Brussels fines AliExpress โฌ550 million, and actually means it
Good morning ๐ The EU's been writing rules about platform risk for years. Today Brussels put a price on ignoring them: โฌ550 million, care of AliExpress.
In today's issue:
- ๐ญ Brussels fines AliExpress โฌ550 million, and actually means it
- ๐ง Open-source cheerleading, brain-shaped nets, and a phone-friendly dashboard
- ๐ฌ Memory tricks for MoE serving, analog computing's long shot, and cluster-aware matching
- ๐ ๏ธ Run your own DSA gap check before the Commission runs it for you
- ๐ A serial SEC filer, a quiet S-1, and Domino's earnings
Get tomorrow's issue in your inbox.
One concise AI brief, sent after the signal clears the noise.
๐ญ THE ONE THING
๐ Brussels fines AliExpress โฌ550 million, and actually means it
The European Commission fined AliExpress โฌ550 million today for failing to police its own marketplace: the DSA's language on "diligently assessing and mitigating risk" from illegal, unsafe, and counterfeit goods, turned into an actual number. Miss the corrective-action deadline and periodic penalty payments start stacking on top. That's a real enforcement mechanism, not a press release with a threat attached to it. For any marketplace holding "very large online platform" status in the EU, moderation spend just moved from nice-to-have to line item, and โฌ550 million is what it costs to have treated that line item as optional.
๐ง MODELS & RELEASES
- ๐ค Hugging Face Clem Delangue ran his usual open-source case this week: "just wait for open weights and inference optimization" to catch closed labs, and that open science leads the frontier a few years out, the way it did for the US in AI. Two tweets, one thesis, no product attached. link link
- ๐ Sakana AI New work called "Diffusing Blame" trains neural nets that obey Dale's principle (a neuron is excitatory or inhibitory, never both, same as real ones) and still holds its own on image classification and RL. Accepted at ALIFE 2026. Brain-constrained learning that actually competes, not just a novelty result. link
- ๐ Amazon QuickSight AWS shipped a mobile layout for Quick dashboards: desktop views now auto-render single-column and touch-friendly on a phone, no rebuild needed. Handy if you check revenue between meetings. Not a model release, but it's what shipped today. link
๐ฌ RESEARCH HIGHLIGHTS
- PagedWeight MoE serving has a memory fight nobody talks about enough: the expert weights and the growing KV cache both want the same GPU room. This paper pages weight precision dynamically, keeping the bits that matter and thinning the ones that don't, so you can stretch context length without adding GPUs. Builder read: promising if the quality claims survive outside the paper's own benchmarks. link
- Thermodynamic computing A group sketched a blueprint for analog chips that compute by settling into physical equilibrium, using the hardware's own noise as the substrate instead of fighting it. No chip exists yet, just the math and the architecture. Watch, don't act, but it's a genuinely different bet on where post-GPU compute comes from. link
- Cluster-aware optimal transport Most point-matching methods treat every point as interchangeable. This one matches cluster to cluster first, then refines inside each pair, which should generalize better on data with real group structure. Narrow use case (dataset alignment, domain adaptation) but a clean idea. link
๐ ๏ธ TRY THIS
Run your own DSA gap check before the Commission runs it for you
A โฌ550 million fine isn't a rounding error, and AliExpress isn't a random target: it's a designated Very Large Online Platform, which layers risk-assessment and recommender-transparency duties on top of the baseline rules. Run a marketplace, an app store, or anything with an EU user base and a ranking algorithm, and the same articles apply to you at a smaller scale. Fifteen minutes with an LLM gets you a gap list before your lawyer bills you for the same exercise.
1. Paste your terms of service, recommender description, and ad-targeting policy into Claude.
2. Check the text against DSA Articles 14, 25-27, 30, and 34-35 (dark patterns, ad transparency, recommender transparency, trader traceability, risk assessment).
3. Have it rank the gaps by exposure: which one a regulator flags first.
4. Turn the top three into a one-page remediation brief for legal this week.
Prompt: You are a DSA compliance auditor. Given the following terms of service, recommender system description, and advertising policy, list every obligation under EU Digital Services Act Articles 14, 25-27, 30, and 34-35 that the text does not currently satisfy. For each gap, name the article, the missing element, and the enforcement risk if unaddressed. Text: [paste]๐ QUICK LINKS
- Artificial Intelligence Technology Solutions (AITX) filed four SEC documents in four days this week: three 8-Ks and a fresh S-1. Serial dilution is the business model here as much as the security robots are. S-1, 8-K
- Spectral Capital (FCCN) registered shares with a new S-1. No offering terms on the cover page, just the paperwork stage. Filing
- Domino's (DPZ) filed an 8-K with an EX-99.1 exhibit, the slot the chain reserves for earnings. Timing lines up with Q2 season. Filing
Run the gap check in Try This this week. Cheaper than finding out in a fine.
Pradeep Perugu