Tag: AI Slop

The ‘AI Slop’ Crisis: 21% of YouTube Recommendations are Now AI-Generated

In a startling revelation that has sent shockwaves through the digital creator economy, a landmark study released in late 2025 has confirmed that "AI Slop"—low-quality, synthetic content—now accounts for a staggering 21% of the recommendations served to new users on YouTube. The report, titled the "AI Slop Report: The Global Rise of Low-Quality AI Videos," was published by the video-editing platform Kapwing and details a rapidly deteriorating landscape where human-made content is being systematically crowded out by automated "view-farming" operations.

The immediate significance of this development cannot be overstated. For the first time, data suggests that more than one-fifth of the "front door" of the world’s largest video platform is no longer human. This surge in synthetic content is not merely an aesthetic nuisance; it represents a fundamental shift in the internet’s unit economics. As AI-generated "slop" becomes cheaper to produce than the electricity required to watch it, the financial viability of human creators is being called into question, leading to what researchers describe as an "algorithmic race to the bottom" that threatens the very fabric of digital trust and authenticity.

The Industrialization of "Brainrot": Technical Mechanics of the Slop Economy

The Kapwing study, which utilized a "cold start" methodology by simulating 500 new, unpersonalized accounts, found that 104 of the first 500 videos recommended were fully AI-generated. Beyond the 21% "slop" figure, an additional 33% of recommendations were classified as "brainrot"—nonsensical, repetitive content designed solely to trigger dopamine responses in the YouTube Shorts feed. The technical sophistication of these operations has evolved from simple text-to-speech overlays to fully automated "content manufacturing" pipelines. These pipelines utilize tools like OpenAI's Sora and Kling 2.1 for high-fidelity, albeit nonsensical, visuals, paired with ElevenLabs for synthetic narration and Shotstack for programmatic video editing.

Unlike previous eras of "spam" content, which were often easy to filter via metadata or low resolution, 2026-era slop is high-definition and visually stimulating. These videos often feature "ultra-realistic" but logic-defying scenarios, such as the Indian channel Bandar Apna Dost, which the report identifies as the world’s most-viewed slop channel with over 2.4 billion views. By using AI to animate static images into 10-second loops, "sloppers" can manage dozens of channels simultaneously through automation platforms like Make.com, which wire together trend detection, script generation via GPT-4o, and automated uploading.

Initial reactions from the AI research community have been scathing. AI critic Gary Marcus described the phenomenon as "perhaps the most wasteful use of a computer ever devised," arguing that the massive computational power required to generate "meaningless talking cats" provides zero human value while consuming immense energy. Similarly, researcher Timnit Gebru linked the crisis to the "Stochastic Parrots" theory, noting that the rise of slop represents a "knowledge collapse" where the internet becomes a closed loop of AI-generated noise, alienating users and degrading the quality of public information.

The Economic Imbalance: Alphabet Inc. and the Threat to Human Creators

The rise of AI slop has created a crisis of "Negative Unit Economics for Humans." Because AI content costs nearly zero to produce at scale, it can achieve massive profitability even with low CPMs (cost per mille). The Kapwing report identified 278 channels that post exclusively AI slop, collectively amassing 63 billion views and an estimated $117 million in annual ad revenue. This creates a competitive environment where human creators, who must invest time, talent, and capital into their work, cannot economically compete with the sheer volume of synthetic output.

For Alphabet Inc. (NASDAQ: GOOGL), the parent company of YouTube, this development is a double-edged sword. While the high engagement metrics of "brainrot" content may boost short-term ad inventory, the long-term strategic risks are substantial. Major advertisers are increasingly wary of "brand safety," expressing concern that their products are being marketed alongside decontextualized, addictive sludge. This has prompted a "Slop Economy" debate, where platforms must decide whether to prioritize raw engagement or curate for quality.

The competitive implications extend to other tech giants as well. Meta Platforms (NASDAQ: META) and TikTok (owned by ByteDance) are facing similar pressures, as their recommendation algorithms are equally susceptible to "algorithmic pollution." If YouTube becomes synonymous with low-quality synthetic content, it risks a mass exodus of its most valuable asset: its human creator community. Startups are already emerging to capitalize on this frustration, offering "Human-Only" content filters and decentralized platforms that prioritize verified human identity over raw view counts.

