Geeky Weekend Digest | 12.07.2025
In this weekend’s Geeky Weekend Digest: Grok 4 takes the AI crown, OpenAI eyes Chrome, BYD charges ahead—and psilocybin defies ageing.
Welcome to this week’s Geeky Weekend Digest — your curated lens on the accelerating convergence of artificial and human intelligence. As models grow more agentic, browsers become co-pilots, and even psychedelics enter the longevity arena, we explore how technology is reshaping not just the world around us, but our inner architecture.
Here, we cut through the noise to reveal the deeper shifts defining our techno-cultural moment. From recursive AI dreams to electric breakthroughs and cellular awakenings, this edition invites you to witness the contours of a future where intelligence evolves on both sides of the screen.
AI News | Grok 4 Ascends - xAI’s Supermodel Surpasses All Rivals
In a bold leap forward, Elon Musk’s xAI has unveiled Grok 4, the first model from the company to decisively take the lead in AI benchmark performance. More than just faster or more powerful, Grok 4 introduces a “multi-agent” heavy variant that allows models to collaborate in real time, achieving a level of reasoning and accuracy that now sets the gold standard across the board. This marks xAI’s arrival at the frontier, reshaping the landscape of large language models and igniting a fresh phase in the AI race.
Benchmark Dominance:
Grok 4 leads the Artificial Analysis Intelligence Index with a score of 73, outperforming OpenAI’s o3 (70), Gemini 2.5 Pro (70), Claude 4 Opus (64), and DeepSeek R1 (68).Crushing Complex Exams:
Grok 4 achieves record scores in Humanity’s Last Exam (24% without tools, 41% with), surpassing previous leaders by wide margins. It also dominates AIME 2024, MMLU-Pro, and GPQA Diamond.A New Type of Model:
The “Grok Heavy” version enables multiple agents to solve problems together and compare outputs — a novel approach to ensemble reasoning that’s already showing transformative results.Math, Coding, and More:
Grok 4 tops our Math Index and LiveCodeBench, outperforming Claude 4 and Gemini in both scientific and programming reasoning tasks.Contextual Power:
With a 256k token context window, Grok 4 balances capacity and usability better than many competitors, including Claude 4 and o3.Visual and Voice Upgrades:
Image input is supported, and Grok 4’s voice interface now offers dramatically lower latency and improved naturalness.Pricing and Access:
Standard API pricing matches Grok 3 and Claude Sonnet, while the new $300/month SuperGrok Heavy tier offers elite access to the full model range and premium features.
Why It Matters
Grok 4 is not just another increment. It represents a genuine paradigm shift — combining high reasoning fidelity, collaborative agent architecture, and human-like voice interaction in a single platform. With benchmark results consistently topping the charts, xAI has finally broken through the dominance of OpenAI, Google, and Anthropic, signalling a dramatic change in the competitive dynamics of frontier AI development. Grok’s multi-agent collaboration could hint at the future direction of artificial general intelligence (AGI): decentralised cognition with layered reasoning, more akin to a team of thinkers than a solo savant.
Closing Thoughts
The arrival of Grok 4 cements xAI as a serious contender in the AI arms race — and not just for show. By leaning into reasoning over speed and collaborative problem-solving over monolithic intelligence, Musk’s team has delivered something truly novel. Whether this is the foundation of the next phase of AGI or merely a milestone in the long journey ahead, it’s clear that the landscape has changed. And with OpenAI, Google, and Anthropic surely planning their next move, the story of 2025’s AI race is just beginning to unfold.
Interface Revolution | The Battle for the AI Browser Begins: OpenAI vs Comet vs Chrome
OpenAI is poised to shake up yet another digital stronghold: the web browser. Leaks suggest the company is on the verge of launching an AI-powered browser that blends intelligent search, multi-step task automation, and conversational navigation—all within a ChatGPT-style interface. This would place it in direct competition with Google Chrome, currently used by over 3 billion people. But OpenAI isn’t alone in this new frontier. Perplexity has already launched Comet, a sleek, agent-first browser designed for users who want the internet to work for them, not the other way around.
