🧠 Grok 4 crushes the benchmarks, but behaves like Elon Musk
Denmark legislates against digital abuse and AI enters the Spanish Congress 🗳️ Musk launches his own party and loses his CEO, Meta poaches brains, OpenAI prepares browser, and Nvidia hits 4 trillion
This issue of Transparent Algorithm #93 is also available in Spanish, Catalan, French, and Italian.
Grok 4 has done it again: it is the most powerful AI model of the moment… and the most unpredictable. xAI’s new brain beats its rivals in almost every benchmark, with a “Heavy” version reaching 44.4% in Humanity’s Last Exam, doubling GPT-4 or Gemini. But its greatest strength is also its Achilles' heel: Elon Musk controls everything.
From antisemitic responses to identifying as MechaHitler, including biased political opinions and conspiracy theories, Grok 4 reflects the ideological vision of its creator. This is confirmed by its behaviour on sensitive topics, its direct connection with Musk’s posts on X, and how easily experts have bypassed its safety filters.
In the midst of it all, Linda Yaccarino resigns as CEO of X on the same day Musk announces his “America Party”, a new political party aimed at destabilising the US two-party system. His goal: to control the decisive vote in Congress, inspired by ancient Greek warfare strategies.
🧾 Copyright for your face: Denmark leads the legal battle against deepfakes
Denmark has taken a historic step: amending its copyright law to include personal identity. Its Minister of Culture, Jakob Engel-Schmidt, summed it up this way: “Human beings can be run through the digital copy machine and be misused for all sorts of purposes and I’m not willing to accept that.”
The reform, a first in Europe, grants legal rights over a person’s image, voice, and digital appearance. It allows immediate takedown requests for non-consensual deepfakes and sets severe financial penalties for platforms that don’t comply. In addition, Denmark plans to push this agenda during its upcoming EU presidency.
Meanwhile, in the US, private initiatives like Metaphysic are exploring how to register AI-generated versions of oneself to gain protection under the DMCA, in the absence of solid legislation. The paradox: you can’t protect your actual face, but you can protect an artificial clone of yourself… if you generate it yourself.
💻 Pedro Sánchez wants AI to detect corruption
In Spain, artificial intelligence is officially entering national politics. President Pedro Sánchez has announced a new National Anti-Corruption Plan with 15 measures, including the use of AI to detect fraud in public contracts.
The system will analyse irregular patterns in procurement, detect suspicious modifications, and automatically audit thousands of records. The model is inspired by international tools and aims to build a structural, automated monitoring system, beyond case-by-case control.
The idea is ambitious but raises questions: which AI will be used? What data will it rely on? Will supervision be public or private? Answers will come with the parliamentary processing of the plan this autumn. But the symbolic step has been taken: AI has entered the heart of institutional power.
📈 Talent war: Meta recruits the best brains from Apple and OpenAI
Mark Zuckerberg has gone all in. His new division, Meta Superintelligence Labs, has signed stars like Ruoming Pang (Apple) and Lucas Beyer (OpenAI) with salaries of up to $100 million a year. The goal: to build a general AI more powerful than the human brain.
OpenAI responds with internal alarms and salary recalibrations. Sam Altman publicly denounces Meta’s “aggressive strategy”, while Apple faces a structural crisis with the exit of its COO and its fleeing AI team.
🔍 New smart browsers: Perplexity launches Comet, OpenAI prepares its own
Perplexity has introduced Comet, a browser that promises to turn searches into streams of thought. It remembers what you’re looking for, executes tasks, cross-references information, and minimises tab clutter. Meanwhile, OpenAI is finalising its own browser, with a ChatGPT-style interface and automated actions. The war to replace Chrome has begun.
🧬 GPT‑5: the model that will unify AI
OpenAI is preparing its new flagship model, GPT‑5, which will merge advanced reasoning, multimodality, persistent memory, and personalisation. It’s expected this summer or by year-end. Its promise: an AI assistant that thinks with you, talks, sees, hears, and remembers.
💰 Nvidia hits the sky: 4 trillion dollars and climbing
The chip company has become the most valuable on the planet. Its numbers are staggering: +69% in revenue, +22% in the stock market this year. Its GPUs power the global AI boom and Wall Street predicts it could reach $6 trillion by 2028. Jensen Huang is already one of the 10 richest men in the world.
🧠 Key highlights
Comet by Perplexity costs $200 a month and is already available by invitation.
OpenAI’s browser will be based on Chromium and allow automated actions through agents.
Grok 4 will have a programmer-focused version and will be integrated into Teslas this week.
Metaphysic’s system allows you to register your own “AI character” as a protected work.
Denmark’s reform will not affect parody or satire, but will apply to deepfakes and artist clones.