Germain

We will use Grok 3.5 (maybe we should call it 4), which has advanced reasoning, to rewrite the entire corpus of human knowledge, adding missing information and deleting errors. Then retrain on that. Far too much garbage in any foundation model trained on uncorrected data.

“If your $500K engineer isn’t burning at least $250K in tokens, something is wrong.”

This gets to the core of the issue more than any debate about specific terms. Do you believe in democracy? Should our military be regulated by our elected leaders, or corporate executives? Seemingly innocuous terms from the latter like "You cannot target innocent civilians" are actually moral minefields that lever differences of cultural tradition into massive control. Who is a civilian and not? What makes them innocent or not? What does it mean for them to be a "target" vs collateral damage? Existing policy and law has very clear answers for these questions, but unelected corporations managing profits and PR will often have a very different answer. Imagine if a missile company tried to enforce the above policy, that their product cannot be used to target innocent civilians, that they can shut off access if elected leaders decide to break those terms. Sounds, good, right? Not really - in addition to the value judgement problems I list above, you also have to account for questions like: -What level of information, classified and otherwise, does the corporation receive that would allow them to make these determinations? How much leverage would they have to demand more? -What if an elected President merely threatens a dictator with using our weapons in a certain way, ala Madman Theory/MAD? Is the threat seen as empty because the dictator knows the corporate executives will cut off the military? Is the threat enough to trigger the cutoff? How might either of those determinations vary if the current corporate executive happens to like the dictator or dislike the President? -At what level of confidence does the cutoff trigger, both in writing and in reality? The fact that this is a debate over AI does not change the underlying calculus. The same problems apply to definitions and use of ethically fraught but important capabilities like surveillance systems or autonomous weapons. It is easy to say "But they will have cutouts to operate with autonomous systems for defensive use!", but you immediately get into the same issues and more - what is autonomous? What is defensive? What about defending an asset during an offensive action, or parking a carrier group off the coast of a nation that considers us to be offensive? At the end of the day, you have to believe that the American experiment is still ongoing, that people have the right to elect and unelect the authorities making these decisions, that our imperfect constitutional republic is still good enough to run a country without outsourcing the real levers of power to billionaires and corpos and their shadow advisors. I still believe. And that is why "bro just agree the AI won't be involved in autonomous weapons or mass surveillance why can't you agree it is so simple please bro" is an untenable position that the United States cannot possibly accept.

what Anthropic just announced with Claude code & COBOL modernization is quietly one of the biggest AGI signals we’ve had COBOL still runs the vast majority of global banking & insurance infrastructure, written decades ago by people who are retired or dead in a language almost nobody alive understands anymore & modernizing it isn’t translating old syntax into new syntax these codebases are full of undocumented business logic, edge cases that existed only in the head of some developer in 1974, dependencies nobody ever mapped because the guy who built it just knew… so what Claude Code has to do is read millions of lines of ancient code & figure out what the original developer meant not what he wrote, reconstruct intent from context with incomplete information across a system duct taped for 50 years if you can do that you can reason IBM built an empire on being the only ones who could touch these systems, if an AI does in weeks what their teams bill over years that part of their business model is dead & the evaluation metric here is perfectly binary, either the modernized system runs & not a single transaction breaks or it doesn’t, no subjective benchmarks just does the global financial system keep working yes or no but here’s what nobody is connecting yet ??? if an AI can reverse-engineer the intent of dead programmers across decades of undocumented code in a dead language then what exactly can’t it reverse engineer, proprietary algorithms from bytecode, legacy military systems, abandoned scientific simulations, medical devices running firmware nobody understands anymore…COBOL is just the proof of concept the real product is an AI that reads human intent through the artifacts we leave behind & that changes everything

In France, we believe in science. That is why, on May 5, I issued a clear and open call to the world: for science, choose France. I am very proud to see that this call has resonated so strongly. Around forty leading researchers have chosen France. Through “France 2030”, we have invested more than €30 million to advance health, climate action, artificial intelligence, and fundamental sciences. Science has found its home.