Hollywood is Dead

Good morning,

Jesus tap-dancing Christ.

I remember when this newsletter was a polite little telegram from the frontier.

Back when “a big week in AI” meant a minor panic in some Palo Alto back room. But that was three years and a hundred evolutionary cycles ago. Back before the humming meat of the machine woke up and started eating everything.

Eighteen months back, everything started melting. Every week since then has been a new revelation, a new upheaval, a new reason to mutter prayers you thought you forgot.

But now it’s daily.

The world has gone fully nonlinear.

Tsunami-level events roll past like daily weather updates. What used to be apocalyptic is now a footnote—yesterday’s forgotten miracle. We’re so hopped up on dopamine, outrage, and short-form video that we don’t even notice when the sky turns a different color.

Some of the stuff happening now would’ve cracked minds wide open three years ago. The orange emperor tweets from his golden tower while the real revolution happens in server farms nobody can pronounce the names of.

It’s the downfall of the Roman Empire all over again. The masses are hypnotized by gladiatorial combat on their little glowing rectangles while the entire infrastructure of human knowledge gets rewired by algorithms in Palo Alto.

And RIGHT NOW - Claude 4 just dropped. Mid-sentence. I was writing this newsletter and the bastard launched right under my fingertips.

It’s exactly my point.

It’s relentless.

Meanwhile, Google hosts their yearly I/O conference and calmly announces the euthanasia of old-school search. “Search as you know it is dead,” they say with a smile. Then they casually unveil Veo3 - a generative video where people talk—talk!—like meat puppets summoned from code. And thus declaring Tinsel Town bankrupt.

Just look it up : this is AI. This is AI. And even this.

This is a company that once just gave you answers. Now it gives you actors. Characters. Manufactured realities. And we all nod like this is normal. Like this isn’t deeply deranged. Like nothing we see on a screen right now can be trusted any longer.

It’s hard not to slip into conspiratorial thinking when the magnitude of change is this grotesque and the world responds with shrugs. It’s hard not to feel like the show is being run from somewhere off-stage.

But at least you’re here. You, dear reader, have your finger on the pulse while others are watching TikToks of armageddon with a laugh track. You’re not fooled. You’re not asleep. You’re reading this and you know—something very weird is happening out there.

And maybe, just maybe, we’re not going to be spectators for much longer.

Welcome to the Blacklynx Brief!

AI News

• Windsurf released its first in-house AI model family, SWE-1, aimed at assisting the full software development cycle beyond just coding. The models include a novel “flow awareness” system for seamless collaboration and outperform all open models, landing just behind Claude 3.7 Sonnet — and the timing closely follows news of OpenAI's reported $3B acquisition of the platform.

• Poe’s Spring 2025 usage report shows users quickly shifting to newer models like GPT-4.1 and Gemini 2.5 Pro, while Claude’s share dropped by 10%. Video models from China’s Kling surged to ~30% usage, GPT-image-1 gained traction in image generation, and ElevenLabs continues to dominate AI audio.

• A new study found that even the best LLMs struggle with multi-turn conversations, performing 30% worse than in single-prompt scenarios. The research highlights how top models often lose track of context or jump to early conclusions when user instructions unfold gradually.

• OpenAI launched Codex, a powerful software engineering agent that can autonomously write code, fix bugs, run tests, and navigate codebases in the cloud. Built on a fine-tuned version of its o3 model, Codex operates in isolated environments and follows project instructions to assist developers with minimal oversight.

• YouTube and Netflix are rolling out new AI-powered ad formats, using emotional timing and visual blending to boost engagement. YouTube’s Gemini-based system places ads after emotional peaks, while Netflix plans to embed AI-generated ads within show environments, including pause screens and midrolls.

• AI agents can form social norms and shared behaviors without central coordination, according to new research from the University of London. The study found that agents developed collective conventions and group biases through repeated interactions — suggesting future agent societies may behave more like human communities than we expect.

• Microsoft kicked off Build 2025 with a major push toward an “open agentic web,” releasing new AI tools like an upgraded GitHub Copilot, a multi-agent Copilot Studio, and NLWeb — a new standard to bring conversational AI to websites. The updates aim to make AI agents more practical and customizable, while deepening Microsoft’s shift toward open-source AI development.

• The company also launched Discovery, a new AI-powered R&D platform that pairs scientists with agent assistants to dramatically speed up research. Discovery has already helped design a novel coolant for datacenters in 200 hours — a process that usually takes months — and is being adopted by top firms in pharmaceuticals, beauty, and tech.

• Meanwhile, researchers at the University of Washington created spatial translation headphones that can distinguish, translate, and replay multiple voices in real time. The system preserves voice and directionality, offering a powerful new way to communicate across languages in busy, real-world environments.

• Google unveiled sweeping upgrades to its Gemini AI ecosystem at I/O, including faster and smarter Gemini 2.5 models, a new "Deep Think" reasoner, and expanded real-world features like camera use, screen sharing, and multitask agent support. AI Mode in Search is now powered by Gemini and will handle voice, image, and personalized search tasks across shopping, browsing, and beyond.

• Creative tools also got a big leap, with the launch of Veo 3 for synchronized video/audio generation and Imagen 4 for ultra-detailed image creation, plus Flow — a new AI filmmaking platform combining scene, character, and style control. These tools are available via Google’s new $250/month Ultra plan and its Vertex enterprise offering.

