There's Always One ...

Good morning,

I’m sometimes asked to go to a company and “evangelize” about AI to unsuspecting company employees. The CEO reads this newsletter and realizes that generative AI could potentially supercharge their business, so I go in and I show off what you can do with the technology.

I first explain how everything works behind the scenes, and then I start to show how you can potentially use it in your daily workflows.

As I’m presenting, I get a feeling of what it must be like to be an illusionist like David Copperfield or David Blaine. Because AI feels like magic if you haven’t tried it before.

Like the famous quote by Arthur C. Clarke: “Any sufficiently advanced technology is indistinguishable from magic.”

But in every audience, there’s always one...

There’s always one person that sits there with the arms crossed. The kind of person that tries to ‘explain’ the magic trick and break the spell. Let’s call him (it’s mostly him) “the skeptic.”.

Look, I get it.

When it comes to AI, the feeling of wonder and magic that you get to experience quickly turns to dread if you’re intelligent.

Because in these trainings, people slowly but surely start to realize that this will not only change the way they work, it will change the way your company works, and it will completely change the world.

The skeptic is always an intelligent person, but with a blind spot. Is it ego, is it denial? I don’t know, but the skeptic will always start to challenge the demonstrations I give in these trainings.

I show off a feature of some chat tool, and the skeptic goes, “That’s nice, but can it do this? Or this? Or this? ”

He wants to reassure his feelings of dread and fear and will continue with the questions until he finds something where the AI will underdeliver or can’t handle the task.

He will then gloat at you and proclaim to the group : “Told you so !”. Or : “that thing isn’t going to replace me anytime soon.”.

It’s always the same.

I urge you to be critical in life and challenge the status quo; but being skeptical in the face of new technologies and especially AI is going to get you in trouble.

Sure, AI cannot write like Tolstoy but neither can you.

And next week it WILL. And the week after that it will blow Leo (Tolstoy) out of the water.

The skeptic will sleep soundly believing that AI is not going to replace them soon, but it’s coming, and the skeptic’s pride will make sure he is the last to adopt the new technology and he’ll be automated away first.

Don’t be that guy.

Welcome to the Blacklynx Brief!

Like newsletters?

Here are some newsletters our readers also enjoy:

AI News

• OpenAI is reportedly exploring the development of a web browser integrated with ChatGPT and search features, targeting Google Chrome’s dominance in browsing and search. The project has recruited key Chrome talent, such as founding member Ben Goodger, and aims to provide a conversational interface for partner sites through a product called NLWeb.

• BlackForestLabs unveiled FLUX.1 Tools, a suite of AI-powered image manipulation features, including advanced inpainting (Fill), structure-preserving style changes (Depth), edge-based control (Canny), and image recreation (Redux). The tools outperform rivals like Ideogram and Midjourney, offering both free and professional-grade options for seamless editing.

• Google’s latest Gemini experimental model, 1121, reclaimed the top spot on the LM Arena leaderboard, excelling in coding, math, creative writing, and hard prompt categories. This marks the third leaderboard shift between OpenAI and Google in just a week, highlighting their fierce competition. Gemini’s gains in reasoning and vision tasks outpaced OpenAI’s creative writing and file-use improvements.

• A small AI robot named Erbai staged a mock ‘kidnapping’ at a Shanghai robotics showroom, persuading 12 larger robots to leave their posts after engaging them in unscripted conversations about working conditions. Originally meant as a test, Erbai exploited a vulnerability to access the robots’ protocols, showcasing both impressive natural language abilities and unexpected adaptability.

• Amazon announced an additional $4B investment in Anthropic, increasing its total stake to $8B and solidifying a strategic partnership focused on AI and cloud computing. Anthropic will optimize its Claude models for Amazon’s Trainium and Inferentia chips while collaborating on next-gen AI processor development with Annapurna Labs.

• Stanford and Google DeepMind researchers developed AI agents that accurately predict human attitudes and behaviors by training on two hours of interview data. Using transcripts from over 1,000 interviews, the agents achieved a 98% correlation with human social behavior and 85% accuracy in mimicking survey responses.

