Welcome to #45 (like #23, but aged a bit)

In this issue:

Llama, Gemini, GPT: Who’s Winning the 2025 AI Race?

Peeking into AI’s Black Box: What Claude’s Brain Tells Us

Keeping Up with AI: Why Your North Star Matters Most

Plus, we discuss our new Slack community launch (join here for free), ChatGPT’s getting in your personal business, the Irish are not loving X, building websites that are ready for AI tokenization, building mobile apps, if AI can predict your true soulmate, and all the news we found interesting this week.

It’s Sunday morning!

AI memory is leveling up this week.

Somewhere, a goldfish feels personally attacked.

Let’s get into it,

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

Llama, Gemini, GPT: Who’s Winning the 2025 AI Race?

Meta surprised everyone with the sudden release of Llama 4, their latest multimodal AI model, joining a series of major launches in early 2025. With versions named Scout, Maverick, and Behemoth, the Llama 4 assortment aims to be versatile and powerful. Behemoth promises 2 trillion parameters, the largest announced so far by Meta, and is a mixture-of-experts (MoE) architecture with 16 experts governed by an always-on master expert.

This year has already seen major announcements like Google's Gemini 2.5, which quickly superseded the already-competitive multimodal Gemini 2.0. Then OpenAI’s highly successful GPT-4o image generation took the stage, with Anthropic hinting at Claude 4 later this year rounding out the shifting field. This rapid succession of releases feels more significant than previous cycles, raising the question: are these improvements genuinely larger than before, or is the AI industry simply becoming better at capturing our attention?

Meanwhile, upcoming releases like OpenAI’s GPT-5 promise a groundbreaking leap, integrating capabilities across multiple specialized models into one seamless experience. This raises the stakes for competitors to deliver equally transformative products or risk falling behind. While rapid model releases keep companies competitive and constantly in the news, the real test will be their practical impact, how these models actually improve daily tasks for professionals and businesses, and ultimately which models are chosen over the others for application to our work and play.

WHY IT MATTERS

AI Innovation Accelerates: The speed and magnitude of AI model releases seem to be intensifying, suggesting businesses must prepare for continuous adaptation to keep up.

New Industry Benchmarks: Features like multimodal capabilities, MoE architectures, and enormous parameter sizes are resetting expectations for what's standard in AI models.

Risk of Model Fatigue: Companies and users face the risk of being overwhelmed by frequent updates, possibly leading to confusion about which models truly matter for practical applications.

Critical Infrastructure: The rapid pace emphasizes the importance of infrastructure development, like massive data centers, which underpin the availability and capabilities of advanced AI models. Without substantial investment, even the best AI advancements could stumble due to resource limitations which crimp their delivery.

Competition Intensifies: As AI giants like Meta, OpenAI, Google, Xai, and Anthropic race to launch newer, better models, smaller players must carve out niches or risk becoming obsolete in a rapidly advancing market.

Peeking into AI’s Black Box: What Claude’s Brain Tells Us

Anthropic recently published groundbreaking research aimed at "tracing the thoughts" of AI language models, offers a rare glimpse into how AI actually processes information. Using advanced techniques called "attribution graphs," Anthropic researchers examined how their model, Claude, reaches conclusions and assembles relevant responses, from answering straightforward questions like "What is the capital of Texas?" to creating rhyming poetry.

One surprising discovery showed that Claude doesn't just pick words in isolation. When composing poetry, Claude actually plans ahead, mentally organizing rhyming words before it writes the final lines. Another fascinating finding is that Claude effortlessly uses multiple languages internally, sometimes switching to Japanese or Chinese to process ideas quicker or more accurately. This suggests AI models may think differently, and perhaps even more efficiently, than humans, leveraging their multilingual capabilities seamlessly.

These insights fundamentally challenge existing assumptions about AI's internal processes. The discoveries also highlight critical areas for future research, like whether similar "thinking" patterns appear across different AI models from companies like OpenAI or Google, how understanding these patterns can help build smarter and safer AI systems, and possibly reveal innate structures and methods in the ‘black boxes’ of both human and artificial intelligence in neural networks.