Algorithmic Pollution and the "Dead Internet" Reality

The broader significance of the 21% slop threshold lies in its validation of the "Dead Internet Theory"—the once-fringe idea that the majority of internet activity and content is now generated by bots rather than humans. This "algorithmic pollution" means that recommendation systems, which were designed to surface the most relevant content, are now being "gamed" by synthetic entities that understand the algorithm's preferences better than humans do. Because these systems prioritize watch time and "curiosity-gap" clicks, they naturally gravitate toward the high-frequency, high-stimulation nature of AI-generated videos.

This trend mirrors previous AI milestones, such as the 2023 explosion of large language models, but with a more destructive twist. While LLMs were initially seen as tools for productivity, the 2026 slop crisis suggests that their primary use case in the attention economy has become the mass-production of "filler." This has profound implications for society, as the "front door" of information for younger generations—who increasingly use YouTube and TikTok as primary search engines—is now heavily distorted by synthetic hallucinations and engagement-farming tactics.

Potential concerns regarding "information hygiene" are also at the forefront. Researchers warn that as AI slop becomes indistinguishable from authentic content, the "cost of truth" will rise. Users may lose agency in their digital lives, finding themselves trapped in "slop loops" that offer no educational or cultural value. This erosion of trust could lead to a broader cultural backlash against generative AI, as the public begins to associate the technology not with innovation, but with the degradation of their digital experiences.

The Road Ahead: Detection, Regulation, and "Human-Made" Labels

Looking toward the future, the "Slop Crisis" is expected to trigger a wave of new regulations and platform policies. Experts predict that YouTube will be forced to implement more aggressive "Repetitious Content" policies and introduce mandatory "Human-Made" watermarks for content that wishes to remain eligible for premium ad revenue. Near-term developments may include the integration of "Slop Evader" tools—third-party browser extensions and AI-powered filters that allow users to hide synthetic content from their feeds.

However, the challenge of detection remains a technical arms race. As generative models like OpenAI's Sora continue to improve, the "synthetic markers" currently used by researchers to identify slop—such as robotic narration or distorted background textures—will eventually disappear. This will require platforms to move toward "Proof of Personhood" systems, where creators must verify their identity through biometric or blockchain-based methods to be prioritized in the algorithm.

In the long term, the crisis may lead to a bifurcation of the internet. We may see the emergence of "Premium Human Webs," where content is gated and curated, existing alongside a "Public Slop Web" that is free but entirely synthetic. What happens next will depend largely on whether platforms like YouTube decide that their primary responsibility is to their shareholders' short-term engagement metrics or to the long-term health of the human creative ecosystem.

A Turning Point for the Digital Age

The Kapwing "AI Slop Report" serves as a definitive marker in the history of artificial intelligence, signaling the end of the "experimentation phase" and the beginning of the "industrialization phase" of synthetic content. The fact that 21% of recommendations are now AI-generated is a wake-up call for platforms, regulators, and users alike. It highlights the urgent need for a new framework of digital ethics that accounts for the near-zero cost of AI production and the inherent value of human creativity.

The key takeaway is that the internet's current unit economics are fundamentally broken. When a "slopper" can earn $4 million a year by automating an AI monkey, while a human documentarian struggles to break even, the platform has ceased to be a marketplace of ideas and has become a factory of noise. In the coming weeks and months, all eyes will be on YouTube’s leadership to see if they will implement the "Human-First" policies that many in the industry are now demanding. The survival of the creator economy as we know it may depend on it.

This content is intended for informational purposes only and represents analysis of current AI developments.

TokenRing AI delivers enterprise-grade solutions for multi-agent AI workflow orchestration, AI-powered development tools, and seamless remote collaboration platforms.
For more information, visit https://www.tokenring.ai/.

January 5, 2026
The Great Digital Decay: New 2025 Report Warns ‘AI Slop’ Now Comprises Over Half of the Internet

As of December 29, 2025, the digital landscape has reached a grim milestone. A comprehensive year-end report from content creation firm Kapwing, titled the AI Slop Report 2025, reveals that the "Dead Internet Theory"—once a fringe conspiracy—has effectively become an observable reality. The report warns that low-quality, mass-produced synthetic content, colloquially known as "AI slop," now accounts for more than 52% of all newly published English-language articles and a staggering 21% of all short-form video recommendations on major platforms.

