Warum die nächsten Sprünge in Richtung AGI möglicherweise „geborenes Geheimnis“ sind // Ein neues Manhattan-Projekt zum Bau von AGI / ASI steht vor der Tür

Sounds like wasted money and will likely devolve to “we dont have the technology yet”, also OP is a bot

On multiple fronts, this proposal is beyond silly.

* „the inability of frontier labs to protect them from espionage disqualifies them from being the stewards of those models“ – [Handy link to list of spies on the atomic project](https://en.wikipedia.org/wiki/Atomic_spies). Also a reminder that Russia lagged the US in atomic bomb testing by 4 years and hydrogen bomb testing by 3 years. The original Manhattan Project succeeded in developing atomic weapons before Japan or Germany, but failed to give the US a durable lead. We’re not trying to develop AGI before a hard deadline, like landing US troops on Japanese homeland. And there is no reason to believe a crash government project would give the US a sustained lead.
* How would this be staffed? Tons of people working at US AI labs are not US citizens. Take a look at the author lists for any major AI project ([GPT-4](https://arxiv.org/abs/2303.08774), [Gemini 1.5,](https://arxiv.org/abs/2403.05530) [LLama 3](https://arxiv.org/abs/2407.21783)). Be sure to click the links at the end of the list that say things like „(1035 additional authors not shown)“ to get more data. Now, a name certainly does not convey citizenship, but it is a hint. Not clear that even a majority of people working on AI in the US are actually US citizens. So… what? Are we going to fire all all citizens of China, India, and elsewhere, crippling all the top US AI labs? Good plan to maximize velocity.
* The field is moving far too fast to assemble a competitive government effort before major milestones are reached, especially under a president consistently partial to old school technologies ([coal vs. wind](https://www.eenews.net/articles/what-trumps-love-affair-with-fossil-fuels-means-for-us-emissions/), [steam vs. electromagnetic launchers](https://news.usni.org/2019/05/28/experts-navy-would-spend-billions-to-answer-trumps-call-to-return-carriers-to-steam-catapults), [gasoline vs. electric](https://www.reuters.com/business/autos-transportation/trumps-transition-team-aims-kill-biden-ev-tax-credit-2024-11-14/)). The proposal (all half-dozen sentences of it) is apparently to give additional funding to US AI labs. So… the US government will start shoveling money into Google, Microsoft, Meta, etc.? Absolutely impossible politically.
* We’ll know the US is serious about prioritizing AI development relative to other societal concerns when Congress changes copyright law to explicitly allow training on any data set, rather than leaving the issue to wind slowly through the courts for the next 5 years while AI companies live in legal jeopardy. And how about expediting development of nuclear plants to power datacenters, instead of letting them languish in regulatory red tape for a decade? Do not hold your breath on either one. It is easy to say „Manhattan Project“ and copy-paste in a mushroom cloud. But when it comes to making the hard tradeoffs that really could accelerate AI (instead of just throwing money somewhere), I think there’s very little chance politically.