Welcome to #43

In this issue:

AI Apps Shake-Up: Which Tools Rose, Fell, and Vanished in 6 Months?

Why Your Company’s Biggest Threat Might Be AI

Vibe Coding: Can Talking Replace Typing Code?

Is Diffusion Dead?

Plus, we discuss the potential repercussions of curing disease with AI, the power and efficiency of using AI for police case work, and all the news we found interesting this week.

It’s Sunday morning!

AI’s been busy learning everything. You just have to read this email.

Seems fair.

Let’s get into it,

The DAS Crew - Andy, Beth, Brian, Eran, Jyunmi, and Karl

AI Apps Shake-Up: Which Tools Rose, Fell, and Vanished in 6 Months?

Andreessen Horowitz recently released their fourth edition of the "Top 100 Gen AI Consumer Apps" report, highlighting dramatic shifts in the AI landscape over just the last six months. Some notable newcomers like Deep Seek skyrocketed into prominence, while former favorites such as Midjourney and Leonardo saw significant declines in popularity. Perhaps most surprising: nearly half of the apps featured six months ago have disappeared completely.

Chinese AI platforms like Deep Seek, Doubao, and Kling experienced rapid growth, showing China's increasing influence in the AI space. Interestingly, conversational AI apps and multimodal tools that combine several capabilities, such as text, images, and video, gained substantial traction. Conversely, specialized image-generation tools seem to be losing momentum, likely due to saturation and competition from multifunctional platforms.

The report also highlighted that consumer spending remains strong in practical AI applications such as photo and video editing tools, despite the widespread availability of free alternatives. Additionally, apps offering conversational companionship, like Character AI, continue to hold top spots, pointing to broader societal shifts towards personalized AI interactions.

WHY IT MATTERS

Rapid Shifts in AI Preferences: The rapid disappearance and emergence of AI tools underscore the volatility of consumer preferences, highlighting the need for businesses to remain adaptable.

Rise of Chinese AI Platforms: China's growing challenge to the leaders in AI, mostly U.S. companies, signals a major shift, emphasizing the importance of global strategy for Western businesses seeking to stay competitive. China is competing and taking share, even in the face of U.S. sanctions on advanced GPU sales to China.

Multimodal Dominance: Tools that integrate multiple data modes—video, images, text and audio—as part of their training, enable their models to accept inputs and outputs across multiple modes. Applications that do are winning over users, suggesting specialized single-purpose tools like text-only chat may struggle to compete.

Conversational AI Popularity: The sustained popularity of conversational and companion-style AI apps reflects deeper cultural shifts toward digital companionship and personalized interactions, particularly among younger generations.

Paying for Practicality: Despite the prevalence of free tools, consumers are increasingly willing to pay for practical, high-quality editing and productivity-focused applications, pointing toward monetization strategies businesses should consider.

Why Your Company’s Biggest Threat Might Be AI

Businesses aren't just competing on product features or pricing anymore. They're now battling an invisible force reshaping entire markets: AI-driven disruption.

Consider Chegg, once a disruptor itself by pioneering textbook rentals and online learning, now struggling as students turn to generative AI tutor apps like Khan Academy’s "Khanmigo." Similarly, WebMD’s specialized health content is losing users to AI-powered conversational interfaces like ChatGPT and Perplexity, which offer instant personalized answers, bypassing traditional information hubs altogether.

Even industries beyond information delivery, like banking and creative media, aren't safe. Fintech companies powered by AI are challenging traditional banks with faster, more personalized services. At the same time, established creative industries face disruption from user-generated, AI-assisted content, cutting out traditional agencies, studios and middlemen.

Companies face critical decisions: pivot swiftly, rethink their business models, or risk becoming irrelevant in an AI-driven world.

WHY IT MATTERS

Product-Market Fit is Fragile: Companies once dominating their markets can see rapid collapse as AI provides consumers with better, cheaper, or faster alternatives.

