Welcome to The Daily AI Show Newsletter, your deeper dive into AI that goes beyond the latest news. In this issue:

• Is AI Slowing Down? Why the Question Misses the Point

• The Secret to Going Viral: What AI Success Stories Teach Us

• Elevating AI Prompting: Making the Most of Tools Like Anthropic Console

Plus, we discuss Runway’s Act-One and show you a workflow for creating your own, ChatGPT’s absolute dominance in September, Siri’s connection to DARPA, the tipping point for trusting AI over humans, hacking Anthropic for good, and all the new stories that caught our eyes and ears this past week.

It’s Sunday morning and it is getting cooler here in the States.

But AI doesn’t care if it’s winter—it’s still suggesting sandals based on that one vacation search you did in July.

The DAS Crew

Is AI Slowing Down? Why the Question Misses the Point

Recent debates around AI’s pace of innovation—spurred by questions like “Is AI slowing down?”—may be missing a critical point. While discussions about scaling compute, data availability, or the release timelines of new models like GPT-5 grab headlines, they overlook a fundamental reality: we’ve only just begun to scratch the surface of how current AI tools can be applied in business, life, and society. Even if AI development temporarily "slowed," the untapped potential of what already exists could keep businesses innovating for years.

What’s more, innovation isn’t just about bigger or faster models. Advances in areas like inference efficiency, domain-specific applications, and cost reductions are creating new opportunities for small and medium-sized enterprises to adopt AI at scale. As AI gets cheaper and more accessible, innovation will shift from big tech labs to the creativity of everyday users finding novel applications for existing tools.

WHY IT MATTERS

Room for Business Growth: Many organizations are still in the early stages of adopting AI. Current tools offer years of potential for streamlining operations, improving customer experiences, and driving revenue growth.

Cost Drives Accessibility: As the cost of AI tools continues to drop, smaller businesses can afford to implement AI solutions, driving innovation across more industries and use cases.

Creativity Over Computation: Innovation is shifting toward how users creatively apply existing AI models, rather than relying on massive leaps in model performance.

Evolving Use Cases: AI is enabling entirely new workflows and efficiencies, from interactive customer support to more dynamic internal collaboration tools.

A Long-Term Revolution: The full impact of AI isn’t defined by short-term advances; it will depend on how well businesses and individuals integrate these tools into daily operations over the next decade.

The Secret to Going Viral: What AI Success Stories Teach Us

What makes an AI product go viral? It’s not always the technology itself. Some tools with advanced capabilities fizzle out, while others, like ChatGPT, achieve explosive adoption despite offering features similar to existing tools. The secret lies in understanding the factors that drive widespread acceptance and use of new technology.

The "Unified Theory of Acceptance and Use of Technology" provides four key elements that explain why some tools succeed:

• Performance Expectancy: People must see the tool as genuinely useful. ChatGPT's versatility, from writing emails to brainstorming ideas, instantly resonated with users.

• Effort Expectancy: Simplicity drives adoption. ChatGPT feels intuitive, like texting or searching, making it accessible to users without technical expertise.

• Social Influence: People adopt tools their peers or influencers endorse. ChatGPT’s virality was boosted by media coverage and tech leaders promoting its capabilities.

• Facilitating Conditions: Users need accessible support and resources to solve problems. Companies like Perplexity and Replit excel in creating responsive communities and offering real-time help.

WHY IT MATTERS

Adoption Over Innovation: Businesses can focus on improving user experience and support to drive adoption, even if their technology isn’t groundbreaking.

User-Centered Design: Tools that feel intuitive and align with user expectations gain traction faster, helping businesses scale.

Community Drives Success: Creating engaged, supportive user communities can foster loyalty and attract new users through word-of-mouth.

Sustainable Growth: Products that balance rapid adoption with user retention build long-term value, rather than being flashes in the pan.

Framework for Success: Leveraging these four principles can help businesses design tools that achieve both virality and lasting impact.

Elevating AI Prompting: Making the Most of Tools Like Anthropic Console

The rise of tools like Anthropic Console is revolutionizing how we create and refine prompts for AI, making the process accessible to more people than ever before. Whether you’re an AI novice or a seasoned expert, these tools offer structured, iterative approaches to help craft better instructions for large language models (LLMs) like Claude, ChatGPT, and Gemini. At the heart of these advancements in building quality prompts is translating your simple statements into detailed prompts that focus on clarity, specificity, efficiency, and adaptability.

