Subtitle: "Reflections on the importance of technological independence for individual states, individual organizations and, why not, individual people".
These reflections stem from Anthropic's recent announcement that it is making the Fable 5 model unavailable. The news strikes me both in its substance and in the way this ban came about. It is no surprise that Anthropic's latest models are considered “dangerous”. Much has been said about this, and it is the reason we moved from Mythos to Fable. The existence of these concerns is probably justifiable, and being cautious is probably a reasonable stance. There are security aspects (cybersecurity in particular) that must be managed, and perhaps it is even understandable and acceptable that a company should protect itself by making sure that its “own” software cannot be used to improve products that might be competitors. These are topics one can discuss and debate. But what troubles me — without turning this into a political argument — is something else: it is the way in which, if the reconstruction of the facts is correct, the "ban" came about. Institutional motivations that override any technical reasoning. The bottom line, as I see it, is this: the United States decrees that citizens of other nations may not use those Anthropic products.
Those models were trained on "everyone's" data; the theory behind them is "everyone's" heritage, the fruit of the work of many people, of many countries.
It is certainly true that the US — and in particular the companies operating there — have enjoyed access to vast computing power, low-cost energy and abundant capital, and they have made the most of this advantage; in that sense I believe nothing illegitimate has been done. I hope that American companies, and those across the rest of the world, will keep making progress and achieve ever more astonishing results. And yet I reflect on the access restrictions that have been imposed, because security and prudence are matters that concern everyone and should never become a tool (or an alibi) for creating inequality of opportunity. I hope these events shake us up a little, here in Europe in particular. I hope we reflect, and that — while we will in any case try to close the gap in computing power and access to low-cost energy — we put more and more resources into designing and developing models, into designing and developing fine-tuning, into local inference with open-weight LLMs, and that we prefer (where possible) open-source solutions, and so on.
With a thought also for data in the cloud, we must think hard and dwell on the importance of technological independence — for individual states, for individual organizations, and for individual people...