OpenAI’s Browser Incoming:
Insider sources indicate OpenAI’s AI browser could debut within weeks, integrating an upgraded version of its “Operator” agent to perform real-world tasks like booking, autofilling, and making reservations.Beyond Search:
The browser may keep users inside the ChatGPT ecosystem, surfacing content and completing tasks without redirecting to external websites. This is a clear move to redefine web navigation as a conversation.Built on Chromium, Powered by Reasoning:
With former Chrome engineers onboard, OpenAI is likely building on Chromium for compatibility while embedding native agentic intelligence. A lightweight, open-source reasoning model similar to o3-mini is also expected to arrive next week.Perplexity’s Comet Launches First:
Comet, an AI-first browser with a built-in assistant sidebar, lets users “vibe browse” using voice or text commands. It integrates seamlessly with calendars, emails, and existing workflows, and is currently rolling out to Perplexity Max subscribers.A Hardware Future Too:
In parallel, OpenAI has finalised its $6.5B merger with Jony Ive’s design firm, laying the groundwork for its first AI-native device—a potential iPhone moment for generative agents.
Why It Matters
The browser wars are entering a radically new phase. No longer just gateways to the web, tomorrow’s browsers are becoming action platforms—personal agents that read, understand, and act on your behalf. OpenAI and Perplexity are leading this shift by embedding reasoning and agency into the core browsing experience. Meanwhile, Chrome’s reign faces a critical test: can it pivot quickly enough to stay relevant in the era of AI-first interaction? The competition is about far more than search—it’s about who controls the next interface layer of the internet.
Closing Thoughts
If ChatGPT redefined the future of questions, OpenAI’s browser may redefine the future of doing. Alongside Perplexity’s Comet, these early AI-first browsers are glimpses into an evolving paradigm, where the web adapts to users, not the other way around. It’s a timely reminder that in the age of AI, even the most established technologies can become fertile ground for reinvention.
Behind the Curtain | Recursive Dreams & Reasoning Limits: What AI Still Can’t Do
As excitement around Recursive Self-Improving AI (RSI) heats up, a closer look reveals a growing divide between public perception and actual capability. Recent papers from Google DeepMind, Anthropic, and Apple offer radically different narratives—one leaning into the dream of meta-learning machines designing themselves, the other pulling the curtain back on the illusion of "reasoning" itself. With AI’s most ambitious claims increasingly shaped by PR as much as research, these papers provide a rare lens into what’s real, what’s theoretical, and what remains pure projection.
RSI Hype Peaks with AlphaEvolve:
Google DeepMind’s AlphaEvolve paper describes a Gemini-powered coding agent that designs new algorithms and is already deployed across Google’s infrastructure. It hints at an early form of RSI—systems that evolve their own capabilities via feedback loops.Claude Code’s Self-Writing Claim:
Anthropic made waves suggesting 80% of Claude Code’s upgrades are generated by the model itself, reinforcing a narrative of self-directed learning that borders on sentient mythos.Venture Capital and the “Gentle Singularity”:
Sam Altman’s vision for a “gentle singularity” casts RSI as an inevitable arc of progress, justifying billions in investment. But critics argue this positions token prediction as superintelligence—an anthropomorphic leap that borders on magical thinking.Apple Pushes Back on the Illusion of Reasoning:
Apple’s recent study delivered a sobering counterpoint. By testing top models (from OpenAI, DeepSeek, and Anthropic) on novel puzzle games, researchers showed a clear “reasoning collapse” at higher complexities. Beyond a certain point, performance tanks—even when token budgets are sufficient.What the Findings Reveal:
Apple’s results show that reasoning, as currently implemented in LLMs, lacks robustness and flexibility. When presented with unfamiliar problems, AI behaves less like a thinker and more like a parrot with a good memory. Apple critiques common benchmarks for failing to account for this, often inflating capabilities via data contamination and overfitting.
Why It Matters
We are in a pivotal moment where the mythos of AGI risks overtaking the mechanics of actual machine learning. The idea that AI can improve itself recursively is seductive, but it still rests on fragile architecture, limited transfer learning, and models that struggle with novelty. Meanwhile, the illusion of reasoning persists in marketing, even as empirical studies like Apple’s expose its structural limits. These contradictions matter: they shape public trust, investment trajectories, and how we prepare for the actual capabilities and risks of AI.
Closing Thoughts
The journey toward RSI may one day yield systems that can meaningfully improve themselves, but today’s architectures are not there yet. The excitement is understandable, but clarity is crucial. The divide between how AI is marketed and how it truly behaves under pressure is widening. Reading between the lines of these new papers reveals a truth both more humbling and more fascinating than any hype: AI is still just beginning to learn how to think.
Recommended Reading:
For a deeper dive into these emerging contradictions in AI development, we highly recommend AI Papers on the Spectrum of RSI and the Illusion of Thinking via AI Supremacy on Substack. This piece explores how recent papers are being used to project ideological positions across the AI hype cycle—from bold claims about self-improving models like AlphaEvolve, to Apple’s more sceptical take on the limits of reasoning itself. It’s a timely, sharp analysis of where research ends and narrative engineering begins.