Prompt improvement isn’t just about cleaning up wording or making tasks easier to follow—it’s about unlocking the full potential of AI models. By using techniques like example enrichment, chain-of-thought reasoning, and variable structuring, these tools ensure your prompts are optimized for precision and reliability of LLM responses. For businesses and individuals alike, this means better results with less effort, whether you’re generating content, analyzing data, or building workflows.

WHY IT MATTERS

Democratizing Prompt Engineering: Tools like Anthropic Console simplify complex prompt design, allowing anyone to craft effective instructions without a steep learning curve.

Efficiency and Scalability: By improving prompts iteratively, these systems enable users to get accurate, repeatable results, saving time and resources in both personal and professional applications.

Cross-Platform Insights: While not all prompts transfer seamlessly across LLMs, tools like Anthropic’s prompt improver help bridge differences, making it easier to adapt prompts for various models.

Tailored Outputs: Features like XML tagging and structured variable inputs allow users to design highly specific, reusable prompts tailored to their requirements.

Enhancing AI Collaboration: The iterative process of refining prompts fosters better collaboration between humans and AI, improving trust and understanding in the outcomes delivered by these systems.

Just Jokes

What if Abbott and Costello did a routine about how confusing AI names and terms can be?

That was Brian’s inspiration for creating his recent short animation made with Runway’s Act-One.

“If I Can Do This, Anyone Can” - Brian

Brian shared his workflow for creating the short animation above on the show, but here it is if you want to follow along. Brian mentioned he would do it differently the next time after learning what worked and what didn’t. So while this first try took him around 3 hours total, he thinks he could recreate something similar in about half that time. Stay tuned!

Here is how Brian explained the process:

1. I start with an idea for a funny skit about the complexities of understanding AI by doing a parody of Who's On First by Abbott and Costello.

2. Using Fal.com and Flux 1.1 pro (ultra), I worked to get my cartoon versions of Abbott and Costello. I ran into several issues here because many of the much better cartoon/Pixar versions had round noses more like clown's nose than a real one. Act-One couldn't identify the face though in those images, so I had to ultimately go with characters that had more realistic facial features. Once I did that, Act-One had no issue identifying and using the reference characters.

3. I used ChatGPT to help write the 2 person script. It started with my idea, ChatGPT wrote about 80% of it out of the gate, I made some suggestions and had some new ideas based on what I got, and ChatGPT wrote the final script. Total time was probably 10 min.

4. I split the script up by speaker and then recorded myself saying the lines with a slight pause between each one.

5. To get different voices, I used ElevenLabs Voice Changer and created a new audio track.

6. Using Canva, I synced the new audio track with my original video and muted the original audio.

7. The overall length was over 30 seconds (the limit for a single video in Act-One), so I chopped up each video into segments in Canva

8. Using Act-One, I fed the 8 smaller clips in and got the rendered characters now saying my lines, mimicking my facial expressions and using the ElevenLabs new voices.

9. Then I put everything into DaVinci Resolve 19 and chopped up the individual lines. Since I didn't mark the times when one character wasn't speaking, I decided to do zoom-in shots on one character after the original opening where you see both characters saying thank you to the crowd.

10. Finally, I added a few laugh tracks just for fun and then I rendered it.

Anthropic Gives Love to Hackers

Karl shared this story about Anthropic working with hackers to see what they can do with Computer Use Agents.

Here is some of what they saw according to Alex Reibman on X:

LLM (Love Language Model)

Agent that automatically finds matches on Hinge with an iPhone simulator controller

Love is blind but smart

Task Bridge

Computer Use agent that reads through your slack messages, identifies tasks, and submits them to Asana

Zero effort task management with computer use

Tool use team

Multi-agent “bureaucracy” for instructing Computer Use agents to perform tasks better than a single standalone agent

Multi-agent computer use

Open Source Computer Use 🥉3rd Place

Computer Use system built from the ground up with Open Source and tools.

Converts screenshots to grids and uses pyautogui, element localization, image-to-text, and an agent loop to perform tasks onscreen

Phaedrus - 🥈2nd Place

Quality assurance agent that goes through UX-flow task lists, attempts to test they work, and automatically creates Jira tickets for broken features

Automates QA testing of user experiences

Tests AI - 🥇1st place

Generates UI tests described in natural language that execute computer use agents to test frontend interfaces and user-actions.

Simplifies creating QA test scripts from text descriptions

Takeoff!

Andy shared this graphic showing just how far ChatGPT has pulled away from the competition in terms of desktop and mobile visits in September.

ChatGPT had more traffic than Amazon.com! 🤯

Did you know?

