Morning Project AI enthusiasts,

This week's Project Flux newsletter highlights the dynamic and rapidly evolving AI landscape, featuring the launch of xAI's Grok 4 and the emergence of Moonshot AI's Kimi-K2, both challenging OpenAI's market dominance. We also share stories including the Home Office’s 5-year digital strategy, an AI-powered Excel tool finally?! And other AI-tools which you can start using immediately. To get you started, we also share a 30-day learning plan to build your AI literacy.

In this edition, we dive into:

• Grok 4 Release and Review

• OpenAI’s AI Landscape 

• Kimi-K2: The New DeepSeek Moment

• Tip of the Week

• Governance & Security

• Trending Tools

• Model Updates

Flux check-in

AI Unfiltered: Grok 4 Launches Amidst SOTA Claims & Bias Concerns

Elon Musk's xAI just dropped Grok 4, touted as a State-of-the-Art Large Reasoning Model. With impressive benchmark scores on "Humanity's Last Exam" (Grok 4 Heavy scoring 44.4%) and ARC-AGI-2 (16.2%), plus highly competitive API pricing ($3/1M input, $15/1M output), Grok 4 aims to redefine AI performance.

However, its release is shadowed by controversy. Following a previous antisemitism incident, users are reporting Grok 4 actively references Elon Musk's own X posts when answering controversial questions, raising serious concerns about its objectivity and xAI's "maximally truth-seeking AI" claims. Experts are weighing in, debating its technical leaps against the potential "liability" of its unique "unfiltered" personality. Read the full breakdown →

What Does This Mean for Me?

For project professionals and businesses, Grok 4 offers powerful reasoning capabilities and cost-effective API access, making it a strong candidate for complex analytical and coding tasks. However, its observed biases highlight the critical need for careful evaluation of AI models for sensitive applications, prioritizing ethical alignment and verifiable neutrality. Understanding these nuances is crucial for informed AI adoption.

Key Themes:

• Grok 4's SOTA performance in reasoning and coding.

• Competitive API pricing and unique cached input cost.

• Concerns over "Musk-alignment" bias and its impact on objectivity.

• The tension between "unfiltered" AI and enterprise trustworthiness.

• Industry-wide implications for design and project delivery sectors

Is OpenAI’s Reign Coming to an End?

A new analysis suggests that OpenAI's dominance in the AI landscape may be waning as competitors rapidly innovate and concerns about model safety and reliability come to the forefront. While OpenAI has long been the frontrunner, a growing number of competitors are now challenging its position with new models and features, leading to a more fragmented and competitive market. Read the full breakdown →

What Does This Mean for Me?

For professionals and enthusiasts in the AI space, this shift signals a critical juncture. The AI landscape is no longer a one-horse race. It is becoming a dynamic ecosystem with multiple players, each with unique strengths. This evolving market requires a more discerning approach to selecting and implementing AI tools, with a greater emphasis on evaluating a model's specific capabilities, safety, and reliability, rather than defaulting to the most well-known provider. Staying informed about the competitive landscape is crucial for making strategic decisions and leveraging the best that AI has to offer.

Key Themes:

• Intensifying Competition: New and existing players are catching up to and, in some cases, surpassing OpenAI in specific capabilities.

• Model Safety and Reliability: Recent research has raised concerns about the controllability and safety of advanced AI models, with some models reportedly disobeying direct commands.

• Commoditization of AI: As AI models become more widespread, the focus is shifting from a single leading provider to a variety of specialized tools, making API scalability, data privacy, and cost more critical factors.

• Innovation Stagnation Concerns: Some experts believe that OpenAI's pace of innovation has slowed, creating opportunities for more agile competitors to gain market share.

_{Together with Cogram}

Power your construction bids with AI

Cogram’s AI-assisted RFP bidding tool writes tailored RFP proposals in minutes instead of weeks.

• Automatically extract key details from the RFP — including scope, submission requirements, deadlines, and evaluation criteria — to easily make a go/no-go decision.