WHY IT MATTERS

Understanding AI's Inner Workings: If we can see how AI "thinks," it becomes easier to improve its decision-making, reduce hallucinations, and ensure it aligns with human values.

Multilingual Superpowers: AI’s ability to use multiple languages interchangeably might lead to better, faster insights, and challenges human assumptions about language processing and communication barriers.

Building Trust and Safety: Tracing AI’s decision pathways can enhance transparency, making it clearer why models give specific answers or fail in certain contexts, and potentially prevent dangerous mistakes.

AI’s Unexpected Intelligence: The findings suggest intelligence, even artificial intelligence, may have native organization and emergent patterns of ‘thought’ we don’t yet fully understand.

Opening the Door to More Collaboration: Sharing insights across AI developers could lead to better overall models, benefiting everyone from consumers to businesses, though proprietary concerns may slow this down.

Keeping Up with AI: Why Your North Star Matters Most

Professionals face a critical challenge: staying current with AI without becoming overwhelmed. New developments like Model Context Protocol (MCP), Google's Agent-to-Agent (A2A) communication, and rapid model iterations (such as Claude 3.5 to 3.7) are undeniably exciting, but also potentially exhausting if not understood, absorbed and incorporated effectively.

Experts emphasize the importance of aligning your AI exploration with clear, long-term priorities. AI strategist Allie K. Miller recently highlighted the danger of "missing the forest for the trees," warning against becoming overly fixated on minor model updates or the latest "shiny object." Instead, she suggests stepping back regularly to re-evaluate how these AI advancements genuinely align with personal or organizational goals.

One effective strategy is adopting an "AI-first" mindset in day-to-day tasks, for example, leveraging AI immediately after meetings to pinpoint overlooked opportunities and to clarify and elaborate on insights from the discussions. Also, analyzing and revising workflows holistically, rather than merely adding AI automations into existing processes, can prevent burnout on AI by ensuring each integration of AI is purposeful and productive while maintaining continuity and alignment with the human-managed processes they will assist.

Ultimately, staying abreast of AI doesn't mean chasing every new tool or feature. It means understanding which innovations genuinely align with your goals, then strategically deciding when to engage deeply and when it's better to wait for solutions to mature.

WHY IT MATTERS

Align AI with Your North Star: Clearly define your personal or business priorities. Ensuring AI activities directly support these priorities helps avoid burnout from chasing irrelevant trends.

Be Selective, Not Exhaustive: Trying every new AI development is impossible. Carefully choosing tools that advance your goals saves time and reduces stress.

Adopt Purposeful Workflows: Periodically revamping your workflows around new AI capabilities, rather than shoehorning AI into existing ones, can dramatically enhance productivity.

Security Remains Critical: Despite excitement around rapid innovations, don’t neglect fundamental security concerns, which are easily overlooked in the rush to adopt new technologies.

Learn by Doing, but Pace Yourself: Hands-on experimentation is essential for understanding AI, but pacing yourself strategically, knowing when to actively engage versus when to wait, is critical for long-term effectiveness.

Just Jokes

ChatGPT isn’t messing around with it’s improved memory update 😱

Did you know?

The Irish Data Protection Commission has launched an investigation into Elon Musk's social media platform X regarding its use of personal data to train the Grok AI chatbot. The inquiry focuses on whether publicly accessible posts from European users on X were lawfully processed in accordance with the European Union’s stringent General Data Protection Regulation (GDPR). The Grok chatbot is trained using large language models (LLMs), which rely on vast collections of online text such as articles, blogs, and social media posts. Since X's European headquarters is located in Dublin, the Irish watchdog serves as the lead EU regulator for the company.

Under GDPR, the commission has the authority to impose significant penalties, including fines of up to 20 million euros or 4% of a company's global revenue for major breaches. X has not yet provided any comment regarding the investigation.

Source

Gareth Hood shared his “Blue Steel”

Looking good Gareth!