Apple’s Siri wasn’t originally an Apple creation—it started as a project funded by the U.S. Department of Defense’s DARPA program. Siri was developed by SRI International, a nonprofit research institute, and was part of the CALO project (Cognitive Assistant that Learns and Organizes), aimed at creating AI that could adapt and learn from users over time.

When Siri launched as an iPhone app in 2010, it caught Apple’s attention for its innovative voice recognition and AI capabilities. Apple acquired the company behind Siri just two months later, integrating it into the iPhone 4S in 2011 as the first major voice assistant in a smartphone.

Here’s something even less known:

Siri’s creators originally envisioned it as a multi-platform assistant that could integrate with Android and BlackBerry. Apple’s acquisition, however, made it exclusive to iOS, cementing Siri’s role as a key part of Apple’s ecosystem.

Today, Siri uses advanced machine learning and natural language processing to handle billions of requests per month, evolving from simple commands to a sophisticated AI assistant.

Its origins as a government-funded project highlight how technologies designed for research often find unexpected applications in consumer products.

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

The AI Agent Error Trade-Off Conundrum:

AI agents are becoming capable of handling complex tasks independently, from managing finances to diagnosing medical issues. Like self-driving cars, these agents could dramatically reduce errors compared to human performance, improving efficiency and accuracy on a large scale.

However, no AI system is perfect, and some errors—whether due to misinterpretation or unforeseen circumstances—will inevitably occur.

While these mistakes may be rarer than human errors, their consequences could be more far-reaching, raising difficult questions about accountability and societal tolerance for AI-driven failures.

The conundrum: Should society accept that some mistakes by AI agents are inevitable, tolerating these errors as part of a broader benefit to humanity? Or should the expectation of perfection for AI systems remain higher than for humans, even if it means delaying their deployment or limiting their autonomy?

Leibniz Institute's AI Models for Ecosystem Monitoring
The Leibniz Institute for Zoo and Wildlife Research has developed AI models that track vultures in the wild, enabling them to act as "death detectors." This innovative use of AI aids in monitoring ecological cycles, disease outbreaks, and illegal hunting activities, offering a unique blend of technology and environmental conservation.

Factiverse Tackles Disinformation with AI
Norwegian company Factiverse has launched an advanced AI system for real-time fact-checking across 114 languages. The platform evaluates the credibility of sources and claims in text, video, and audio formats, outperforming existing tools like GPT in accuracy.

Ilya Sutskever: Scaling Laws and the Future of AI Models
Ilya Sutskever from OpenAI has highlighted the limitations of scaling laws in large language models, emphasizing the need for alternative methods to achieve efficiency and progress in AI development. This revelation has prompted companies to explore reasoning and other innovative approaches to improve AI models.

Meta Expands into Business-Focused AI Tools
Meta has formed a new business unit to create AI tools beyond ad optimization. Led by Clara Shih, formerly of Salesforce AI, this initiative aims to provide businesses with resources for customer relationship management and operational improvements using AI.

Google Gemini Introduces Memory Feature
Google's Gemini AI has launched a memory feature, available for Google One AI Premium users. This enhancement allows the model to retain conversational context, making interactions more seamless and informed.

OpenAI Adds Advanced Voice Mode to Web Version
OpenAI has announced the integration of advanced voice mode into its web interface for ChatGPT, unifying its features across platforms and improving accessibility for users.

Mistral's Free ChatGPT Rival, Le Chat, Updated
Mistral has updated its free ChatGPT rival, Le Chat, with new features like collaborative image generation and editing. The move underscores the growing competition in the AI chatbot space, with Le Chat offering significant capabilities at no cost.

Perplexity Adds E-Commerce Functionality
Perplexity AI has introduced a shopping capability, allowing Pro users to make purchases directly through its platform. This feature highlights Perplexity’s pivot toward monetizing its search and recommendation services.

ElevenLabs Launches Conversational Agents
ElevenLabs has expanded its voice cloning technology to include conversational agents, enabling businesses to build interactive AI-driven customer service tools directly on its platform.

Ben Affleck and Deloitte Discuss AI in Hollywood
Ben Affleck remarked on generative AI's role in Hollywood, emphasizing its utility for repetitive tasks rather than artistic creation. Deloitte's analysis aligns, suggesting AI will mostly support operational processes like contracts and distribution rather than creative roles.

Moon Valley Secures $70 Million Seed Round for Generative Video
Moon Valley, a Hollywood-based company, raised $70 million to develop ethical generative video models for advertising and film. Their focus is on partnering with creatives for opt-in, indemnified content development.

Did You Miss A Show Last Week?

Enjoy the replays here on YouTube or take us with you in podcast form on Apple Podcasts or Spotify.