"All Human Work Obsolete in 3 Years"

Good morning,

First of all : grab yourself a coffee. This is a long one.

(sorry)

I’ve been writing this newsletter for almost two years now. There was 1 week where there was hardly any AI news. That was more than a year ago—somewhere near the end of 2023.

Since then a non-stop barrage of releases and evolutions.

A Hell of A Week

I used to start out the newsletter with “another crazy week” or something similar. I can honestly say that this past week has been INSANE. The most crazy of them all.

We surpassed the 1000 subscribers this week, and I wanted to make it about that. But the news simply caught up with me. And to be fair, it was not only the news, but I had one of those old-school "aha moments.”.

I’ve considered the viewpoint of the AI haters in this newsletter, saying it’s all a bubble, and since this week I’m firmly dismissing their opinion.

They are wrong.

Artificial general intelligence is happening, and the singularity is happening as well. Sooner than we might think.

Tweets like this are showing up:

I want to write about everything that happened this week—about the political ramifications. About what’s going to happen to our jobs. To our society.

But even if I switched to a daily newsletter, I couldn’t cover the intricacies of what has all happened this week.

What Happened

So what happened this week that I am posting a little more panicked and excited than last week?

A lot happened but these things matter most. The implications of these two things put together has me now convinced Skynet is coming:

• The Chinese company DeepSeek released an open source AI model that you can run locally. It performs better than OpenAI’s o1 Pro (the one you pay 200$ per month for) without needing nuclear-powered datacenters. It’s capable of reasoning and improving itself. Locally. The implications are enormous. This means OpenAI might be IN TROUBLE. Why pay OpenAI for something you can run on your laptop? Only a day before the release of Deepseek there was an article in The Economist talking about how the “thinking” the new models do costs a lot of money and it will make sure that powerful AI is for the few, not for the masses.
  
  And then a Chinese lab that doesn’t even have AI research as their main activity releases something that completely democratizes AI.
  

  LAST MINUTE UPDATE !! : This newsletter was already scheduled and it looks like OpenAI got spooked and quickly released their new AI Agent tool called Operator. I’ll look at that next week. Feels underwhelming at the moment. Anyway, on with the show.

  As a life-long cybersecurity professional I have a deep distrust of everything Chinese (I don’t even want my solar panels to run on Huawei hardware) but this seems like a magical breakthrough. And why did they release it ? What else are they sitting on over there ?

• On the other hand , the US announced Project Stargate - 500 billion dollars (BILLION with a B) will be invested in building 10 huge server farms in Texas in collaboration with Oracle, NVIDIA and OpenAI. It cannot be overstated how important this is. This is the brain for the singularity being built.

This is what my favorite AI thought leader Dave Shapiro has to say about this:

These two evolutions together: giant amounts of “compute” in combination with a solution to running these models without giant costs is going to change the world

Pandora’s Box has been opened

On top of that I looked further into OpenAI’s o3 model.

These new family of models can "generalize outside their training distribution" - meaning they can solve complex problems they weren't explicitly trained on, similar to human reasoning. Through an innovative process of iterative training and computation, these models can apparently synthesize new insights from first principles rather than just recalling memorized information, with benchmarks suggesting they're outperforming human domain experts even on highly specialized questions. This development signals a potential shift toward "cognitive hyperabundance," where human cognitive limitations may no longer constrain scientific and economic progress, much like how machines overcame the boundaries of human physical strength.

In other words, the criticism towards AI and LLMs was that it was just a prediction machine. It couldn’t think or reason. With o3, it can not only do that—it can self-improve.

And since this week DeepSeek has shown us this does not necessarily need 5 nuclear plants but can be done efficiently and run locally.

Nobody is talking about this …

Mathematics and coding are about to undergo a revolution. Within months, we'll likely see AI systems that can handle complex mathematical proofs and generate sophisticated code with minimal errors. The jobs of tomorrow won't look anything like today's. Software engineers aren't going away, but their role is evolving rapidly—think less about writing code and more about orchestrating AI systems to solve complex organizational problems.

