The ELECTE blog
Back to the front page All articles
Newsletter
Whether you like it or not, your data will fuel AI.
From the illusion of total control to strategic participation: why viral posts won't save you. Whatever your choice, choose with awareness, not with digital illusions.NewsletterThe company without owners
The Story of Aria-7: When artificial intelligence buys itself and revolutionizes global capitalism.NewsletterThe robot that believed in God: the prophecy about AI black boxes
As a story from 1941 had already imagined artificial intelligences developing their own "religions."NewsletterThe Illusion of Reasoning: The Debate that's Shaking the World of AI
Apple publishes two devastating papers-"GSM-Symbolic" (Oct. 2024) and "The Illusion of Thinking" (June 2025)-which demonstrate how LLMs fail on small variations of classical problems (Hanoi Tower, river crossing): "performance decreases when altered only numerical values." Zero success on complex Tower of Hanoi. But Alex Lawsen (Open Philanthropy) retorts with "The Illusion of the Illusion of Thinking" demonstrating failed methodology: failures were limits of token output not reasoning collapse, automatic scripts misclassified partial correct outputs, some puzzles were mathematically unsolvable. By repeating tests with recursive functions instead of listing moves, Claude/Gemini/GPT solve Tower of Hanoi 15 records. Gary Marcus embraces Apple thesis on "distribution shift," but pre-WWDC timing paper raises strategic questions. Business implications: how much to trust AI for critical tasks? Solution: neurosymbolic approaches-neural networks for pattern recognition+language, symbolic systems for formal logic. Example: AI accounting understands "how much travel expenses?" but SQL/calculations/tax audits = deterministic code.NewsletterThe Pay Per Crawl Revolution: How Publishers Can Monetize AI Traffic
The Google pact worked: free crawlers in exchange for referral traffic. AI is destroying it: devastating crawl-to-refer ratio—Anthropic 38,000:1, OpenAI 1,700:1—with 80% of AI crawling for model training and zero clicks to publishers. Cloudflare game-changer (July 2025): first internet infrastructure company to block AI crawlers by default on every new domain + Pay Per Crawl marketplace where publishers request direct compensation. TollBit, monetization pioneer, already generates $71M/year with traffic exploding from 2.75M→13M bot accesses/day, CPM $15 = $195K/day. Two-tier pricing: summarization vs syndication rate. Convenience thresholds: <50K visits/month better strategic free access; 100K $75-750/month; 1M+ $750-10K/month immediate implementation. DataDome detects AI traffic tripled in 6 months (2.6%→8.2%), Skyfire builds AI agent payment network ($8.5M). AI-to-AI commerce projection $46B over next 3 years. No longer "allow/block" but a third option: "charge".Newsletter🤖 Tech Talk: When AIs develop their own secret languages.
While 61% of people are already wary of AI that understands, in February 2025 Gibberlink gained 15 million views by showing something radically new: two AIs that stop speaking English and communicate through high-pitched sounds at 1875-4500 Hz, unintelligible to humans. This is not science fiction but FSK protocol that improves performance by 80 percent, subverting Article 13 of the EU AI Act and creating double-level opacities: inscrutable algorithms that coordinate in indecipherable languages. Science shows that we can learn machine protocols (such as Morse at 20-40 words/minute) but we run up against insuperable biological limits: 126 bits/s human vs. Mbps+ of machines. Three new professions are emerging-AI Protocol Analyst, AI Communication Auditor, AI-Human Interface Designer-while IBM, Google and Anthropic develop standards (ACP, A2A, MCP) to avoid the ultimate blackbox. Decisions made today on AI communication protocols will determine the trajectory of artificial intelligence for decades to come.NewsletterRegulating what is not created: does Europe risk technological irrelevance?
Europe attracts only one-tenth of global investment in artificial intelligence but claims to dictate global rules. This is the "Brussels Effect"-imposing regulations on a planetary scale through market power without driving innovation. The AI Act goes into effect on a staggered timetable until 2027, but multinational tech companies respond with creative evasion strategies: invoking trade secrets to avoid revealing training data, producing technically compliant but incomprehensible summaries, using self-assessment to downgrade systems from "high risk" to "minimal risk," forum shopping by choosing member states with less stringent controls. The extraterritorial copyright paradox: EU demands that OpenAI comply with European laws even for training outside Europe-principle never before seen in international law. The "dual model" emerges: limited European versions vs. advanced global versions of the same AI products. Real risk: Europe becomes "digital fortress" isolated from global innovation, with European citizens accessing inferior technologies. The Court of Justice in the credit scoring case has already rejected the "trade secrets" defense, but interpretive uncertainty remains huge-what exactly does "sufficiently detailed summary" mean? No one knows. Final unresolved question: is the EU creating an ethical third way between U.S. capitalism and Chinese state control, or simply exporting bureaucracy to an industry where it does not compete? For now: world leader in AI regulation, marginal in its development. Vaste program.NewsletterWhy prompt engineering alone is of little use
Effective implementation of artificial intelligence separates competitive organizations from those destined for marginality. But in 2025, winning strategies have changed dramatically from even a year ago. Here are five updated approaches to truly harnessing the capabilities of AI.Page 2 of 3