Pablo Moreno 🔸 🇪🇺 🇺🇦

More practical idea: make it easy for frontier AI companies to build datacentres in the UK, in return ask for guaranteed access to frontier models.

@AlexTPet @lugaricano Separately, it is non-obvious to me that Coasean dynamics won't kill a strategy based on diffusing AI in the economy. Model providers could build mostly automated companies if they achieve sufficient capabilities.

@AlexTPet @lugaricano I am worried about betting on non-frontier models. It's good to have bargaining chips in open-weight models, but the returns to marginal capabilities seem very significant. It'd be good to see a more concrete formalisation of this question, but my intuition is that it won't work.

I write today in Silicon Continent with Jesús Saa-Requejo "Three Theses on AI Value Capture". We argue that the leading AI labs are betting hundreds of billions on the idea that holds the best model captures the value. We think that's the wrong bet. The model layer is squeezed between customers who can switch with a simple change in configuration and suppliers who are each monopolists. Our hypothesis is that the surplus flows past the labs, to chips above and implementation below. Hence the country that wins AI is not the one with the tokens on the frontier model. It's the one that is best able to implement the technology. siliconcontinent.com/p/three-theses…

@AlexTPet @lugaricano Anton Leicht believes import agreements (writing.antonleicht.me/p/import-imper…) and datacenters in non-US soil (writing.antonleicht.me/p/can-you-poac…) are what will provide geopolitical leverage and prevent being cut off (writing.antonleicht.me/p/cut-off). But he believes access to frontier AI is necessary.

@lugaricano makes sense! I do wonder what is driving the gap though, and whether it will continue to grow.

@lugaricano If I think more deeply about it, I think I am worried about coasean dynamics of AGI: it is not clear to me that automated firms are off the table here. (dwarkesh.com/p/ai-firm) If broadly automated firms become a thing, the model will retain the whole application-layer value.

@lugaricano I unfortunately think that deciding to go for non-frontier models is a losing proposition. Empirically, the practical usefulness of increased AI capabilities seems massive. AI is not about to get commoditised. I think Anton Leicht has better suggestions writing.antonleicht.me/t/geopolitics.

@rdomenechv @lugaricano If I think more deeply about it, I think I am worried about coasean dynamics of AGI: it is not clear to me that automated firms are off the table here. (dwarkesh.com/p/ai-firm) If broadly automated firms become a thing, the model will retain the whole application-layer value.

@rdomenechv @lugaricano I unfortunately think that deciding to go for non-frontier models is a losing proposition. Empirically, the practical usefulness of increased AI capabilities seems massive. AI is not about to get commoditised. I think Anton Leicht has better suggestions writing.antonleicht.me/t/geopolitics.

Three theses on AI value capture, by @lugaricano & Jesús Saa-Requejo Instead of trying to win the frontier model race, middle powers should focus on implementing the organizational work and form a coalition to fund a shared open model 1–2 steps behind the frontier as a strategic hedge against supply rationing or vendor lock-in. open.substack.com/pub/siliconcon…

Today @cusp_ai and @KemiraGroup announce a milestone in AI-driven materials discovery. We have used generative AI to design new materials targeting PFAS removal from drinking and process water at trace concentrations.

Preventing an AI cyber bloodbath is becoming difficult, but not impossible. I urged the European Commission to build digital resilience swiftly. All of our society will be tested this summer, so act NOW #Mythos

@NXT4EU @StanisP32956 Such a project can only work at the European level. The budget will be absolutely ridiculous compared to that of the US. When will national politicians learn? The minimum would be to make it work with a small coalition of European countries.

A bittersweet announcement! For family reasons, I will be leaving AISI soon to move back to the Bay Area. I will be starting a new nonprofit alignment research org (more to come). I will miss this place! Here are some reflections about my time at AISI. 🧵❤️

Almost no one gives significantly to charity, but that's seen as normal, which leaves EAs in a weird place where of three possible people: Person 1: No money donated Person 2: $500 to their local community Person 3: $5,000 to a far away malaria charity People get way way more mad at 3 than 1.

