AI ·
Forgetting in machines: when deleting data isn't enough
A datum erased from a table disappears. One erased from a model does not. On machine unlearning, weights that hold on and human forgetting as mirror.
Read post → Personal blog
Notes from the edge of AI
Notes and essays on the silent conversation between machines and those of us who design them. Short pieces to slow down the reading of a technology that moves too fast.
Explore by pillar
Get each note in your inbox
Double email confirmation. No noise. Only when I publish something worth reading slowly.
Latest posts
-
-
Technology ·Infrastructures that forget: from compliance to design
Saving is easy; erasing is hard. How to shift the architecture of storage so that forgetting becomes the default state and not a late add-on.
Read post → -
AI ·The engineering of doubt
Validation bias in AI outputs: code that compiles, well-written reports, closed-loop audits. How to build mechanisms of doubt into the pipeline.
Read post → -
AI ·When agents talk to each other
MCP and A2A protocols, task delegation and diffused responsibility. The machine-to-machine conversation engineering does not yet know how to supervise.
Read post → -
AI ·The hidden cost of reasoning: thinking has a bill
Reasoning models consume 6 to 13 times more energy than standard queries. What data engineering is still not measuring.
Read post → -
AI ·Muse Spark vs. Gemini 3 Deep Think: two ways to think
Two architectures, two philosophies of reasoning. Google bets on depth, Meta on presence. What changes for data engineering.
Read post → -
AI ·When AI learns to think on its own
Reinforcement learning, DeepSeek-R1 and the erosion of collective cognitive immunity. What emerges when we stop teaching and start observing.
Read post → -
AI ·When LLMs deliberate with each other
Bias, identity and exclusion in multi-agent systems. What happens when several language models stop talking to you and start talking among themselves.
Read post → -
AI ·Sora 2 and the vanishing of reality
When video stops being proof: how Sora 2 proposes emotional experiences indistinguishable from the lived, and what that blurred frontier implies.
Read post → -
AI ·The silent dependency
When Barcelona's companies outsource their thinking to opaque algorithms. A tech-ethical analysis of the Barcelona business ecosystem.
Read post → -
AI ·Lumo and sovereign AI
Proton's AI assistant proposes another way to build AI: no mass collection, no centralised infrastructures, and European sovereignty.
Read post → -
AI ·AI embedded in databases
Databases that understand context, generate metadata and participate in governance. From traditional modelling to co-modelling with embedded intelligence.
Read post → -
AI ·The future of AI: from tool to environment
AI will not arrive as a spectacular invention but as silent infrastructure. Five structural scenarios and the ethical tensions of tomorrow.
Read post →