Rapid Adaptation is Critical: Firms must be agile and quickly pivot to integrate AI, or risk being overtaken by new entrants already built around AI-driven solutions.

Changing User Expectations: Consumers now demand instant, personalized interactions. Businesses that fail to meet these new standards risk significant user losses.

Shifting Industry Power Dynamics: Smaller, agile AI-first startups are now serious threats to established giants, highlighting the importance of continuous innovation.

Employee Awareness is Key: Workers should evaluate their company's long-term viability in an AI-driven market, staying aware of disruption trends is essential for career security.

Vibe Coding: Can Talking Replace Typing Code?

"Vibe coding", a term coined by AI expert Andrej Karpathy, promises to make software development accessible to everyone—no coding experience required. Instead of writing code, you simply describe what you want in natural language, and AI tools generate your app in real-time. Platforms like Lovable, Replit, and Cursor already enable users to visually build and refine software through continuous conversations with AI assistants.

Yet, despite the excitement, vibe coding isn't without its challenges. As easy as it sounds, there's still a significant skill gap between seasoned developers and total beginners. Common pitfalls include security vulnerabilities, inefficient architectures, and problems scaling from prototypes to full-scale applications. Experienced developers highlight the importance of understanding at least basic coding principles or being able to read code to effectively manage these AI-generated projects.

As vibe coding improves, it could significantly reduce barriers for businesses and creators to test new ideas quickly and cheaply. However, the question remains whether it can evolve enough to reliably produce robust, scalable applications without expert supervision.

WHY IT MATTERS

Accessible App Development: Vibe coding democratizes software creation, enabling non-developers to rapidly prototype ideas without deep technical knowledge.

Persistent Challenges: Despite the ease of use, current limitations in AI-driven coding, particularly security, scaling, and efficiency, mean expert oversight remains critical.

Changing Developer Roles: Rather than writing code, developers' roles may shift toward overseeing, refining, and troubleshooting AI-generated code, emphasizing strategic thinking and quality assurance.

Rapid Prototyping Advantage: Businesses that embrace vibe coding could quickly iterate on ideas and bring innovations to market faster than traditional coding allows.

Skills Shift: The ability to read and understand code, rather than write it, may become the most valuable skill in the evolving software landscape.

Is Diffusion Dead?

The world of AI-generated images has fundamentally changed with the recent release of OpenAI’s GPT-4o image generation, shaking up the competition using established diffusion models. Traditionally, AI images were generated through a diffusion process, starting with noise, slowly refined into the desired image. GPT-4o takes a different approach, directly predicting and generating pixels from its generative ability to infer what you have asked to be depicted, similar to painting with pinpoint brushstrokes across a canvas.

This new method offers impressive benefits, including significantly improved accuracy with intelligent text integration. Unlike traditional diffusion methods that struggle with coherent text generation within images, GPT-4o excels in maintaining textual integrity. Early tests suggest that this direct pixel-generation approach could redefine creative workflows, especially in marketing, branding, and rapid content creation.

However, diffusion models, championed by tools like Midjourney, Gemini, and Grok, aren't going away soon. They still offer distinct strengths, Grok's is creativity and Gemini's is precise editing capabilities; GPT-4o has yet to fully match these. Businesses will need to navigate these nuances carefully, selecting tools based on specific creative requirements and desired outcomes.

WHY IT MATTERS

Better Text Accuracy: GPT-4o dramatically improves text coherence within images, opening up new possibilities in advertising, marketing collateral, and instructional content.

Efficiency and Speed: Although GPT-4o currently runs slower due to demand, its direct pixel-generation method theoretically promises faster and more streamlined production processes once scaled.

Creative Competition: The rise of GPT-4o forces existing platforms like Gemini, Grok, and Midjourney to innovate faster, potentially accelerating the pace of AI image-generation advancements industry-wide.

Impact on Marketing Teams: Businesses can quickly create highly customized, brand-specific visuals without the need for extensive design resources, significantly reducing costs and turnaround times.