But here's what's keeping me up at night: The pace of change is unprecedented in human history. We're discussing the future of humanity through newsletters like this one, while governments and the general public barely grasp what's happening. The disconnect between the speed of advancement and our societal preparedness is alarming.

Anthropic’s Claudio Amodei has stated that within 3 years all human work will be obsolete. We cannot grasp that concept!

The physical world won't be far behind. While robotics faces different challenges due to the constraints of the physical world, the combination of advanced AI models with improving hardware will eventually bridge that gap. We're looking at a future where autonomous systems could handle everything from manufacturing to space exploration.

The wildest part? This isn't science fiction anymore. We're talking about real possibilities within years, not decades. The barriers aren't technical anymore- they're human.

How will we regulate this? How will society adapt? How will we ensure this technology benefits humanity rather than destabilize it?

I might have been covering AI developments for 2,5 years, but this week feels different. The acceleration is palpable. We're not just seeing incremental improvements - we're witnessing the early stages of a transformation that will reshape every aspect of human civilization.

The TLDR basically is that

1. We all thought a lot of processing power is needed to get to ASI

2. Chinese company disproves the theory

3. Big AI Studios are panicking

4. Lots of processing power + this new model = probably ASI

Until next week,

That being said. In a Linkedin post this week I told the story of an entrepreneur who just ordered 15 social media posts for the exorbitant sum of 10.000€. In that post I promised to show you how I would create and schedule 50 posts within the hour.

Even though I HATE filming myself I did the thing : I made the video 

Welcome to the Blacklynx Brief!

Like newsletters?

Here are some newsletters our readers also enjoy:

AI News

• As mentioned above Chinese AI lab DeepSeek launched DeepSeek-R1, an open-source reasoning model with 671B parameters, matching or exceeding OpenAI’s o1 on benchmarks like AIME and MATH-500. Available under an MIT license, R1 includes smaller, laptop-compatible versions and offers dramatically lower API costs ($0.14 vs. $7.5 per million input tokens).

• A World Bank-backed study in Nigeria found that students using AI as an after-school tutor achieved learning gains equivalent to two years of traditional education in just six weeks. The program, focusing on English skills and digital literacy, was especially impactful for girls and helped improve performance in regular school exams.

• Apple paused its AI-powered news summary feature after generating false headlines, including fabricated stories about arrests and deaths. News organizations like the BBC criticized the feature for contradicting reporting and eroding trust, prompting Apple to promise better labeling and user controls. The incident highlights ongoing challenges in ensuring accuracy for AI in real-time, high-stakes applications like news reporting.

• Microsoft Research unveiled MatterGen, an AI model capable of generating new materials with specific properties, using a diffusion architecture to predict atom types and structures. The system outperforms previous methods in generating stable, optimized materials and integrates with MatterSim for validation.

• OpenAI announced the upcoming release of o3-mini, a faster, scaled-down reasoning model set to launch in both API and ChatGPT simultaneously. While o3-mini is less capable than o1 Pro, it prioritizes speed and lower costs, with o3 and o3-Pro models slated for future releases at the $200/month Pro tier. OpenAI’s rapid model rollout signals an accelerated innovation cycle for 2025.

• Sam Altman will present “PhD level” AI agents capable of complex problem-solving to U.S. officials on Jan. 30th, alongside OpenAI’s broader initiatives for economic growth. The briefing comes amid internal excitement over new capabilities, including GPT-4b micro, a model engineering proteins for cellular reprogramming with 50x higher efficiency than human scientists. OpenAI’s behind-the-scenes advancements suggest a fast-approaching leap in AI capabilities.

• Runway officially launched Frames, an AI image generation model featuring photorealism and stylistic control through customizable ‘Worlds’ for consistent aesthetics. Outputs integrate seamlessly with Runway’s video tools and include advanced features like aspect ratio control and fixed seeds. Frames expands Runway’s creative platform, offering powerful tools for artists and professionals to craft cohesive, high-quality visual content.

