Not Good Enough ... Yet

Good morning,

Welcome from the rollercoaster. What the hell is happening?!

If you have just emerged from a 10-day silent vipassana retreat and this newsletter is the first thing you read, I would say go read up on our extra edition from Monday on all things ‘Deepseek’. (As a bonus - at the end of this newsletter I have included a quick tutorial on how to install Deepseek locally !)

While the debate on what this all means is still raging a week later and the stock market is recovering from its initial shock from this, I would like to talk about something completely different.

At the end of last week, I made a LinkedIn post. It was all about how an entrepreneur that I know is going to pay 10,000€ to a branding agency in exchange for 15 social media posts. A day later I showed him how I could generate 50 posts within the hour using a combination of AI tools.

You can watch a video of me doing that (more or less) here.

Back to that post. Let’s say that the branding agency consultants and digital marketers were “not amused.”.

Their main reaction can be summarized as follows: AI gives crappy results, nobody can replace human creativity, and you’re stupid for even suggesting it.

I was called a shill for something I’m not selling and a prophet of doom. (It’s ok; i’ve been called worse.)

I’ve had some time to formulate a response. So here it is.

WARNING: LONG RANT!

(Perhaps grab a coffee.)

First of all, if you have an emotional response to the advent of AI, as in getting angry, defensive, and defiant, I have news for you: you might be in trouble.

If your first reaction to new technology is to try to dismiss it, you might be in trouble.

We’ve seen it with the advent of DeepSeek this week. Ha, it cannot do this or it cannot do that. Or my favorite: ask it about Tiananmen Square in 1989.

Or: it vacuums up all your data and sends it straight to China! This is not true, by the way; you can run it locally, and it does not phone home.

If you make it political and are looking at the flaws only, you might be in trouble.

All the while you’re missing the point.

The point is not the new model and if it can answer your question or not. It’s how this model came into existence (much cheaper) and how it works internally (much more efficient). And yes, it has its flaws, and it probably stole a lot from other models.

You have to cut through the noise—listen to the facts and realize that the world around us is changing.

There is a huge disconnect between the people that are following this space (and are slowly starting to see where this is going) and people for whom AI is ‘just another tool’ or worse—are blissfully unaware.

The improvement that was made in only a year is mind-blowing.

Dismiss this technology at your own risk.

Sure,it’s not as good as you.

But it WILL be soon.

———

Back to the article and the premise:I could create 50 social media posts for 1 hour of my time versus a brand agency with 15 media posts and 10.000€.

Mostly Inferior—Vastly Quicker

At this moment in time, I agree completely that the output of AI-generated text is mostly inferior to what trained and experienced people are able to bring to the table.

I repeat:it is indeed,for the most part,INFERIOR.

An AI will not write a novel in the vein of Hunter S. Thompson or JRR Tolkien.

Yet!

What an AI has going for it,however, is it works 100 times faster than any human.

For example, say you need inspiration for a blog post. You go look for a popular Twitter thread, copy/paste it and you tell the chatbot to convert it into a blogpost.

You can even train the AI to analyze your writing style and use that prompt to generate posts that look like you.

Then you use that baseline to mold it so it has your personal style written all over it. You make it yours.

It’s Going To Change

What some of you don’t seem to realize is that the AI chatbot you’re using now is the worst it will ever be.

Today, the AI is writing better than the marketing intern; in a few months, it will write better than the experienced copywriter, and in some more months it will reach the level of the very best in the world. The gap is going to get closed.

Some people got quite emotional in the comments, and understandably so.

This is pure disruption at work, and it’s not pretty. And you can tell yourself as much as you want how AI is not able to do what YOU do, that is going to change soon.

But it’s not here yet,which is why many people don’t believe it’s going to happen.

And yes, there’s a chance this whole thing is a bubble and implodes. That is also still very much a possibility.

It has always been possible this goes nowhere, but with the arrival of DeepSeek,the chances of this happening have become a lot slimmer. The barriers to entry have been removed and the deep-pocketed gatekeepers have been shoved aside.