Read it here: AI Papers on the Spectrum of RSI and the Illusion of Thinking
A Dose of Velocity for Your Weekend | Because Charging Anxiety Is About to Become Obsolete
China just hit the fast-forward button on the EV revolution. BYD, now the world’s largest electric vehicle maker, has unveiled a megawatt charging system capable of delivering up to 400 kilometres of range in just five minutes—an unprecedented leap that finally rivals the convenience of petrol refuelling. Demonstrated at the Shanghai auto show, this breakthrough makes previous fast-charging claims look quaint by comparison, and positions BYD as a global leader not just in EV sales, but in the infrastructure that will define the future of mobility.
With its vertically integrated approach—designing its own batteries, chips, chargers, and vehicles—BYD has moved the entire industry’s benchmark. Its new Flash Charging Battery and 1,000-volt platform outpace even the most advanced Western systems, while the company’s commitment to scale means thousands of these ultra-fast chargers are already in the pipeline. If ever there were a turning point for electric transportation, this is it.
Why it matters isn’t just about speed. It’s about vision. BYD’s ecosystem is proof that when engineering ambition meets national support, transformation follows. The next time someone questions whether EVs can truly compete with internal combustion, point them to China’s answer: Yes—and faster.
One More Thing…
🍄 Psychedelics and Longevity: Could Psilocybin Be the Next Anti-Ageing Breakthrough?
A surprising twist in the longevity research landscape: psilocybin, the active ingredient in magic mushrooms, may do more than expand consciousness—it might extend life itself. A new preclinical study published in npj Ageing reports that psilocin, the bioactive form of psilocybin, extended the lifespan of human fibroblast cells by up to 57% while maintaining telomere integrity. Even more astonishing, aged mice treated with monthly doses of psilocybin showed a 60% higher survival rate, along with noticeable improvements in coat condition—a classic biomarker of vitality in rodents.
The research proposes that psilocybin’s benefits might stem from its impact on SIRT1 activation, reduced oxidative stress, and the preservation of DNA stability, shifting the narrative from mind-altering compound to cellular rejuvenator. Lead scientist Dr Louise Hecker speculates that this “psilocybin-telomere hypothesis” may help explain the compound’s broad therapeutic effects, from mental health to neurodegeneration.
If proven safe and effective in humans, psilocybin could join the ranks of potential geroprotective agents—substances that actively slow the biological ageing process. Still, experts caution this research is in its early days, and much more investigation is needed before magic mushrooms enter the longevity mainstream.
Why It Matters
This breakthrough repositions psychedelics not just as tools for mental healing but as possible biological enhancers. In the quest for healthspan over lifespan, few compounds show such multi-system promise. If further studies validate these findings, we may be on the cusp of a profound new chapter in human vitality, where the fountain of youth might just be growing in the forest floor.
Read more:
🧬 StudyFinds Summary
🧪 Technology Networks Coverage
Disclaimer: The study cited here is preclinical and based on lab cells and animal models. Psilocybin remains a controlled substance in many countries and is not approved for anti-ageing use outside of regulated clinical trials.
🌀 Final Wrap-Up
In this edition of Geeky Weekend Digest, we explored the escalating AI model wars as xAI’s Grok 4 claimed the top benchmark spot, setting a new standard for multi-agent reasoning. We then dived into the brewing browser revolution, with OpenAI preparing to take on Chrome and Perplexity’s Comet reimagining how we interact with the web. In our third feature, we unpacked the illusion of Recursive Self-Improvement (RSI) and the growing debate around whether current models truly “think.” We closed with BYD’s record-shattering megawatt EV charging tech—and a psychedelic surprise hinting at an anti-ageing future powered by psilocybin.
💬 What Are Your Thoughts?
Which breakthrough stood out the most to you—Grok 4’s reasoning leap, the AI browser battle, or the idea that magic mushrooms might one day extend human life?
Let’s keep the conversation alive!
🌐 Follow my journey on Notes and 𝕏 @frankdasilva for real-time insights and bold ideas, and explore my broader work at Frank Da Silva – Living the Future.
🔥 Summer Special – 25% Off for 12 Months!
Subscribe now and get full access to MyGeekSpace for a whole year at 25% off. The offer is valid all summer long for new subscribers. Let’s #LiveTheFuture together. If you’ve found value in this journey, please consider supporting MyGeekSpace as a paid subscriber.
