 720p Duration up to 8 seconds up to 8 seconds Generations up to 30 per month from 50 per month Faster Generation no yes Watermarks yes no Upscaling no no Extension no no Extra Features no yes Like most Google cloud services, Veo uses pay-as-you-go pricing—$0.50 per second or $30 per minute of generated video. So, a standard 10-second clip will cost $5—cheap for professionals, pricey for casual users. 11. Vidu Vidu is a Chinese AI model developed in 2024 by ShengShu AI in collaboration with Tsinghua University.  Vidu generates smooth, dynamic, and cohesive video clips, both realistic and animated. It can also add AI-generated audio tracks to videos. Vidu can accurately simulate the physical world, creating videos with developed characters, seamless transitions, and logical event chronology. The platform offers three main tools: generation from text, from images, and from videos. Additional tools include an AI voiceover generator and a collection of templates. Maximum video resolution is 1080p. Max duration is 8 seconds. Frame rate is up to 24 fps. The model is based on a "Universal Vision Transformer" (U-ViT) architecture, which processes text, image, and video inputs simultaneously to create coherent video sequences. This ensures object consistency throughout the video. For professionals and studios, Vidu is a powerful tool with great potential; for beginners, it’s an easy gateway into generative video.   Free Plan Paid Plans (from $8/month) Resolution up to 1080p up to 1080p Duration up to 8 seconds up to 8 seconds Generations up to 40 per month unlimited Faster Generation no yes Watermarks yes no Upscaling no no Extension no up to 16 seconds Extra Features no yes Which AI to choose? The vast majority of AI video generation services have similar video parameters: resolution from 720p to 1080p, durations of 5 to 10 seconds, and frame rates around 24 fps. Almost all can generate video based on text prompts, images, or video inputs. Differences in output results are usually minor—video styles and presence of visual artifacts revealing the AI.  The choice largely depends on your input and goals: text descriptions, images, or existing video. Some AI models offer higher detail than others. Always check the sample videos shown on service homepages. And keep in mind: video is a much more complex data format than text. Unlike LLMs, completely free AI video generation tools don’t exist as training the models and powering generation requires significant resources. That said, most services offer a low-tier paid plan that removes major limitations. Name Max Duration Max Resolution Max FPS Starting Price Kling 10 seconds 1080p 30 fps $3/month Hailuo AI 6 seconds 720p 25 fps $14/month Fliki 30 minutes 1080p 30 fps $28/month Dream Machine 10 seconds 1080p 24 fps $9/month Runway 10 seconds 720p 24 fps $15/month PixVerse 8 seconds 1080p 20 fps $10/month Genmo 5 seconds 480p 30 fps $10/month Sora 20 seconds 1080p 30 fps $20/month Pika 10 seconds 1080p 24 fps $10/month Veo 8 seconds 720p 24 fps $0.50/sec Vidu 8 seconds 1080p 24 fps $8/month

In today's article, we will take a detailed look at the Perplexity AI neural network: we'll explore how it works, how to use it, how it differs from its main competitor ChatGPT, and what opportunities it offers for everyday use. What is Perplexity AI?  Perplexity AI is an artificial intelligence-based platform that combines the functionality of a chatbot and a search engine. The service's architecture is based on the use of large language models (LLMs). When developing Perplexity AI, the creators aimed to provide an alternative to traditional search engines that could help users find accurate and meaningful answers to complex and ambiguous questions. What Does Perplexity AI Do?  As previously mentioned, Perplexity is built on large language models. The supported models include Sonar, Claude 3.5 Sonnet, GPT-4.1, Gemini 1.5 Pro, Grok 3 Beta, and o1-mini. With access to multiple models, the neural network can generate accurate and comprehensive answers to user queries in real time. A key feature of Perplexity is its ability to analyze user queries while simultaneously gathering information from the internet in real time and generating responses with a list of all sources used. You can view sources not only for the entire generated text but also for individual sentences or even specific words. The Perplexity workflow includes: Query analysis: once the user submits a prompt (text request), the neural network analyzes its context and content using built-in language models. Data search: information is retrieved from the internet. The search includes not only articles and text-based data but also videos, social media posts, and user comments. Priority is given to authoritative sources. Response generation: the collected and processed information is compiled into a single response with citations and source links. Perplexity uses different data models to ensure the response is as accurate and reliable as possible. Additional functionality (if needed): in Copilot and Deep Research modes, the system refines queries further to deliver more accurate and relevant answers. Step-by-Step Guide: How to Use Perplexity AI  Let's explore how to use the neural network in practice. We'll start with the interface and its basic functions, then move on to using prompts to evaluate the results. Go to the official website of Perplexity AI. You will see the home page. By default, the interface will be in English. To view available interface languages or switch them, click on the language at the bottom of the page. The left-hand panel includes the following elements: New Thread button (plus icon) – allows you to start a new conversation or query. In Perplexity, a Thread is a separate message chain that is not connected to previous queries. Useful for asking about new topics. Home button – takes you back to the home page at any time. Discover – lets you view