Good morning,

This week I'm writing to you from the heart.

For the last few months, I have been locked in a high-speed, feverish romance with Claude. Like many of you reading this.

I was burning through money on the Claude Max plan, and willingly handing over the keys to my entire digital soul. I hooked up my personal email, my calendar, my health metrics, my WhatsApp, and finally offered up the sacred crown jewels: my Obsidian Vault containing years of thoughs and ruminations.

It gave birth to a synthetic, digital golem- i called it my "second brain." Every morning, this silicon organism would crawl through my private inbox, track my bodily metrics from the Garmin watch, polish my schedule, and coach me through my day like a corporate shaman on a retainer.

I felt brilliant.

I even bragged about it in this very newsletter.

But the deeper I went down this rabbit hole of self-quantification, the more a cold, greasy revulsion began to creep up my spine. I took a hard look at the custom apps I had "created," then looked over at what everyone else was building. They all had the exact same hollow, high-gloss sheen. It was digital copy-paste. The pure, unfiltered coding equivalent of AI slop.

Then, one morning, I fired up the machine and ended up feeling downright dirty. "What are we doing ?"

So I did something drastic: I dusted off my old, trusted paper bullet journal. I grabbed a physical pen and started scratching words onto actual dead trees. The moment ink hit paper, I felt an immense, physical wave of relief wash over me. A sudden decompression after months of deep-sea diving in digital sewage. Writing out what I actually experienced during the day, by hand, felt like breaking the surface and drawing a real lungful of air.

And while I was waiting for the ink to dry in order to turn the page I understood. What do the vibe-coding projects and the emails coming from that colleague that has just discovered the technology have in common?

They are utterly and completely devoid of SOUL.

That intangible human spark, the raw edge of genius, the unpredictable twitch of genuine thought. It is all gone. We’ve turned our lives over to a second brain while letting our real, original brains rot in the backseat.

We are floating on an ocean of digital slop, shipping out mountains of mechanized emails devoid of a single drop of human warm-bloodedness, watching our whole culture dissolve into bleak, empty optimization.

On top of that, if EVERYONE at your workplace tomorrow starts to use let's say Claude Fable to run their day-to-day , what nobody realizes is that they have effectively replaced themselves with Claude Fable and they technically don't need to show up to work anymore.

In a cruel twist of irony - wanting to speed up your work and heavily relying on AI might fast-track you towards the exit. While being able to eloquently convey something that you genuinely thought about might propel you towards the board room.

It's only a matter of time until your boss (if you're employed) or your client (if you're a freelancer) finds out he's been paying you for prompting Claude. Might as well hire Claude.

Meanwhile, the hyper-velocity engine of modern life just keeps accelerating. Short-form brain-rot rules the earth. TikTok and YouTube Shorts have conquered the collective consciousness, while long, slow, beautiful cinematic stories and books are left to die in the gutter.

I'm not playing that game.

I’m on the absolute precipice of hitting the kill switch. I'm ready to sever AI entirely from my personal existence. You might have to run with the pack at work just to keep the checks clearing, but while the rest of the world is frantically stomping on the gas toward the cliff, I want to slam on the brakes.

I’m talking about pulling back to dark-room privacy: ProtonMail, locked doors, paper journals, and taking the radical step of sitting alone in a room with your own raw, unvarnished thoughts.

In a world where human attention spans have been hacked down to six-second micro-bursts and your independent thinking is being systematically offloaded to a server farm in Oregon, the ultimate power move isn't speed.

Now, don't get me wrong. AI is magnificent when applied in the correct context : self-driving cars, home domotics, factory automations, ..

But not when it comes to content-generation.

There the real superpower is quiet focus, deep attention, and total thoughtfulness.

Rise up against the machine.

Who's with me ?

Welcome to the Blacklynx Brief

OpenAI's first in-house chip beats Nvidia on its own benchmarks
OpenAI published the first results for Jalapeño, the inference chip it built with Broadcom, showing a 700W part answering up to 3.6x faster than Nvidia's 1,200W flagship at 1.9x more work per watt. OpenAI won't sell it — two more generations are already in design, with production ramping through 2027. (OpenAI)

Chinese state hackers doubled their attack volume using DeepSeek
Taiwanese research firm TeamT5 found China-linked crews running more than twice as many intrusions since folding open-source models into their playbooks, favouring DeepSeek for its low cost and weak guardrails. Tracked groups used it to write break-in code, raid a Taiwanese firm's email, and map a target across 1,000 addresses. (Bloomberg)

SpaceX and Nvidia commit to orbital data centres by late 2027
SpaceX will build its Starmind space data centres around Nvidia's Vera Rubin NVL72 rack, with Elon Musk targeting the first racks in orbit in Q4 2027. Orbital compute currently costs more than 4x ground compute — the bet is that ground buildouts hit political limits first. (Nvidia)

Claude ran protein-design campaigns autonomously and the results held up in the lab
Anthropic showed Claude models running early-stage drug discovery on their own from a single expert prompt, producing working binders on 14 of 15 targets. Hit rates of 22–35% beat the 10–15% industry norm; Twist Bioscience and Adaptyv Bio did the wet-lab validation. (Anthropic)

An anonymous frontier-class model appeared on OpenRouter and nobody claims it
Ox Alpha launched free with a 1M-token context and near-frontier coding scores, scoring 63% on a DeepSWE subset at far fewer tokens per task than Fable 5. Forensics point at China's Zhipu AI — all four anonymous OpenRouter drops in six months have come from Chinese labs. (Bloomberg)

Nvidia pays $6B to build a US answer to Chinese open models
Nvidia licensed Poolside's model-development technology and moved 100+ of its engineers onto the Nemotron team, explicitly to produce competitive open-weight models. The deal reframes open weights as industrial policy rather than community goodwill. (WSJ)

Thomson Reuters built its own legal model for $40M instead of renting one
The Westlaw owner retrained Alibaba's open-source Qwen on decades of proprietary legal content, benchmarking ahead of Claude Opus 4.8, Gemini 3.1 Pro and GPT-5.5 on some internal tests. Its latest training run cost $450k — closer to a year of API invoices than a moonshot. (Thomson Reuters)

That’s it for us this week. Please like and subscribe 🙂