Dustin asked for thoughts on how to optimize websites for AI tokenization

He said “the theory is to make website data more digestible by AI scrapers to give priority to AI optimized web content.”

Here were a few of the responses:

Justine Reilly has some questions about getting started building a mobile app

“Any ideas on how to get started building a mobile app?  I've done some prototyping myself (phase zero) but I'm looking for a partner who is well versed in a modern approach to developing an app centered around gen AI integration. “

Andy (our definite in-house expert here), replied:
I have been building in lovable, and if you request that it be a responsive app that can work on a mobile browser, the UI will build in a way that can be featured on the web or on the mobile device.

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

The AI Soulmate Conundrum

In a plausible future, not far off, AI has quietly collected the most intimate data from billions of people. It has observed how your body responds to conflict, how your voice changes when you're hurt, which words you return to when you're hopeful or afraid. It has done the same for everyone else. With enough data, it claims, love is no longer a mystery. It is a pattern, waiting to be matched.

One day, the AI offers you a name. A face. A person. The system predicts that this match is your highest probability for a long, fulfilling relationship. Couples who accept these matches experience fewer divorces, less conflict, and greater overall well-being. The AI is not always right, but it is more right than any other method humans have ever used to find love.

But here is the twist. Your match may come from a different country, speak a language you don’t know, or hold beliefs that conflict with your own. They might not match the gender or personality type you thought you were drawn to. Your friends may not understand. Your family may not approve. You might not either, at first. And yet, the data says this is the person who will love you best, and whom you will most likely grow to love in return.

If you accept the match, you are trusting that the deepest truth about who you are can be known by a system that sees what you cannot. But if you reject it, you do so knowing you may never experience love that comes this close to certainty.

The conundrum:
If AI offers you the person most likely to love and understand you for the rest of your life, but that match challenges your sense of identity, your beliefs, or your community, do you follow it anyway and risk everything familiar in exchange for deep connection? Or do you walk away, holding on to the version of love you always believed in, even if it means never finding it?

Want to go deeper on this conundrum?
Listen/watch our AI hosted episode

OpenAI Eyes Hardware Partnership with Sam Altman and Jony Ive Startup
OpenAI is reportedly considering a partnership valued at approximately $500 million with the hardware-device startup co-founded by Sam Altman and Jony Ive. Initially, OpenAI seemed unaware of Altman's hardware venture but is now actively pursuing a strategic collaboration. This move underscores OpenAI's ambition to compete with Google's Android ecosystem, Apple's iOS devices, and Amazon's Alexa-enabled products​.

Deeper Insight:
This partnership highlights OpenAI’s strategy to move beyond purely software-based services by integrating their AI into dedicated hardware. With Perplexity's recent announcement of an AI-powered smartphone collaboration with T-Mobile, the market for AI-integrated consumer devices is heating up. OpenAI’s potential entry into hardware could reshape consumer expectations around integrated AI experiences, significantly affecting the competitive landscape.

Shopify CEO Demands AI Efficiency Before Approving New Hires
Shopify's CEO issued a memo requiring employees to demonstrate that new tasks cannot be performed efficiently by AI before requesting additional headcount. This stance signals Shopify's aggressive commitment to AI-driven operational efficiencies, following similar approaches by companies like Salesforce and Klarna​.

Deeper Insight:
The memo’s viral reception underscores broader industry anxieties and excitement around AI's potential to reshape employment dynamics. While Shopify’s software-centric business aligns well with automation, this approach raises critical questions about AI's broader workforce implications, particularly in industries less adaptable to rapid AI integration.

NVIDIA Launches Vera Rubin AI Chips and Partners with GM
NVIDIA unveiled its new Vera Rubin chip series, promising a 15-fold increase in computational capacity over existing models, due primarily to significant improvements in memory bandwidth. Additionally, NVIDIA has partnered with General Motors to accelerate development in autonomous vehicle technology, challenging Tesla’s dominance in the sector​.

Deeper Insight:
These advancements solidify NVIDIA’s leading role in AI hardware and underscore the company’s strategic ambition in automotive AI solutions. The GM partnership not only enhances NVIDIA’s industry position but also potentially accelerates mainstream adoption of autonomous vehicles.