This degradation is not merely a nuisance for users; it represents a fundamental shift in how information is consumed and distributed. With Merriam-Webster officially naming "Slop" its 2025 Word of the Year, the phenomenon has moved from the shadows of bot farms into the mainstream strategies of tech giants. The report highlights a growing "authenticity crisis" that threatens to permanently erode the trust users place in digital platforms, as human creativity is increasingly drowned out by high-volume, low-value algorithmic noise.

The Industrialization of Slop: Technical Specifications and the 'Slopper' Pipeline

The explosion of AI slop in late 2025 is driven by the maturation of multimodal models and the "democratization" of industrial-scale automation tools. Leading the charge is OpenAI’s Sora 2, which launched a dedicated social integration earlier this year. While designed for high-end creativity, its "Cameo" feature—which allows users to insert their likeness into hyper-realistic scenes—has been co-opted by "sloppers" to generate thousands of fake influencers. Similarly, Meta Platforms Inc. (NASDAQ:META) introduced "Meta Vibes," a feature within its AI suite that encourages users to "remix" and re-generate clips, creating a feedback loop of slightly altered, repetitive synthetic media.

Technically, the "Slopper" economy relies on sophisticated content pipelines that require almost zero human intervention. These systems utilize LLM-based scripts to scrape trending topics from X and Reddit Inc. (NYSE:RDDT), generate scripts, and feed them into video APIs like Google’s Nano Banana Pro (part of the Gemini 3 ecosystem). The result is a flood of "brainrot" content—nonsensical, high-stimulation clips often featuring bizarre imagery like "Shrimp Jesus" or hyper-realistic, yet factually impossible, historical events—designed specifically to hijack the engagement algorithms of TikTok and YouTube.

This approach differs significantly from previous years, where AI content was often easy to spot due to visual "hallucinations" or poor grammar. By late 2025, the technical fidelity of slop has improved to the point where it is visually indistinguishable from mid-tier human production, though it remains intellectually hollow. Industry experts from the Nielsen Norman Group note that while the quality of the pixels has improved, the quality of the information has plummeted, leading to a "zombie apocalypse" of content that offers visual stimulation without substance.

The Corporate Divide: Meta’s Integration vs. YouTube’s Enforcement

The rise of AI slop has forced a strategic schism among tech giants. Meta Platforms Inc. (NASDAQ:META) has taken a controversial stance; during an October 2025 earnings call, CEO Mark Zuckerberg indicated that the company would continue to integrate a "huge corpus" of AI-generated content into its recommendation systems. Meta views synthetic media as a cost-effective way to keep feeds "fresh" and maintain high watch times, even if the content is not human-authored. This positioning has turned Meta's platforms into the primary host for the "Slopper" economy, which Kapwing estimates generated $117 million in ad revenue for top-tier bot-run channels this year alone.

In contrast, Alphabet Inc. (NASDAQ:GOOGL) has struggled to police its video giant, YouTube. Despite updating policies in July 2025 to demonetize "mass-produced, repetitive" content, the platform remains saturated. The Kapwing report found that 33% of YouTube Shorts served to new accounts fall into the "brainrot" category. While Google (NASDAQ:GOOGL) has introduced "Slop Filters" that allow users to opt out of AI-heavy recommendations, the economic incentive for creators to use AI tools remains too strong to ignore.

This shift has created a competitive advantage for platforms that prioritize human verification. Reddit Inc. (NYSE:RDDT) and LinkedIn, owned by Microsoft (NASDAQ:MSFT), have seen a resurgence in user trust by implementing stricter "Human-Only" zones and verified contributor badges. However, the sheer volume of AI content makes manual moderation nearly impossible, forcing these companies to develop their own "AI-detecting AI," which researchers warn is an escalating and expensive arms race.

Model Collapse and the Death of the Open Web

Beyond the user experience, the wider significance of the slop epidemic lies in its impact on the future of AI itself. Researchers at the University of Amsterdam and Oxford have published alarming findings on "Model Collapse"—a phenomenon where new AI models are trained on the synthetic "refuse" of their predecessors. As AI slop becomes the dominant data source on the internet, future models like GPT-5 or Gemini 4 risk becoming "inbred," losing the ability to generate factual information or diverse creative thought because they are learning from low-quality, AI-generated hallucinations.