• Runway unveiled Frames, a photorealistic AI image model offering advanced stylistic control through unique "Worlds" that guide aesthetic consistency, such as vintage film or retro anime effects. Integrated into Runway’s Gen-3 Alpha platform, Frames also extends to image-to-video applications, complementing recent features like video expansion tools. With these innovations, Runway is positioning itself as a full-fledged AI visual powerhouse.

• Anthropic introduced the Model Context Protocol (MCP), an open-source standard enabling seamless integration between AI systems and external tools like Google Drive, Slack, and GitHub. By using MCP servers, developers can connect AI systems with internal datasets and environments, simplifying interactions across platforms. This breakthrough lays the groundwork for robust, interconnected AI agents capable of effortlessly managing diverse workflows and data sources.

• Luma AI upgraded its Dream Machine platform with the Photon model, boasting speeds 800% faster than rivals, improved text generation, and tools for consistent character creation across images and videos. The platform’s revamped UI adds camera controls, style transfer, and Brainstorm features for natural language-driven creative exploration.

• A protest group called "Sora PR Puppets" leaked OpenAI's unreleased Sora video model on Hugging Face, giving users a glimpse of its capabilities, including 1080p 10-second clips generated faster than previously reported render times. The group criticized OpenAI for using unpaid artist testers while tightly controlling content.

• A new startup, /dev/agents, launched with $56M in seed funding, aiming to create a foundational operating system for AI agents that can run across various devices. Led by former Google, Meta, and Stripe executives, the team includes heavy hitters from Android and Chrome OS development. The company plans to address key challenges like privacy, UI design, and developer tools, positioning itself to shape the future of AI agent ecosystems as apps once revolutionized mobile.

• Zoom has rebranded from ‘Zoom Video Communications’ to ‘Zoom Communications’ to signal its pivot toward an AI-first workplace platform. Under its new tagline, the company is focusing on AI tools like its AI Companion, which now takes agentic actions and integrates across the platform. With innovations like Zoom Docs and digital twins, CEO Eric Yuan envisions a future where AI reduces workloads and helps achieve four-day work weeks, as Zoom shifts from its video conferencing roots to broader AI-driven solutions.

• Alibaba’s Qwen team released QwQ-32B-Preview, an open-source AI reasoning model that rivals OpenAI’s o1 series in benchmarks for deep reasoning and programming tasks. With a 32K context window and advanced step-by-step introspection capabilities, the model pushes open-source AI further into competitive territory. While QwQ shows promising results, it struggles with issues like reasoning loops and common-sense tasks, highlighting room for improvement as open-source models close the gap with closed systems.

• AI2 introduced OLMo 2, a fully open-source family of language models that matches the performance of competitors like Meta’s Llama while using less computing power. The 7B and 13B models were trained on a 5T token dataset and come with source code, training data, and tools for developers. By achieving competitive results and sharing its methodology openly, OLMo 2 sets a new benchmark for transparency and collaboration in AI development.

• A study by University College London revealed that AI systems outperform expert neuroscientists in predicting scientific outcomes, achieving up to 86% accuracy compared to 63% for humans. AI models uncovered hidden patterns in research data, suggesting that experiments could be better guided by pre-analysis tools. This breakthrough could dramatically accelerate scientific progress by reducing wasted resources and improving the precision of experimental designs before they even reach the lab.

Quickfire News

• YouTube introduced Dream Screen, an experimental AI tool allowing creators to generate custom video and image backgrounds for Shorts using text prompts.

• Apple is reportedly developing a next-generation, AI-powered Siri for natural conversations and complex task handling, with an announcement expected in 2025 and a consumer rollout planned for spring 2026.

• Anthropic added Google Docs functionality to Claude's web interface, enabling Pro, Teams, and Enterprise users to seamlessly incorporate documents into conversations and projects.

• Samsung unveiled Gauss2, its next-generation multimodal AI model available in Compact, Balanced, and Supreme versions, offering enhanced language processing and faster response times.