• Foxconn has signed a deal with UBTech to use Walker S1 humanoid robots for iPhone assembly tasks, including quality inspection and heavy lifting. These 5’6” robots, trained at Foxconn's Shenzhen factories, will focus on tasks that reduce strain on human workers. With humanoid robots now entering production lines at major manufacturers like Foxconn, the automation revolution in manufacturing is accelerating.

• The Isambard-AI supercomputer, a $276M UK investment, uses AI to simulate molecular interactions at the atomic level, revolutionizing drug and vaccine development. Already advancing treatments for Alzheimer’s, heart disease, and melanoma, the system can rapidly test millions of drug combinations. Expected to be operational by summer, Isambard-AI underscores the transformative potential of AI-driven medical research to tackle global health challenges.

• OpenAI announced The Stargate Project, a $500B venture with SoftBank, Oracle, Nvidia, and others to build AI infrastructure, starting with $100B for Texas-based data centers. The initiative, promising "hundreds of thousands" of jobs, coincides with President Trump’s repeal of Biden's AI safety executive order, removing oversight requirements for AI labs. This massive collaboration signals a U.S. focus on dominating the global AI race, prioritizing scale and speed over safety concerns.

• Tencent’s Hunyuan3D 2.0 generates high-quality 3D shapes and textures, combining Hunyuan3D-DiT for detailed geometry and Hunyuan3D-Paint for realistic textures. The platform includes tools like sketch-to-3D and stylization, outperforming competitors in detail and quality. By open-sourcing this system, Tencent empowers developers and creators in fields like gaming and film to streamline complex workflows and unlock new creative possibilities.

• Perplexity launched Sonar and Sonar Pro, APIs that provide fast, reliable, and citation-backed search capabilities for third-party apps. Priced at $1/M tokens for standard and $15/M tokens for Pro, the APIs outperform rivals like Gemini and GPT-4o on accuracy benchmarks. Sonar addresses a critical market need for factual, real-time AI search, making it an attractive option for developers seeking to integrate trustworthy search features.

• Google DeepMind launched Gemini 2.0 Flash Thinking, an experimental AI model excelling in math, science, and multimodal reasoning, with a record-breaking 73.3% on AIME and 74.2% on GPQA Diamond. Featuring a 1M-token context window and built-in code execution, the model processes vast data efficiently while providing explicit reasoning transparency. Currently free in beta, Gemini 2.0 offers unmatched capabilities compared to OpenAI’s costly premium models, setting a new standard in AI performance.

• Elon Musk publicly questioned the financial backing for OpenAI’s Stargate Project, claiming SoftBank has "well under $10B secured," despite its partnership with Oracle and Nvidia. Sam Altman dismissed the critique and invited Musk to visit the project site, while Microsoft CEO Satya Nadella confirmed an $80B commitment to Stargate in a CNBC interview. The exchange highlights rising tensions as Musk navigates his dual roles as AI competitor and political influencer.

• ByteDance released Doubao 1.5 Pro, a multimodal AI outperforming GPT-4o and DeepSeek-R1 on benchmarks while using fewer compute resources and offering lower pricing. The company also open-sourced veRL, the library powering Doubao’s reasoning, and UI-TARS, an agent model capable of interacting with GUIs using screenshots. Despite geopolitical tensions and chip restrictions, ByteDance continues pushing the boundaries of efficient, high-performance AI.

Quickfire News

• Mistral AI partnered with Agence France-Presse to integrate real-time multilingual news coverage into its Le Chat assistant, providing verified information in six languages.

• Krea AI introduced an image-to-3D object feature in its creative suite, enabling users to convert generated images into editable 3D objects for real-time placement and use.

• Princeton University researchers launched the Holistic Agent Leaderboard (HAL), a cost-aware benchmarking platform evaluating AI agent performance across 11 standardized tests.

• Chinese robotics firm Mirror Me debuted Black Panther 2.0, a robotic dog capable of sprinting 100 meters in under 10 seconds, featuring spring joints for animal-like movement.

