Jouko Salonen

the past week i've been publishing head to head tests between nvidia vs alibaba AI models on consumer hardware like RTX 3090. same GPU, same tests, same inference engine. letting the architectures fight. nvidia failed twice on my 3090. cascade 2 at 3B active hit 187 tok/s but couldn't build a working game in five attempts. openreasoning 32B dense couldn't stop solving math problems long enough to write a single line of code. two architectures, two failures on same card. alibaba's qwen has been winning on every card i've tested. 27B dense on a single 3090 is still undisputed for agent coding. 9B also built a full game on a 12GB card from 2020. so i'm done with consumer hardware for this fight. loading both flagships on 2x H200 NVL. 287GB of VRAM. full BF16 precision. no quants. nvidia nemotron super 120B-A12B. 120 billion parameters, 12 billion active per token. latent mixture of experts with mamba-2. their GTC flagship. against alibaba's qwen 3.5 122B-A10B. 122 billion parameters, 10 billion active per token. the lineup that has been winning at every tier. both unquantized. both on the same hardware. same octopus invader game test, same agent, same prompt. the only variable is the model. if nvidia can't code at full precision on datacenter hardware it can't code anywhere. downloading now.

BREAKING 🔴 Reports that a nuclear scientist was killed in an airstrike in Dorud, Lorestan Province, Iran.

Humans: 100% Gemini 3.1 Pro: 0.37% GPT 5.4: 0.26% Opus 4.6: 0.25% Grok-4.20: 0.00% François Chollet just released ARC-AGI-3 -- the hardest AI test ever created. 135 novel game environments. No instructions. No rules. No goals given. Figure it out or fail. Untrained humans solved every single one. Every frontier AI model scored below 1%. Each environment was handcrafted by game designers. The AI gets dropped in and has to explore, discover what winning looks like, and adapt in real time. The scoring punishes brute force. If a human needs 10 actions and the AI needs 100, the AI doesn't get 10%. It gets 1%. You can't throw more compute at this. For context: ARC-AGI-1 is basically solved. Gemini scores 98% on it. ARC-AGI-2 went from 3% to 77% in under a year. Labs spent millions training on earlier versions. ARC-AGI-3 resets the entire scoreboard to near zero. The benchmark launched live at Y Combinator with a fireside between Chollet and Sam Altman. $2M in prizes on Kaggle. All winning solutions must be open-sourced. Scaling alone will not close this gap. We are nowhere near AGI. (Link in the comments)

The head of Europe's central bank just went on record and said financial markets don't understand what they're in for (Save this). This is Christine Lagarde saying the damage is already done, and most people have no idea. Here is what she actually said. The US and Israel struck Iran on and within days, Iran closed the Strait of Hormuz, the narrow chokepoint that carries roughly 20% of the world's oil and natural gas. Markets shrugged and investors assumed it would blow over fast but Lagarde says that assumption is dangerously wrong. She told The Economist that technical experts are not talking about months for a recovery, they are talking about years. The infrastructure, extraction, refineries, distribution networks has already taken damage that cannot simply be rebuilt on a short timeline. Then she dropped a detail most investors have completely missed, Helium travels through the Strait of Hormuz. Helium is the invisible ingredient inside every advanced microchip on earth. It cools the machines, purges the chambers, and delivers the precision modern semiconductors depend on. Qatar supplies roughly 35% of the world's commercial helium and Qatar's facilities have gone dark. Helium spot prices have already surged past $450 per thousand cubic feet and most chip fabricators carry less than three months of inventory. The world is building AI data centers at record speed, and the raw material that makes those chips possible is suddenly scarce. Meanwhile, Brent crude has already hit $99 a barrel, with earlier spikes past $120 and gasoline in the U.S. has jumped over 30% since the war began. Iraq has cut 1.5 million barrels per day while Saudi Arabia paused its largest refinery. Europe is now staring at a second energy crisis with gas storage at just 30% capacityheading into this. And the ECB, rather than cutting rates to soften the blow, is now considering hiking rates to fight the inflation surge. This means a slow economy, rising prices and tightening monetary policy, all at once. Lagarde's core warning is this, markets are not pricing in reality.

This applies to academic labs too — maybe more than anywhere. Defense in academia looks like rigor: protect your niche, optimize your pipeline, wait for the field to settle. It isn't rigor. It's just slower irrelevance. The labs that matter in five years aren't asking "how do we use AI to do our old experiments faster." They're asking "what questions are now askable that weren't before?" That's the offensive posture. And academics have something founders don't — the freedom to chase questions with no obvious payoff. That's an enormous edge, if you use it.

German journalist Hüseyin Doğru has had no access to his bank accounts and has not been allowed to travel since the EU sanctioned him in May. The sanctions came following his reporting on the repression of pro-Palestinian activists. This week, a Frankfurt court upheld the blocking of his accounts. Doğru and his family now face homelessness. Journalists critical of Western foreign policy are increasingly subject to sanctions and legal harassment designed to make their lives impossible.