Today, the marketers are angry. Tomorrow it’ll be the truck drivers or taxi drivers who get replaced by self-driving rigs and robotaxi’s.

And might I add for the benefit of society. No more road deaths. No more traffic jams.

Can you imagine the societal uproar once AI gets to the AGI level and becomes better at all of our jobs.

Don’t be arrogant and think that it couldn’t happen.

Don’t think you are irreplaceable.

Just consider the possibility that you too will be forced to leave the workforce one day. Because your replacement is 1. better than you , 2. costs next to nothing 3. doesn’t sleep or go on extended holidays and 4. doesn’t try to unionize against the employer.

Embrace The Change

You need to keep an open mind and just consider the possibility that it might happen.

It takes a few sleepless nights to get to grips with it and then you try to bend it to your advantage. Position yourself where you will be either not replaced or the last to be replaced.

But consider also that by reading this very newsletter you are vastly ahead.

You are keeping your finger on the pulse of revolution.

You’re paying attention.

I think no more than 5% of the general population is paying attention at the moment or even considering the implications.

The warnings are everywhere.

Anthropic’s Dario Amodei claims that the end of all human work arrives in under 3 years and the end of all disease in 5 years.

Have you ever looked up why he’s claiming this?

That’s the kind of stuff we’re doing in this newsletter.

Welcome to the Blacklynx Brief!

BUT … WE’RE NOT FINISHED

To close out this post … I have an offer for you.

If you want to take this a step further. Recently I gave a few “inspiration”-sessions for companies where most people are in “knowledge work”.

In these sessions I show people what AI is, what you can do with it and what you can’t. These sessions really wake up people to the possibilities of the future and are fun for me as I can see people’s head explode (not literally of course) . On top of that - it includes some consulting where I can really help in making your company more efficient in a very short time.

If you want me to come do a workshop in your company or a keynote - click the button below and let’s have a quick little chat about it.

AI News

• OpenAI launched Operator, an autonomous AI agent capable of navigating web browsers to complete tasks like booking reservations and ordering groceries. Built on a new Computer-Using Agent model, Operator integrates GPT-4o’s vision with advanced reasoning and partners with DoorDash, Instacart, and Uber for seamless interactions. Currently in a U.S. Pro user preview, this release marks OpenAI’s first major move into agentic AI, shifting how we interact with AI assistants.

• Perplexity launched Perplexity Assistant, a voice- and gesture-controlled AI tool that interacts with Android apps like Uber and OpenTable to complete real-world tasks. The assistant maintains context across interactions and supports multimodal inputs using voice and camera, allowing for advanced research-to-action workflows. By letting users replace Google’s default assistant for free, Perplexity is directly competing in the growing AI agent space.

• The Center for AI Safety and Scale AI launched Humanity’s Last Exam (HLE), a 3,000-question benchmark covering 100+ subjects to test the limits of AI knowledge. Even top models currently score below 10%, with questions designed by 500+ institutions worldwide, some requiring multimodal reasoning. With AI rapidly surpassing existing academic tests, HLE aims to set a new standard for evaluating advanced AI systems.

• Meta is investing $60-65B in AI infrastructure for 2025, planning to deploy 1.3M GPUs and build a 1GW data center—one of the largest globally. This represents a 70% increase from 2024’s spending, with Meta AI targeting 1B users and Llama 4 aiming to be the industry’s top open-source model. As OpenAI’s $500B Stargate Project and DeepSeek’s R1 reshape AI, Meta is aggressively scaling to stay in the race.

• Alibaba’s Qwen2.5-1M models (7B and 14B parameters) support 1M-token context windows, outperforming rivals like Llama-3 and GPT-4 in long-text processing. Using a custom vLLM-inference framework, Qwen models run 7x faster than existing long-context AI. With Google, OpenAI, and DeepSeek also pushing ultra-long context, the ability to process massive data streams is becoming the next competitive frontier.