This digital pollution has also triggered what sociologists call "authenticity fatigue." As users become unable to trust any visual or text found on the open web, there is a mass migration toward "dark social"—private, invite-only communities on Discord or WhatsApp where human identity can be verified. This trend marks a potential end to the era of the "Global Village," as the open internet becomes a toxic landfill of synthetic noise, pushing human discourse into walled gardens.

Comparisons are being drawn to the environmental crisis of the 20th century. Just as plastic pollution degraded the physical oceans, AI slop is viewed as the "digital plastic" of the 21st century. Unlike previous AI milestones, such as the launch of ChatGPT in 2022 which was seen as a tool for empowerment, the 2025 slop crisis is viewed as a systemic failure of the attention economy, where the pursuit of engagement has prioritized quantity over the very survival of truth.

The Horizon: Slop Filters and Verified Reality

Looking ahead to 2026, experts predict a surge in "Verification-as-a-Service" (VaaS). Near-term developments will likely include the widespread adoption of the C2PA standard—a digital "nutrition label" for content that proves its origin. We expect to see more platforms follow the lead of Pinterest (NYSE:PINS) and Wikipedia, the latter of which took the drastic step in late 2025 of suspending its AI-summary features to protect its knowledge base from "irreversible harm."

The challenge remains one of economics. As long as AI slop remains cheaper to produce than human content and continues to trigger algorithmic engagement, the "Slopper" economy will thrive. The next phase of this battle will be fought in the browser and the OS, with companies like Apple (NASDAQ:AAPL) and Microsoft (NASDAQ:MSFT) potentially integrating "Humanity Filters" directly into the hardware level to help users navigate a world where "seeing is no longer believing."

A Tipping Point for the Digital Age

The Kapwing AI Slop Report 2025 serves as a definitive warning that the internet has reached a tipping point. The key takeaway is clear: the volume of synthetic content has outpaced our ability to filter it, leading to a structural degradation of the web. This development will likely be remembered as the moment the "Open Web" died, replaced by a fractured landscape of AI-saturated public squares and verified private enclaves.

In the coming weeks, eyes will be on the European Union and the U.S. FTC, as regulators consider new "Digital Litter" laws that could hold platforms financially responsible for the proliferation of non-disclosed AI content. For now, the burden remains on the user to navigate an increasingly hallucinatory digital world. The 2025 slop crisis isn't just a technical glitch—it's a fundamental challenge to the nature of human connection in the age of automation.

This content is intended for informational purposes only and represents analysis of current AI developments.

TokenRing AI delivers enterprise-grade solutions for multi-agent AI workflow orchestration, AI-powered development tools, and seamless remote collaboration platforms.
For more information, visit https://www.tokenring.ai/.

December 29, 2025
Journalists Unite Against ‘AI Slop’: Safeguarding Truth and Trust in the Age of Algorithms

New York, NY – December 1, 2025 – As artificial intelligence rapidly integrates into newsrooms worldwide, a growing chorus of unionized journalists is sounding the alarm, raising profound concerns about the technology's impact on journalistic integrity, job security, and the very essence of truth. At the heart of their apprehension is the specter of "AI slop"—low-quality, often inaccurate, and ethically dubious content generated by algorithms—threatening to erode public trust and undermine the foundational principles of news.

This burgeoning movement among media professionals underscores a critical juncture for the industry. While AI promises unprecedented efficiencies, journalists and their unions are demanding robust safeguards, transparency, and human oversight to prevent a race to the bottom in content quality and to protect the vital role of human-led reporting in a democratic society. Their collective voice highlights the urgent need for a balanced approach, one that harnesses AI's potential without sacrificing the ethical standards and professional judgment that define quality journalism.

The Algorithmic Shift: AI's Footprint in Newsrooms and the Rise of "Slop"

The integration of AI into journalism has been swift and pervasive, transforming various facets of the news production cycle. Newsrooms now deploy AI for tasks ranging from automated content generation to sophisticated data analysis and audience engagement. For instance, The Associated Press (NASDAQ: AP) utilizes AI to automate thousands of routine financial reports quarterly, a volume unattainable by human writers alone. Similarly, German publication EXPRESS.de employs an advanced AI system, Klara Indernach (KI), for structuring texts and research on predictable topics like sports. Beyond basic reporting, AI-powered tools like Google's (NASDAQ: GOOGL) Pinpoint and Fact Check Explorer assist investigative journalists in sifting through vast document collections and verifying information.