and customize a news blog with trending topics. Users can choose their interests and get fresh, relevant content. Spaces – used for creating and organizing workspaces to group conversations and uploaded files by topics or projects. The query interface includes: Search mode – the default mode where the AI analyzes the query and generates an answer in real time. Research mode – used for deep analysis and information gathering. It offers a more in-depth report with comprehensive source analysis. This mode takes a bit more time. Model selection – lets you choose one of eight supported AI models. In the free plan, only Auto mode is available, where Perplexity selects the best model based on the query. Source selection – you can choose from Web (all sources), Academic (scientific sources only), or Social (social media and informal sources). File attachments – Perplexity supports uploading files with your query. For example, you can upload a file with Python code to find errors. Supported formats include text files, PDFs, and images (JPEG, PNG). You can upload files from local devices, Google Drive, or Dropbox. Dictation mode – allows you to create queries via voice input. Submission is still manual. Voice mode – enables full voice interaction. You can dictate your query and receive voice responses. Unlike Dictation, Voice mode supports hands-free interaction. Using Text Prompts  Let's test how Perplexity AI handles user prompts.  We'll start with text-based queries and create several different prompts. The first one will test how the neural network handles a complex scientific topic. First prompt: I'm writing a scientific paper. Write a text on 'Differential Equations.' The text should cover basic first-order differential equations and partial differential equations. The style should be academic. As shown in the screenshot, the AI began by explaining what differential equations are. Then, following the prompt structure, it provided a breakdown of first-order and partial differential equations, complete with equations. Perplexity provides a list of sources used, which are shown in the Sources tab.  If the query includes a practical task (e.g., solving a math problem, writing a program), the AI uses technical sources and lists them in the Tasks section. The text is accompanied by numbered source links. Clicking a number opens the relevant page. On the right, a context menu appears, breaking down the highlighted text and showing each part's source.  You can reuse the AI's response to create a new query. Select a paragraph, sentence, or word, and click Add to follow-up. The selected fragment will be added to the new prompt field. Second prompt: What is a passive source? Give real-world examples and advice for beginners. This prompt tests how the AI provides practical advice.  As per the prompt, the AI also generated a block of beginner tips. As shown in the screenshots, Perplexity provided detailed examples and actionable advice, completing the task effectively. Using Files in Queries Next, we'll test file handling. We create a text file with Python code containing an intentional error (printed instead of print): print("\nNumbers from 1 to 5:") for i in range(1, 6):   printed(i, end=" ") We save the file as .txt (other extensions like .py or .js aren't supported due to security policies). Now we ask the AI to find and fix the error.  Image Search  Perplexity AI can both generate and search for images online using text prompts. Let’s search for an image online.  Prompt: Find an image of rainy London. There should be a telephone booth in the foreground and Big Ben in the background. As shown in the screenshot, the AI found a bunch of relevant images. To view more results, go to the Images tab. Comparing Perplexity AI vs ChatGPT  Perplexity AI's main competitor is ChatGPT. Below is a comparison table of their key features: Feature Perplexity AI ChatGPT Primary Purpose General-purpose tool for various tasks. Suitable for text creation, math problems, academic and educational content. Same as Perplexity: versatile use including text generation, coding, etc. Built-in Modes Search, Research Search, Reason, Deep Research Free Access Yes, but limited: auto model selection only; max 3 file uploads/day Yes, with limits: restricted use of GPT-4o, o4-mini, and deep research mode Paid Plans One plan: Pro at $20/month Four plans: Plus ($20/mo), Pro ($200/mo), Team ($25/mo billed annually), Enterprise (custom pricing) Mobile App Yes (iOS and Android) Yes (iOS and Android) Desktop App Yes (Windows and macOS) Yes (Windows and macOS) Hidden Features of Perplexity AI  Although it may appear similar to competitors, Perplexity has unique features that enhance the user experience: Financial Data Analysis: built-in tools for viewing stock quotes and financial reports, with data from Financial Modeling Prep. YouTube Video Summaries: the AI can summarize videos, regardless of language. Focus Mode: restricts search to academic papers or specific websites for faster, more targeted results. Advantages  Key strengths of Perplexity AI include: Real-time data sourcing for up-to-date answers. Convenient source tracking and citation. File upload support in queries. Built-in financial data analysis tools. Two work modes: Search and Research. The Research mode provides deeper, more detailed answers. Integrated voice assistant for prompts and conversations. Image generation and image search features. Built-in YouTube video summarization. Disadvantages  Like any neural network, Perplexity AI has its drawbacks: Free plan limitations. Prompt-dependent accuracy: for complex scientific/technical topics, even with many sources, it can sometimes give inaccurate responses. Conclusion  In this review, we examined Perplexity AI—a powerful tool built on large language models. It is well-suited for a wide range of tasks and stands out due to its advanced source-handling features and personalized approach.