AI Literacy Essential: With the rapid evolution of these technologies, professionals must continually update their skills and understanding of which AI tool best fits their creative and operational needs.

Just Jokes

The current scene at most brand ad agencies after 4o’s new image generator

Did you know?

Brisbane-based firm Comtrac has been honored with an AI Excellence Award from the US Business Intelligence Group for its innovative digital evidence platform. Founded by former police officer Doran, Comtrac's technology streamlines evidence collection and case preparation, significantly reducing the administrative burden on law enforcement officers. Currently, over 40 law enforcement and regulatory agencies utilize this platform, enhancing the efficiency and accuracy of their investigative processes.

This Week’s Conundrum
A difficult problem or question that doesn't have a clear or easy solution.

The AI Disease Inequity Conundrum

As AI breakthroughs rapidly transform medicine, cures for previously incurable diseases are becoming inevitable. Advanced algorithms are discovering personalized treatments for cancer, genetic disorders, and chronic illnesses, promising a healthier future. But this certainty of progress raises uncomfortable, deeper questions beyond simply having or not having cures.

If AI-generated medical breakthroughs initially favor wealthier nations or individuals due to costs or access, healthcare inequity could sharply increase, not simply between rich and poor, but between entire populations. Over time, the healthiest segments of humanity might gain genetic, biological, or cognitive advantages, effectively creating two distinct classes: those whose health and lifespan are AI-enhanced, and those left behind in a biological status quo.

This isn't a debate about whether we will use AI to cure disease as we surely will. Instead, it’s a complex ethical question of what happens after: Who gets prioritized? Who decides? And how will society manage a potentially permanent divide?

The conundrum: As AI inevitably leads to disease cures, should society actively intervene to ensure these breakthroughs are evenly and immediately accessible, even if it slows innovation or limits investment? Or should we prioritize speed and progress first, accepting initial inequality in the hope it eventually balances out, at the risk of permanently dividing humanity into biological “haves” and “have-nots”?

Ready To Dive Deeper?
Listen to this week’s episode where AI debates both sides using the latest research and perspectives.

Google Launches Gemini 2.5 Pro, Shaking Up AI Rankings
Google surprised everyone by releasing Gemini 2.5 Pro, an advanced multimodal AI model designed to excel at visual web apps and agentic coding. Gemini 2.5 quickly topped LLM rankings, surpassing OpenAI’s GPT-4.5 and previous Gemini model 2.0 Flash Thinking. On benchmarks, Gemini 2.5 leads significantly in coding tasks, achieving top scores in major coding evaluations.

Deeper Insight:
Google’s Gemini 2.5 positions itself directly against coding-focused competitors like Anthropic's Claude 3.7 Sonnet. By excelling in complex agentic tasks and coding, Gemini 2.5 could shift the balance toward Google as developers increasingly rely on AI for coding assistance, and can easily shift to the best-performing model. The major no-code “vibe-coding” platforms like Replit, Loveable, and Bolt have access to multiple models in their platforms. Outside the software development market, Google's challenge remains in breaking ChatGPT’s hold on broader consumer recognition and subscription.

OpenAI Unveils Native Image Generation in an LLM
OpenAI launched its latest update of the GPT-4o model, now featuring native image generation capabilities. This unexpected update was live-streamed as an announcement-with-demo by Sam Altman and principal engineers. This essentially marks the end of DALL-E as a standalone service. By integrating autoregressive visual generation directly within ChatGPT, users are given a new range of text-to-image manipulations, GPT-4o now excels at text rendering, creating more realistic images, and following complex instructions.

Deeper Insight:
By merging image generation into its core ChatGPT product, OpenAI simplifies the user experience and bolsters ChatGPT’s position as the default AI platform. This integration hints at an upcoming UX battle among AI companies aiming to offer seamless multimodal interactions, changing the competitive landscape significantly.