Technically, modern generative AI, particularly large language models (LLMs) like OpenAI's (Private Company, backed by Microsoft (NASDAQ: MSFT)) GPT-4 and Google's Gemini, can produce coherent and fluent text, generate images, and even create audio content. These models operate by recognizing statistical patterns in massive datasets, allowing for rapid content creation. However, this capability fundamentally diverges from traditional journalistic practices. While AI offers unparalleled speed and scalability, human journalism prioritizes critical thinking, investigative depth, nuanced storytelling, and, crucially, verification through multiple human sources. AI, operating on prediction rather than verification, can "hallucinate" falsehoods or amplify biases present in its training data, leading to the "AI slop" that unionized journalists fear. This low-quality, often unverified content directly threatens the core journalistic values of accuracy and accountability, lacking the human judgment, empathy, and ethical considerations essential for public service.

Initial reactions from the journalistic community are a mix of cautious optimism and deep concern. Many acknowledge AI's potential for efficiency but express significant apprehension about accuracy, bias, and the ethical dilemmas surrounding transparency and intellectual property. The NewsGuild-CWA, for example, has launched its "News, Not Slop" campaign, emphasizing that "journalism for humans is led by humans." Instances of AI-generated stories containing factual errors or even plagiarism, such as those reported at CNET, underscore these anxieties, reinforcing the call for robust human oversight and a clear distinction between AI-assisted and human-generated content.

Navigating the New Landscape: AI Companies, Tech Giants, and the Future of News

The accelerating adoption of AI in journalism presents a complex competitive landscape for AI companies, tech giants, and startups. Major players like Google, OpenAI (backed by Microsoft), and even emerging firms like Mistral are actively developing and deploying AI tools for news organizations. Google's Journalist Studio, with tools like Pinpoint and Fact Check Explorer, and its Gemini chatbot partnerships, position it as a significant enabler for newsrooms. OpenAI's collaborations with the American Journalism Project (AJP) and The Associated Press, licensing vast news archives to train its models, highlight a strategic move to integrate deeply into the news ecosystem.

However, the growing concerns about "AI slop" and the increasing calls for regulation are poised to disrupt this landscape. Companies that prioritize ethical AI development, transparency, and fair compensation for intellectual property will likely gain a significant competitive advantage. Conversely, those perceived as contributing to the "slop" problem or infringing on copyrights face reputational damage and legal challenges. Publishers are increasingly pursuing legal action for copyright infringement, while others are negotiating licensing agreements to ensure fair use of their content for AI training.

This shift could benefit specialized AI verification and detection firms, as the need to identify AI-generated misinformation becomes paramount. Larger, well-resourced news organizations, with the capacity to invest in sophisticated AI tools and navigate complex legal frameworks, also stand to gain. They can leverage AI for efficiency while maintaining high journalistic standards. Smaller, under-resourced news outlets, however, risk being left behind, unable to compete on efficiency or content personalization without significant external support. The proliferation of AI-enhanced search features that provide direct summaries could also reduce referral traffic to news websites, disrupting traditional advertising and subscription revenue models and further entrenching the control of tech giants over information distribution. Ultimately, the market will likely favor AI solutions that augment human journalists rather than replace them, with a strong emphasis on accountability and quality.

Broader Implications: Trust, Misinformation, and the Evolving AI Frontier

Unionized journalists' concerns about AI in journalism resonate deeply within the broader AI landscape and ongoing trends in content creation. Their push for human-centered AI, transparency, and intellectual property protection mirrors similar movements across creative industries, from film and television to music and literature. In journalism, however, these issues carry additional weight due to the profession's critical role in informing the public and upholding democratic values.

The potential for AI to generate and disseminate misinformation at an unprecedented scale is perhaps the most significant concern. Advanced generative AI makes it alarmingly easy to create hyper-realistic fake news, images, audio, and deepfakes that are difficult to distinguish from authentic content. This capability fundamentally undermines truth verification and public trust in the media. The inherent unreliability of AI models, which can "hallucinate" or invent facts, directly contradicts journalism's core values of accuracy and verification. The rapid proliferation of "AI slop" threatens to drown out professionally reported news, making it increasingly difficult for the public to discern credible information from synthetic content.