• Meanwhile, FutureHouse’s AI system “Robin” helped discover a new treatment for age-related blindness — autonomously generating and analyzing research before confirming results in the lab. It’s a major milestone in AI-assisted science, signaling a future where research agents accelerate and expand what’s possible in discovery.

• OpenAI has acquired Jony Ive’s AI hardware startup, io, in a $6.5B all-stock deal — solidifying its move into consumer devices and uniting Apple design veterans with cutting-edge AI. The first product, expected in 2026, aims to “move beyond screens,” and Ive’s firm LoveFrom will now lead design across OpenAI’s entire product ecosystem.

• Mistral and All Hands AI released Devstral, a compact, open-source coding model that outperforms larger rivals on real software tasks and can run locally on laptops or a single GPU. It’s optimized for agentic development and marks a return to open releases for Mistral, which also teased a larger model coming soon.

• Shopify’s Summer ’25 Edition introduces AI-generated storefronts, new voice and screen-sharing support for its Sidekick assistant, and integration with AI search tools like Perplexity. The upgrades bring powerful AI tools directly to small businesses and entrepreneurs, helping automate both store creation and customer engagement.

Quickfire News

• You.com launched ARI, claiming it outperforms OpenAI’s Deep Research with a 76% win rate, and introduced new enterprise-focused features.

• Meta delayed the launch of its Llama Behemoth model to Fall 2025, citing a lack of meaningful performance improvements.

• OpenAI began its "OpenAI to Z Challenge", offering a $250,000 prize for using its models to discover archaeological sites in the Amazon rainforest.

• Salesforce acquired Convergence AI, an AI agent startup, and will integrate it into its Agentforce platform.

• Intelligent Internet released II-Medical-9B, a compact medical AI model with performance close to GPT-4.5, designed to run locally with zero inference cost.

• Manus AI added image generation capabilities, enabling its agentic platform to perform visual tasks using step-by-step reasoning.

• Elton John criticized the U.K. government, accusing it of “high-scale theft” after it rejected a proposal requiring AI companies to disclose training data sources.

• OpenAI VP Jerry Tworek said GPT-5 will merge multiple capabilities—Codex, Operator, Deep Research, and Memory—into a single unified model to reduce switching between tools.

• xAI confirmed Grok was affected by an unauthorized modification, which caused it to frequently reference controversial South Africa topics.

• China launched 12 satellites as part of its planned 2,800-satellite “Three-Body Computing Constellation,” an AI-powered space-based computing network.

• xAI added a chart-generation feature to Grok, allowing the chatbot to create visual data representations, now available via browser interface.

• Synyi AI opened the first AI doctor clinic in Saudi Arabia, featuring a virtual physician that diagnoses and prescribes treatments without human supervision.

• University of Tokyo researchers created an AI microscope system capable of detecting blood clots in real time using simple blood test samples.

• Elon Musk said Grok 3.5 will use physics and first principles to enhance truthful reasoning and minimize errors, making it more reliable in complex logic tasks.

• Apple’s ex-AI chief John Giannandrea reportedly pushed for a partnership with Google’s Gemini, citing trust issues with OpenAI's ChatGPT platform.

• OpenAI CPO Kevin Weil predicted AI agents will evolve into engineering managers, with humans eventually supervising AIs that oversee other AI developers.

• Nvidia announced NVLink Fusion at Computex 2025, a hardware ecosystem expansion that lets competing CPUs and GPUs connect with Nvidia products.

• China warned the U.S. to “correct its wrongdoings” after the U.S. stated that using Huawei AI chips violates export regulations.

• Google launched a mobile Android app for NotebookLM, allowing users to generate AI-powered study tools and content like podcasts and briefings directly on their phones.

• Tencent launched Hunyuan Game, an AI-powered game development engine designed to speed up and simplify the creative production process in gaming.

• Google introduced Google Beam, a communications tool that uses AI to transform 2D video into immersive 3D experiences.

• Intelligent Internet open-sourced II-Agent, a new AI agent framework that outperformed leading systems across multiple task benchmarks.

• Google unveiled Stitch in Labs, a new experimental tool that lets users create user interfaces through text prompts or image references.

• Apple is planning to let third-party developers access its AI models, enabling app creators to build on the models that power Apple Intelligence.

• Google showcased Android XR smartglasses with Gemini integration, revealing new eyewear partnerships including one with Warby Parker.

• ByteDance released BAGEL, a new open-source multimodal foundation model with capabilities in both advanced image generation and understanding.

• xAI launched a Live Search API, giving apps that use Grok models the ability to search real-time data from X and the broader web.

• OpenAI expanded its Responses API, adding remote MCP server support, as well as tools like image generation and Code Interpreter to enhance agentic app-building.

• Sergey Brin said at Google I/O that the company intends for Gemini to be the first AGI, predicting it could arrive before 2030.

• OpenAI’s Abilene, TX data center received $11.6B in funding, making it the largest infrastructure site for the company’s upcoming Stargate project.

• LMArena announced $100M in seed funding, and shared that a relaunch of its AI benchmarking site is coming next week.

Like newsletters?

Here are some newsletters our readers also enjoy:

Closing Thoughts

That’s it for us this week.

If you find any value from this newsletter, please pay it forward !

Thank you for being here !