Apple Falls Further Behind in AI Development, Siri Delayed Until 2026 or 2027
Apple reportedly faces significant delays in advancing Siri and its broader AI initiatives, now possibly postponed until 2026 or even 2027. Internal frustration has led to executive shake-ups, sparking speculation that Apple might consider a major AI acquisition to catch up.

Deeper Insight:
Apple’s struggles illustrate how quickly the AI market is evolving. Without significant action, such as acquiring an advanced AI provider like Anthropic, Apple risks losing strategic control of AI integration in its ecosystem. The delay also underscores Apple's dilemma between perfecting technology and keeping pace in a rapidly advancing AI landscape. Or they may be playing a shrewd wait-and-see before staking their claim through acquisition or “distillation” development on top of SOTA models.

Perplexity Makes a Play for TikTok's U.S. Operations
Amid uncertainty surrounding TikTok’s future in the U.S., Perplexity AI made a serious bid to acquire TikTok’s American operations. This follows Perplexity’s earlier insinuations about merging with TikTok, a combination would be a strategic move to integrate with the source of vast social media user data and informational video content.

Deeper Insight:
Acquiring TikTok could transform Perplexity into a powerful AI competitor by providing extensive real-time social media data, similar to XAI’s advantage through Twitter data. This acquisition could significantly enhance Perplexity’s real-time analytics and consumer insights, possibly reshaping competitive dynamics in AI-driven media analysis.

Rev Image 1.0 Offers New AI Image Generation Capabilities
A new entrant, Rev Image 1.0, has emerged in the AI image-generation space, boasting competitive pricing, excellent prompt adherence, and versatility. It positions itself directly against popular rivals such as Midjourney and Ideogram.

Deeper Insight:
Rev’s arrival indicates increased competition in AI-driven creative tools, pushing image generation toward greater affordability and accessibility. If Rev proves capable of consistently outperforming or matching established players, it could shift market dynamics, influencing how creatives choose AI tools.

DeepSeek V3.1 Can Run Locally on High-End Mac Computers
DeepSeek released version 3.1, notable for running efficiently on powerful local machines, like a Mac Studio priced around $10,000. This version enables high-level AI processing without cloud reliance, addressing data privacy concerns tied to Chinese-developed cloud-hosted models.

Deeper Insight:
Local AI deployments present significant privacy advantages, making DeepSeek particularly appealing to businesses and researchers cautious about data sovereignty. This model could set a new trend for local, enterprise-grade AI solutions, reducing reliance on cloud-based services.

Anthropic Gains Further Enterprise Traction Amid Manus AI Success
Following the successful integration of Anthropic’s Claude models into Manus AI’s multi-agent system, Anthropic continues to attract enterprise interest. This success highlights Anthropic's strength in providing powerful, reliable, and ethically-aligned AI solutions for business.

Deeper Insight:
Anthropic’s continued rise in enterprise applications emphasizes the market’s preference for responsible AI. As companies prioritize ethical deployment and transparency, Anthropic’s constitutional AI approach positions it well against rivals who face ongoing scrutiny about safety and alignment.

Coffee-Making Robot from University of Edinburgh Masters Adaptability
Researchers at the University of Edinburgh have developed a robotic arm capable of making coffee even in changing, unpredictable environments. Unlike earlier robots that require precise conditions, this AI system dynamically adapts to misplaced items or shifting arrangements.

Deeper Insight:
This advancement signals a leap forward in robotic adaptability, crucial for integrating AI into everyday human environments. By enabling robots to handle variability, the research moves us closer to truly collaborative robotic assistants in both domestic and professional settings.

AI Identifies Chemical Compositions Using Simple Images
Florida State University has developed AI that accurately identifies chemical compositions from images alone, achieving 99% accuracy with dried salt solutions. Initially created for NASA, this tool significantly simplifies chemical analysis.

Deeper Insight:
This advancement democratizes access to chemical analysis, potentially lowering barriers for smaller labs and field researchers. The technology’s portability could revolutionize in-field scientific analysis, making precise chemical insights affordable and accessible for diverse scientific applications.