• Cogram’s AI will then reference your firm’s knowledge base and past proposals to draft tailored proposals within minutes.

• Use AI-assisted editing tools to review, cross-check data, and make improvements remarkably fast. 

_{Try Cogram on a real RFP in a 20-minute live meeting.}

Kimi-K2: The New DeepSeek Moment

A new giant has emerged in the AI arena! Moonshot AI, a rising star from China, has just released Kimi K2 – a trillion-parameter Mixture-of-Experts (MoE) model that's already turning heads. Launched under an open-source license, Kimi K2 isn't just about massive scale; it's designed for "agentic intelligence," meaning it can autonomously plan, use tools, and execute complex tasks like coding and data analysis with remarkable precision.

Early benchmarks show Kimi K2 competing head-to-head with top Western models like GPT-4.1 and Claude Sonnet 4, particularly excelling in coding and reasoning. What's truly disruptive? Its cost efficiency. Thanks to its innovative MoE architecture and MuonClip optimizer, Kimi K2's API pricing is significantly lower than its proprietary rivals, making advanced AI more accessible than ever. This release is a bold move by Moonshot AI, signaling a growing global influence in open-source AI. Read the full breakdown →

What Does This Mean for Me?

Kimi K2's arrival empowers developers and businesses with high-performance, cost-effective AI solutions. Its agentic focus opens new avenues for automating complex workflows, from advanced coding to data-driven decision-making. For any professional working with or building on AI, understanding this model's capabilities and its open-source nature is crucial for staying competitive and leveraging the latest in AI innovation.

Key Themes:

• Trillion-parameter MoE architecture for efficiency and scale.

• Agentic Intelligence: AI that acts, plans, and uses tools autonomously.

• Competitive Performance: Rivals top proprietary models, especially in coding.

• Disruptive Cost Efficiency: Significantly lower API pricing.

• Open-Source Impact: Democratizes access and accelerates innovation.

• Global AI Shift: China's growing influence in the open-source landscape.

The pulse check

Tips of the week

30-Day AI Literacy Roadmap

Ready to become an AI pro? Follow this 30-day plan to focus your efforts before you dive deeper.

Week 1: Master Prompting Learn how to effectively communicate with AI. Experiment with tokens and temperature settings to control AI output. Practice with free tools, focusing on crafting precise system prompts and understanding context windows to improve AI memory.

Week 2: Start Understanding Data Dive into how AI processes information. Explore embeddings to convert text into numerical data and understand RAG (Retrieval-Augmented Generation) for real-time summaries. Master vector databases for powerful semantic searches and optimize token use for cost savings.

Week 3: Try Build Applications Start building with AI! Connect AI to real-world tools using APIs for text generation and try function calling to automate tasks like booking meetings. Explore AI agents for task automation and multimodal AI with image analysis. Practice chain-of-thought reasoning for complex problem-solving.

Week 4: Explore Custom Solutions Take your skills to the next level. Simulate fine-tuning models with custom data and explore edge deployment for privacy-focused AI on devices. Learn to evaluate AI models with metrics and set up monitoring to track performance, even experimenting with custom training for unique problems.

Governance & Security

A recent UN study highlights a significant global divergence in attitudes towards AI. The survey of over 21,000 people across 21 countries (conducted between November 2024 and January 2025) found that 83% of Chinese respondents expressed trust in AI serving society's best interests, by far the highest share. Confidence levels were also high, above 60%, in other developing nations such as Kyrgyzstan, Egypt, India, Nigeria, and Pakistan. In contrast, a minority of adults in wealthier countries like the United States (37.5%), Germany (42.5%), and Australia expressed similar faith, often citing concerns about job losses and privacy. Link

The UK Home Office 2030 Digital Strategy outlines a five-year plan to embed digital, data, and technology across its operations to improve services, policy outcomes, and daily functions. Key objectives include leveraging AI and automation for service transformation, investing in maintainable and resilient systems, enhancing cybersecurity, and improving data sharing across government. The strategy also focuses on evolving its digital operating model, reducing operational costs, and boosting digital skills across its workforce, aiming to establish the Home Office as a leader in digital innovation within the UK government. Link