Comparing this to previous AI milestones reveals a stark difference. Early AI, like ELIZA in the 1960s, offered rudimentary conversational abilities. Later advancements, such as Generative Adversarial Networks (GANs) in 2014, enabled the creation of realistic images. However, the current era of large language models, propelled by the Transformer architecture (2017) and popularized by tools like ChatGPT (2022) and DALL-E 2 (2022), represents a paradigm shift. These models can create novel, complex, and high-quality content across various modalities that often requires significant effort to distinguish from human-made content. This unprecedented capability amplifies the urgency of journalists' concerns, as the direct potential for job displacement and the rapid proliferation of sophisticated synthetic media are far greater than with earlier AI technologies. The fight against "AI slop" is therefore not just about job security, but about safeguarding the very fabric of an informed society.

The Road Ahead: Regulation, Adaptation, and the Human Element

The future of AI in journalism is poised for significant near-term and long-term developments, driven by both technological advancements and an increasing push for regulatory action. In the near term, AI will continue to optimize newsroom workflows, automating routine tasks like summarization, basic reporting, and content personalization. However, the emphasis will increasingly shift towards human oversight, with journalists acting as "prompt engineers" and critical editors of AI-generated output.

Longer-term, expect more sophisticated AI-powered investigative tools, capable of deeper data analysis and identifying complex narratives. AI could also facilitate hyper-personalized news experiences, although this raises concerns about filter bubbles and echo chambers. The potential for AI-driven news platforms and immersive storytelling using VR/AR technologies is also on the horizon.

Regulatory actions are gaining momentum globally. The European Union's AI Act, adopted in 2024, is a landmark framework mandating transparency for generative AI and disclosure obligations for synthetic content. Similar legislative efforts are underway in the U.S. and other nations, with a focus on intellectual property rights, data transparency, and accountability for AI-generated misinformation. Industry guidelines, like those adopted by The Associated Press and The New York Times (NYSE: NYT), will also continue to evolve, emphasizing human review, ethical use, and clear disclosure of AI involvement.

The role of journalists will undoubtedly evolve, not diminish. Experts predict a future where AI serves as a powerful assistant, freeing human reporters to focus on core journalistic skills: critical thinking, ethical judgment, in-depth investigation, source cultivation, and compelling storytelling that AI cannot replicate. Journalists will need to become "hybrid professionals," adept at leveraging AI tools while upholding the highest standards of accuracy and integrity. Challenges remain, particularly concerning AI's propensity for "hallucinations," algorithmic bias, and the opaque nature of some AI systems. The economic impact on news business models, especially those reliant on search traffic, also needs to be addressed through fair compensation for content used to train AI. Ultimately, the survival and thriving of journalism in the AI era will depend on its ability to navigate this complex technological landscape, championing transparency, accuracy, and the enduring power of human storytelling in an age of algorithms.

Conclusion: A Defining Moment for Journalism

The concerns voiced by unionized journalists regarding artificial intelligence and "AI slop" represent a defining moment for the news industry. This isn't merely a debate about technology; it's a fundamental reckoning with the ethical, professional, and economic challenges posed by algorithms in the pursuit of truth. The rise of sophisticated generative AI has brought into sharp focus the irreplaceable value of human judgment, empathy, and integrity in reporting.

The significance of this development cannot be overstated. As AI continues to evolve, the battle against low-quality, AI-generated content becomes crucial for preserving public trust in media. The collective efforts of journalists and their unions to establish guardrails—through contract negotiations, advocacy for robust regulation, and the development of ethical guidelines—are vital for ensuring that AI serves as a tool to enhance, rather than undermine, the public service mission of journalism.

In the coming weeks and months, watch for continued legislative discussions around AI governance, further developments in intellectual property disputes, and the emergence of innovative solutions that marry AI's efficiency with human journalistic excellence. The future of journalism will hinge on its ability to navigate this complex technological landscape, championing transparency, accuracy, and the enduring power of human storytelling in an age of algorithms.

This content is intended for informational purposes only and represents analysis of current AI developments.

TokenRing AI delivers enterprise-grade solutions for multi-agent AI workflow orchestration, AI-powered development tools, and seamless remote collaboration platforms.
For more information, visit https://www.tokenring.ai/.

December 1, 2025