• Perplexity AI revised its TikTok U.S. merger proposal, offering the U.S. government up to 50% ownership of a newly formed "NewCo" valued at up to $300B post-IPO. ByteDance would contribute TikTok’s U.S. operations while retaining its recommendation algorithm, though rivals Elon Musk, Microsoft, and Oracle are also exploring bids. With Trump temporarily lifting the ban for 75 days, the race for TikTok’s future is heating up.

• DeepSeek unveiled Janus-Pro, a multimodal AI model that outperforms DALL-E 3 and Stable Diffusion on benchmarks like GenEval and DPG-Bench. Available in 1B and 7B parameter versions under an MIT license, Janus-Pro follows DeepSeek’s viral R1 release, which disrupted AI cost assumptions with o1-level reasoning at lower expenses.

• Alibaba’s Qwen2.5-VL (72B parameters) surpasses GPT-4o and Claude 3.5 Sonnet in document parsing and video understanding while enabling AI control over apps. The model can analyze hour-long videos, extract key moments, and perform tasks like airfare booking and image editing. With Qwen and DeepSeek rapidly advancing, the gap between open-source and closed AI—and China vs. U.S. capabilities—continues to narrow.

• Meta’s AI assistant now remembers past conversations and integrates user data from Facebook, Instagram, and WhatsApp for tailored recommendations. Features include dietary preference tracking, location-based suggestions, and content personalization, but users cannot fully opt out—only manually delete memories. While Meta’s vast social data could enhance AI personalization, privacy concerns loom over the lack of user control.

• OpenAI introduced ChatGPT Gov, a secure AI platform for U.S. agencies, offering access to GPT-4o within Azure environments for handling sensitive data. Agencies can deploy custom GPTs, conversation sharing, and admin controls, with over 90K government employees across 3,500 agencies already using ChatGPT. This move signals a broader integration of AI in federal operations, ensuring security while accelerating adoption.

• Jack Dorsey’s Block launched Goose, an open-source AI agent framework supporting multiple LLMs, including OpenAI, DeepSeek, and Anthropic, while prioritizing data privacy. Goose integrates with APIs like Anthropic’s MCP, enabling live tool connections and mid-session modifications, with early Block tests automating code migrations and test generation. This release brings the open-source movement to AI agents, making automation more accessible and customizable.

• HKUST unveiled YuE, an AI music system that transforms lyrics into full-length songs, rivaling commercial platforms like Suno and Udio. The model supports multiple languages, complex vocal techniques, and customizable production elements, generating songs up to 5 minutes long. With AI music facing legal battles, YuE’s open-source approach could disrupt the industry by democratizing access to advanced AI-powered music creation.

• The U.S. Copyright Office ruled that AI-generated content cannot receive copyright protection unless it includes meaningful human authorship. Simply prompting an AI doesn’t qualify for copyright, but works combining AI and human-created elements can be partially protected. This decision clarifies legal boundaries for creators and businesses, balancing AI integration with human intellectual property rights.

• Anthropic CEO Dario Amodei dismissed DeepSeek R1 as an expected cost reduction rather than a major breakthrough, arguing it only matches the U.S. from months ago. He revealed Claude 3.5 Sonnet was trained for tens of millions, challenging DeepSeek’s $6M efficiency claim, and projected superintelligence will require tens of billions in investment by 2026-2027. His comments suggest chip controls are slowing China’s AI progress despite media hype.

• Chinese robotics firm Unitree wowed audiences with 16 synchronized humanoid robots performing complex folk dance routines using AI motion control and 3D laser SLAM technology. The robots, equipped with 360° panoramic depth awareness, adapted their movements to music in real time. As humanoid robots achieve greater dexterity, coordination, and real-world adaptability, their integration into society is rapidly approaching.

Quickfire News

• Anthropic introduced Citations in the Claude API, enabling automated source attribution and verification to improve response accuracy.

• Google’s Imagen 3.0 debuted at No. 1 in the LM Text-to-Image Arena, securing Google the top spot on both image and LLM leaderboards.

• ByteDance plans to invest $20 billion in AI infrastructure in 2025, with half allocated to international data centers and partnerships with chip suppliers.