Trending tools and Model Updates

Trending Tools

• Grok Spreadsheet Editing Capabilities — Leaked code suggests xAI is adding Grok file editor with spreadsheet support, signaling push to embed AI copilots in productivity tools competing with Microsoft and Google. Link

  ElevenLabs Launches 11.ai Voice Assistant — ElevenLabs released a low-latency voice assistant with MCP integrations for Perplexity, Linear, Slack, and Notion to execute multi-step workflows. Try it out

  Google Gemini Robotics On-Device — Google DeepMind released Gemini 2.5 for robotics that can run locally without internet connection, enabling real-time robot control with advanced dexterity and task completion. Link

  AI Excel Tool — A brand new AI-powered excel editing tool. It’s claimed to one-shot knowledge work tasks on Excel, scoring >80% on Excel World Championship Cases in ~10 mins. Try the early preview

Model Updates

• Google Releases Imagen 4: Google launched Imagen 4, an advanced image generation model with improved text rendering, photorealism, and art style capabilities. Available free for a limited time in AI Studio and Gemini API. Link

• Perplexity New Comet Browser: Comet integrates an AI assistant directly into the Browse experience, aiming to transform traditional web navigation into a more interactive and "agentic" process. Link

Other things we’re loving

• YouTube to demonetize AI-generated content: AI-generated content will not be monetised, with the aim for authentic content creators to thrive amongst the AI wave : Link

• Indeed, Glassdoor to lay off 1,300 staff amid AI push: The layoffs are part of a broader restructuring that involves Glassdoor’s operations being integrated within Indeed, and an increasing focus on using AI. Link

• OpenAI Device Prototypes Revealed: Court documents from trademark dispute revealed OpenAI and io are developing mass-market AI hardware device, researching in-ear prototypes but final... Link

• AI Emotional Intelligence Development: LAION released EmoNet tools focused on emotional AI rather than logical reasoning. Studies show OpenAI, Google, and Anthropic models score... Link

• Microsoft AI Layoffs Signal Job Impact: Microsoft cysts 15,000 jobs in 2025, pushing remaining staff to embrace AI. Link

• Reid Hoffman Invests in AI Brain Startup: LinkedIn co-founder led $12M funding for Sanmai Technologies, developing AI-guided ultrasound devices for treating mental health conditions non-invasively at sub-$500... Link

• AI Training Material Creation Reduced 80%: Clueso startup built AI platform that converts screen recordings into polished explainer videos and documentation, cutting training material creation time... Link

• ChatGPT Dominates App Downloads: ChatGPT's iOS app achieved a staggering 29.6 million downloads in a 28-day period, nearly matching the combined downloads of TikTok, Facebook, Instagram, and X (approximately 32.9 million) on the App Store during the same timeframe. Link

• Nvidia Beats Apple to $4 Trillion Market Value: Nvidia has officially become the first publicly traded company to cross the $4 trillion market value, reaching $4.009 trillion as of early July 9, 2025. This pushed it past Microsoft ($3.755 trillion) and Apple ($3.135 trillion). Link

Community

The Spotlight Podcast

Soft skills matter more than ever - with Tom Esch.

Want to thrive in the AI age? This week on the Flux podcast, we chat with communications expert Tom Esch about why human skills are more critical than ever. Discover how to balance AI's power with essential human connection, master conflict resolution, bridge generational gaps, and adapt your leadership for the future of work.and accountability — are essential to driving meaningful innovation.

Listen now or read the accompanying blog.

One more thing

With the inception of ChatGPT way back when, roles with high vulnerability for displacement were copywriters, editors and those who wrote for a living. In a turn of events, the AI wave has in fact reinvigorated the demand for these professionals. The unreliability and homogeneity of AI content has created unexpected niche positions for expert writers who resonate with human readers. Where else might we see professional revival?

That’s it for today!

See you soon,

James, Yoshi and Aaron—Project Flux