UK AI SECURITY INSTITUTE: - November 2025 estimate that AI cyber abilities double in ability every 8 months - In February 2026, re-estimated doubling to be every 4.7 months - Claude Mythos and GPT-5.5 are more capable than a 4.7 month doubling rate would suggest

AI strategies everywhere hinge on widely available American frontier AI. Post-Mythos, amid compute crunches, security concerns and distillation crackdowns, that paradigm is under threat. Today, I argue the era of widespread access to frontier AI is almost over.

We are getting ready to do some very large runs on MirrorCode to learn whether AI can solve coding tasks that would take months for an engineer to complete. The version of the experiment we would most like to do is very expensive: it would cost us $600,000. And even a more modest version would cost upwards of $100,000. So we can spend this with confidence, we’d like to line up some funding. Please reach out if you want to back the effort!

Read here: open.substack.com/pub/pabloamc/p…

I took part in the @dwarkeshpodcast essay competition.

@miguel_tambien @OperadorNuclear Mi argumento es que si el precio de las baterías sigue bajando, podríamos monetizar las placas solares + baterías en unos años, que era el problema que subrayaba @OperadorNuclear Gemini dice que el coste del panel solar + almacenamiento ~39+78~$117/MWh. lazard.com/media/5tlbhyla…

@Pablomorecasa @OperadorNuclear Mi fuente era la IEA norteamericana. eia.gov/electricity/mo… En Texas dan un uso más sensato a las baterías, para estabilizar la red, no tanto como almacenamiento.

La energía solar fotovoltaica está chocando con una realidad física y económica que muchos llevamos años explicando. Cuanta más potencia solar se instala en las mismas horas del día, más se hunden los precios justo cuando produce. Y cuando caen a cero o incluso negativos, muchos parques dejan de ser rentables. La paradoja es que hace poco asistíamos a una campaña política y mediática celebrando que España tenía durante algunas horas los precios eléctricos más bajos de Europa. Lo que apenas se decía es que eran precios mayoristas, no la factura real que pagan hogares e industrias. «Créeme a mí y no a tu factura», parecían decir. Tampoco se hablaba del coste técnico de mantener estable un sistema con cantidades masivas de generación variable. Tras el apagón, Red Eléctrica opera en modo reforzado para reducir riesgos y evitar otro colapso. Eso implica más respaldo, más servicios de ajuste y más recursos síncronos estabilizando tensión y frecuencia. Todo eso cuesta, aunque la subasta marque cero euros unas horas. A la factura final hay que añadir peajes, cargos, impuestos, pagos por capacidad y costes crecientes de operación del sistema. La electricidad no vale solo lo que aparece en un gráfico del mercado mayorista al mediodía. Y mientras se usaban políticamente esos precios bajos como propaganda energética, parte del propio sector renovable empezaba a admitir que hay plantas fotovoltaicas ingresando cantidades ridículas y proyectos que podrían dejar de construirse o necesitar rescates financieros. Esto no significa que la energía solar no sea útil. Lo es y muchísimo. El problema es haber vendido durante años la idea simplista e infantil de que bastaba con llenar el país de paneles solares para resolver el problema energético. Una transición seria necesita combinar renovables con fuentes capaces de aportar estabilidad, inercia y potencia firme las 24 horas. Y ahí el parque nuclear español sigue siendo estratégico. Produce alrededor del 20% de la electricidad de forma continua, estable y sin emisiones de CO₂. Prescindir de esa capacidad en un sistema cada vez más tensionado es un suicidio energético. elperiodicodelaenergia.com/una-verdad-inc…

@miguel_tambien @OperadorNuclear Texas also leads in battery storage (finance.yahoo.com/sectors/energy…), with much cheaper energy according to your sources (15.41 cents/KWh).

@Pablomorecasa @OperadorNuclear Mucha batería en California, por ejemplo. No se consigue un recibo eléctrico tan absurdamente alto como el de California (que dobla el promedio de EE.UU.) sin hacer muchas tonterías. eia.gov/electricity/mo…