• OpenAI CEO Sam Altman announced that the upcoming o3-mini model upgrade will be available in ChatGPT’s free tier, with usage expansions for Plus subscribers.

• Hugging Face launched SmolVLM 256M and 500M, the world’s smallest vision-language models, which maintain competitive performance against significantly larger rivals.

• LinkedIn is facing a class-action lawsuit alleging that the company used premium subscribers' private messages to train AI models.

• ElevenLabs is reportedly raising a $250 million Series C at a $3 billion+ valuation, driven by increasing demand for its AI voice synthesis and dubbing technology.

• Anthropic CEO Dario Amodei predicted that AI could double human lifespans by 2030, accelerating a century’s worth of research progress into just 5-10 years.

• xAI is developing a voice interface for its Grok iOS app, featuring both proprietary and ElevenLabs voice options, with real-time data integration.

• OpenAI expanded Canvas functionality in ChatGPT, adding new rendering capabilities and o1 model support, while also rolling out desktop app access to all subscription tiers.

• Paul McCartney criticized proposed UK AI copyright law changes, warning they could allow AI to "rip off" musicians without proper compensation.

• OpenAI CEO Sam Altman stated that AI advancements will eventually require changes to the social contract, with "the whole structure of society up for debate and reconfiguration."

• LinkedIn co-founder Reid Hoffman announced a $24.6 million raise for Manas AI, an AI-powered drug discovery platform focused on cancer treatments.

• DeepSeek’s R1 app surged to No. 1 on Apple’s App Store with 2.6 million downloads but has paused new user sign-ups outside China due to reported cyberattacks.

• xAI’s Grok-3 briefly appeared live for some users, showing improved reasoning capabilities, with a full release expected later this week.

• Pika Labs launched v2.1 of its AI video generation model, introducing advanced motion control, realistic physics, and customizable scene elements.

• Apple released iOS 18.3, enabling Apple Intelligence by default, though AI summaries remain disabled.

• The French government suspended its AI chatbot Lucie just days after launch due to widespread factual errors, despite backing from President Macron.

• Alibaba’s Qwen introduced Qwen2.5-Max, a large-scale MoE model that surpasses DeepSeek-V3, GPT-4o, and Claude 3.5 Sonnet on key AI benchmarks.

• OpenAI CEO Sam Altman praised DeepSeek’s R1 release, calling it an impressive model and saying it’s "invigorating" to have a new competitor in the space.

• Figure AI launched the Center for Advancement of Humanoid Safety to establish industry-wide testing standards and release quarterly safety reports for workplace robots.

• Former OpenAI safety researcher Steven Adler expressed concerns on X about the rapid pace of AI development, stating that no lab currently has a viable solution for alignment.

• Convergence AI introduced Proxy, a natural language agent positioned as "Europe’s answer to OpenAI’s Operator."

• Hugging Face integrated four new serverless inference providers — fal, Replicate, Sambanova, and Together AI — enabling direct model deployment and faster inference speeds.

• Microsoft is investigating potential unauthorized data collection from OpenAI’s API by a DeepSeek-linked group, while U.S. AI czar David Sacks claimed there is "substantial evidence" that DeepSeek used OpenAI’s models for training.

• Microsoft AI CEO Mustafa Suleyman announced that the "Think Deeper" feature, powered by OpenAI's o1 reasoning model, is now free for all Copilot users.

• Luma Labs introduced a new "Upscale to 4K" feature, allowing video generations on its Dream Machine platform to be enhanced to 4K resolution.

• The U.S. Navy banned its members from using DeepSeek for both work and personal use, citing security and ethical concerns.

• The "Doomsday Clock" moved to a record 89 seconds to midnight, with scientists pointing to AI-powered military operations as a growing global threat.

• Ragon Institute and MIT scientists unveiled MUNIS, an AI tool that accurately identifies viral targets to accelerate vaccine design and outperform traditional lab methods.

Like newsletters?

Here are some newsletters our readers also enjoy:

Tip of the Week

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 !