• Google consolidated its Workspace AI features into standard plans with a $2 monthly price increase, eliminating the $20 Gemini add-on to compete with Microsoft’s integrated AI strategy.

• Minimax launched T2A-01-HD, a text-to-audio model supporting voice cloning from 10-second samples in 17+ languages with emotional synthesis capabilities.

• DeepSeek unveiled a cross-platform mobile app featuring its V3 model, offering free access to an AI assistant with web search and file processing for iOS and Android users.

• Perplexity acquired professional networking platform Read.cv and is reportedly preparing a $50 billion merger bid for TikTok's U.S. operations.

• Epoch AI is facing criticism after it was revealed that OpenAI financially backed the Frontier Math benchmark, which o3 recently used to achieve record-breaking performance.

• Pew Research Center data showed that U.S. teen usage of ChatGPT for schoolwork has doubled to 26% since 2023, with 79% of teens now familiar with the platform.

• Character AI launched two interactive games, signaling a strategic shift toward entertainment integrated with AI-powered chatbots.

• Cognition Labs released version 1.2 of its Devin AI coding assistant, featuring enhanced context understanding, a browser-based workspace, enterprise accounts, and Slack audio integration.

• OpenAI CEO Sam Altman clarified on X that the company has not built AGI and will not release it next month, dismissing "Twitter hype" as out of control.

• ByteDance launched Trae, a macOS-based AI coding tool offering automated project building and interactive development assistance in both Chinese and English.

• Liquid AI introduced LFM-7B, a language model leveraging LFM architecture for improved chat capabilities in languages such as Arabic and Japanese.

• Codeium released Windsurf Wave 2, an update to its AI development platform, adding web search integration, automated learning patterns, improved code execution, and enhanced enterprise features.

• Moonshot AI, a China-based lab, unveiled Kimi k1.5, a multimodal AI model achieving state-of-the-art performance on the short-CoT benchmark with joint reasoning over text and vision.

• Former OpenAI CTO Mira Murati secured her first hires for her new AI startup, bringing on Jonathan Lachman, OpenAI's ex-head of special projects, along with ten other key hires from top AI firms.

• Anthropic CEO Dario Amodei announced at Davos that the company is developing a two-way voice mode for Claude and smarter AI models, with plans for over 1 million chips powering its systems by 2026.

• Mistral CEO Arthur Mensch confirmed that the company is exploring an IPO, denying acquisition rumors, as the French AI leader nears a $6 billion valuation.

• The X Games will use a Google Cloud-powered AI judging system during the 2025 Aspen competition to analyze and grade snowboarding superpipe performances alongside human judges.

• The UK government introduced Parlex, an AI system within its Humphrey suite that predicts how members of parliament will react to proposed policies by analyzing parliamentary data.

• Goldman Sachs launched GS AI Assistant, an internal tool designed to help over 10,000 employees with tasks such as email drafting and code translation.

• OpenAI CPO Kevin Weil revealed at Davos that the company is already training the successor to its upcoming o3 reasoning model, predicting a "big jump in capabilities" despite shorter release cycles.

• Microsoft adjusted its exclusive cloud agreement with OpenAI, keeping first-refusal rights for computing capacity while allowing OpenAI to explore additional infrastructure partnerships.

• Samsung unveiled its Galaxy S25 series phones at the "Unpacked" event, featuring AI upgrades like enhanced Gemini integrations, multimodal agent capabilities, and context-aware natural language interactions.

• Google reportedly invested an additional $1 billion+ into Anthropic, raising its total stake to over $3 billion, alongside a separate funding deal that could increase Anthropic’s valuation to around $60 billion.

• Salesforce CEO Marc Benioff told CNBC that tensions between OpenAI and Microsoft deepened after Mustafa Suleyman became CEO of Microsoft AI, straining relations with Sam Altman.

• Ramp introduced Ramp Treasury, an AI-powered cash management platform offering automated balance alerts, liquidity forecasting, and self-directed finance options for passive investing.

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 !