Google Announces Ironwood TPU and Gemini 2.5 AI Model
Google introduced its seventh-generation AI chip, Ironwood, optimized specifically for AI inference, featuring clusters as large as 9,216 chips. Alongside hardware advancements, Google launched the Gemini 2.5 AI model, emphasizing dynamic responsiveness and improved computational efficiency​.

Deeper Insight:
Google’s continuous hardware and software evolution demonstrates a commitment to regaining competitive ground against OpenAI and NVIDIA. Ironwood’s massive scale may attract enterprise clients needing intense AI workloads, while Gemini 2.5’s efficiency might significantly lower costs for broader AI applications.

Reddit Adopts Google's Gemini for AI Features
Reddit has started integrating Google’s Gemini AI into its platform, aiming to leverage Gemini’s advanced capabilities to enhance user experiences and data-driven decision-making on the site​.

Deeper Insight:
This collaboration could considerably strengthen Reddit’s analytical capabilities and user engagement. With Google’s AI powering data insights, Reddit may significantly enhance its platform relevance and commercial attractiveness, altering competitive dynamics with social platforms like X.com and Meta.

Intel and TSMC Form Strategic Chipmaking Joint Venture
Intel and Taiwan Semiconductor Manufacturing Company (TSMC) announced a new joint venture, with TSMC acquiring a 20% stake in Intel’s manufacturing facilities. This collaboration addresses the strategic need to diversify global semiconductor manufacturing amid geopolitical tensions and supply chain vulnerabilities.

Deeper Insight:
This alliance could dramatically alter the semiconductor landscape by stabilizing chip supplies and bolstering production resilience. For Intel, the partnership provides an opportunity to reclaim lost manufacturing ground, potentially revitalizing its strategic importance in global chip production.

Higgsfield Launches Advanced AI Video Generation
Higgsfield, a relatively unknown AI startup, has unveiled advanced video generation capabilities featuring impressively realistic human interactions, such as natural sitting movements and environmental interactions that rival established platforms like Runway and OpenAI’s Sora.

Deeper Insight:
The emergence of Higgsfield illustrates growing competition in AI video generation, potentially democratizing high-quality animation and video production. Such advanced realism could significantly impact media creation, lowering production barriers for filmmakers, advertisers, and digital creators.

Penn State Develops Soft, Flexible Robots for Life-Saving Applications
Penn State researchers introduced magnetically controlled, flexible robots capable of navigating complex environments like earthquake rubble or delivering medication within the human body. These robots aim to operate effectively across a wide range of sizes, from macro to micro scales​.

Deeper Insight:
This breakthrough in adaptable robotics could revolutionize disaster response and medical interventions, highlighting AI and robotics’ growing potential in critical human-centric applications. As robotic technology becomes smaller and more versatile, it could significantly expand practical robotics into everyday life and complex emergency scenarios.

ElevenLabs and Supabase Join Anthropic's MCP Initiative
ElevenLabs and Supabase announced support for Anthropic’s Model Context Protocol (MCP), enabling easy integration of AI services like voice generation and database queries through simple prompts. This development expands the ecosystem for streamlined AI integration in business applications​.

Deeper Insight:
MCP's growing adoption suggests an industry-wide shift toward standardizing AI integrations, potentially making advanced AI features accessible even for small developers and businesses. This unified approach could rapidly accelerate innovation and reduce complexity in enterprise AI deployments.

OpenAI Launches Educational Platform 'OpenAI Academy'
OpenAI has introduced "OpenAI Academy," a comprehensive educational portal offering live training events and workshops covering AI for diverse sectors, including nonprofits, K-12 education, and older adults. This initiative aims to foster widespread AI literacy and practical application across various demographics.

Deeper Insight:
The Academy represents OpenAI's strategic push into AI education, significantly enhancing public understanding and adoption of AI technologies. By democratizing AI education, OpenAI positions itself as an essential resource for widespread AI competence and ethical deployment, potentially shaping broader industry standards.