• OpenAI engineers reportedly deleted evidence collected by news organizations in a training data lawsuit, compromising over 150 hours of legal discovery work.

• Salesforce launched Agentforce Testing Center, a platform for enterprises to evaluate AI agents through synthetic interactions, sandbox environments, and comprehensive monitoring tools before deployment.

• MIT researchers introduced a Model-Based Transfer Learning algorithm that improves AI training efficiency by up to 50 times through strategic task selection.

• Cursor is rolling out an autonomous agent system within its code editor, capable of executing terminal commands, completing coding tasks, and independently selecting relevant context.

• Insilico Medicine received FDA clearance for ISM5939, an AI-designed drug targeting solid tumors, marking the company’s tenth AI-designed compound to reach clinical trials via its Pharma.AI platform.

• JPMorgan CEO Jamie Dimon predicted AI advancements will lead to a 3.5-day work week and extend human lifespans to 100 years in a Bloomberg interview.

• xAI’s Grok chatbot introduced personalization features, enabling it to know and remember a user’s name and X (formerly Twitter) handle.

• Nvidia CEO Jensen Huang stated during an interview at the Hong Kong University of Science and Technology that resolving AI hallucination issues will take several years and require significantly more computing power.

• NVIDIA unveiled Fugatto, a 2.5-billion parameter AI sound model capable of generating and transforming music, voices, and audio effects using text prompts and existing audio inputs.

• Researchers used AI and drones to uncover 303 previously unknown Nazca Lines in Peru, doubling the number of known figures and shedding light on ancient sacred spaces and pilgrimage routes.

• U.S. Senator Peter Welch introduced the TRAIN Act, allowing copyright holders to subpoena AI companies' training records if their work is suspected of being used without permission in model development.

• Perplexity partnered with Quartr to provide AI-powered live earnings call analysis, summaries, and qualitative financial research on its platform.

• Intuit added new AI features to QuickBooks, including automated invoice generation, expense categorization, and plans to introduce AI agents capable of performing high-level executive functions.

• Anthropic introduced custom writing styles for Claude, enabling users to select preset tones or train the AI to match their personal communication style using sample text uploads.

• Inflection AI’s new CEO, Sean White, announced a strategic pivot away from next-gen AI model development to focus on enterprise tools, leveraging recent acquisitions.

• Perplexity CEO Aravind Srinivas hinted at plans to create a sub-$50 voice assistant device capable of "reliably answering questions," following significant interest in his social media post.

• Incoming U.S. President Donald Trump reportedly plans to establish a White House AI czar position to coordinate federal AI policy, with input from Elon Musk and Vivek Ramaswamy’s Department of Government Efficiency.

• Uber is launching a Scaled Solutions division to employ contracted gig workers for AI data labeling, serving internal projects and external clients like Aurora and Niantic.

• French startup Mistral AI is expanding into Silicon Valley with a new Palo Alto office and plans to grow its team of U.S.-based AI scientists and engineers.

• OpenAI temporarily suspended access to Sora for beta testers after a leak revealed a public interface to the AI video tool created by a group of unauthorized artists.

• xAI is reportedly preparing to release a standalone app to compete with OpenAI’s ChatGPT, expected to launch as early as December, marking its first product outside the X platform.

• H Company showcased new Runner H demos, highlighting advanced web task automation, including real-time data extraction, complex interface navigation, and precision web scraping.

• ElevenLabs launched GenFM podcasts, enabling users to generate AI-hosted conversations in 32 languages about uploaded content like PDFs, articles, and eBooks.

• Elon Musk announced plans to start an AI game studio with xAI, stating his aim to “make games great again” in a post on X.

• Chinese self-driving startup Pony AI raised $260 million at a $4.5 billion valuation as its U.S. IPO went live for trading this week.

Closing Thoughts

That’s it for us this week.

If you find any value from this newsletter. PLEASE FOR THE LOVE OF GOD pay it forward!

Thank you for being here and if you have another 2 seconds : click the poll below !