gribeill.bsky.social @gribeill.bsky.social · 6hWould be super curious to see code, given the impact of RL... 120
gribeill.bsky.social @gribeill.bsky.social · 14/10/2026To be real though I do think frustration with the defense startups has been broadly good for pushing the big 5 to spend more on IRAD, although hard to disentangle from other happenings. 040
gribeill.bsky.social @gribeill.bsky.social · 14/10/2026As someone from inside one of the primes, all I can say is: 😐😐 270
gribeill.bsky.social @gribeill.bsky.social · 14/10/2026Really curious to know if these people have ever looked up the margins the primes make on programs... 110
gribeill.bsky.social @gribeill.bsky.social · 14/10/2026www.darpa.mil/research/pro... the fact that it's a DARPA program is a clue about how hard this is. Also very hard to beat copper for density, reliability and power, especially once you count the laser.darpa.milHAPPI | DARPAThe Heterogeneous Adaptively Produced Photonic Interfaces (HAPPI) program aims to achieve a 1000x increase in connectivity density within microsystems by developing advanced three-dimensional photonic... 041
gribeill.bsky.social @gribeill.bsky.social · 12/10/2026Specifically for pure math, the "liberal arts" reasons are very much the primary motivation, and most practical application is very much incidental and downstream. Usefulness is not the (only) metric by which we judge mathematical advances. 020
gribeill.bsky.social @gribeill.bsky.social · 12/10/2026Not true for every theorem in general, but yes, broadly, pure math is absolutely something we do for itself. 140
gribeill.bsky.social @gribeill.bsky.social · 08/10/2026Bankrupting my campaign by mailing every voter a copy of L&L volume 9 030
gribeill.bsky.social @gribeill.bsky.social · 25/09/2026Almost none, not counting firearms. You're thinking of the EAR, which catches dual use items but is also much less restrictive. 030
gribeill.bsky.social @gribeill.bsky.social · 20/09/2026This isn't that hard to do, median household income in 1996 was 27,000 loaves of bread, it's 31,000 today. Not really sure what this is supposed to prove, but seems like an increase to me! 1361
gribeill.bsky.social @gribeill.bsky.social · 18/09/2026This was great! In the same vein, I think you'd be interested in the work of Enki Bilal... 010
gribeill.bsky.social @gribeill.bsky.social · 09/09/2026Fair enough, apologies for reading too much into your post, although "noteworthy" is doing a lot of work in the context of claims around RT superconductivity. 010
gribeill.bsky.social @gribeill.bsky.social · 09/09/2026Okay, but your original claim was much closer to something like "Astra++ will find a polynomial time algorithm to a NP-complete problem" than "incrementally advance SOTA in superconductivity theory" 130
gribeill.bsky.social @gribeill.bsky.social · 09/09/2026No argument from me, although I can already give you the answer: make a QC with ~five orders of magnitude more qubits. 140
gribeill.bsky.social @gribeill.bsky.social · 09/09/2026We don't know what to simulate. At this point it's just "invent the theory, invent the simulator, probably invent the quantum computer to run it on". Which sure, maybe! But this is just not a well posed problem at all. 150
gribeill.bsky.social @gribeill.bsky.social · 09/09/2026There's no such thing as a "RT superconductor theory object"; there's not even a theory to predict the transition temperature of a given material from structure. 2110
gribeill.bsky.social @gribeill.bsky.social · 09/09/2026One interesting question is for problems that don't admit a good formalization to Lean, although I don't know enough to know if that's a very big set. 010
gribeill.bsky.social @gribeill.bsky.social · 09/09/2026The new problems are generally discovered through the (social) process of attempting to solve the old problems, so I think in the absence of other changes we would expect fewer hard problems. 102
gribeill.bsky.social @gribeill.bsky.social · 26/08/2026I think the serious claim is more "there is surprisingly little evidence for organized warfare in paleolithic culture". This seems like a good overview: doi.org/10.1177/0022... 100
gribeill.bsky.social @gribeill.bsky.social · 26/08/2026The linked AAAS article makes it sound more like there wasn't actual material evidence for warfare that time/place compared to other activities, hence the surprise. 020
gribeill.bsky.social @gribeill.bsky.social · 24/08/2026Especially coming from people who keep talking about tradeoffs but magically forget their preferrered policy could actually have negative outcomes too 010
gribeill.bsky.social @gribeill.bsky.social · 20/08/2026To a large extent this is cultural and could be solved by better software practices, but AI doesn't incentivize this over the easy path. So yes, we need better ways of thinking about this and we'll probably get there but the transition will be extremely painful 030
gribeill.bsky.social @gribeill.bsky.social · 20/08/2026To give an example: our lab has a repo of shared code for various purposes built up over years. Good use of AI would be to increase test coverage, refactor old code, add new capabilities incrementally. What people actually do is write one-off scripts with no tests that are hard to audit/reproduce 140
gribeill.bsky.social @gribeill.bsky.social · 20/08/2026Depends on your view of code in the scientific process: in the best case it's an instrument that's a shared record of best practices. Its already very hard to get people to contribute back, and I see it as making the problem worse. 140
gribeill.bsky.social @gribeill.bsky.social · 20/08/2026And at least for now, for me, at the research frontier the models really go off a cliff and will absolutely start spouting nonsense very confidently so you need to approach them with extreme caution 040
gribeill.bsky.social @gribeill.bsky.social · 20/08/2026Also in my experience it accelerates the problems of Scientist Code exponentially: people go from shared and tested lab infrastructure code to one off janky scripts 170
gribeill.bsky.social @gribeill.bsky.social · 20/08/2026I'm not going to deny it's utility, but I have very mixed feelings about a tool that automates 20% of my tasks while making it much harder for junior staff to develop taste and judgement about scientific problems. 380
gribeill.bsky.social @gribeill.bsky.social · 19/08/2026Insane leap too where "college" means "Ivy League" and not "affordable state school or community college" which is 95% of how people experience it! 020
gribeill.bsky.social @gribeill.bsky.social · 11/08/2026Kim Stanley Robinson pretty much gets there in Blue Mars 150
gribeill.bsky.social @gribeill.bsky.social · 10/08/2026Contra that random staffer's quote, I don't think this makes very much sense at all. They would very much profit off of someone figuring out how to make them faster/better! 150
gribeill.bsky.social @gribeill.bsky.social · 05/08/2026Unfortunately would require more space than we have here, but imo the problems with US FP are structural. Thanks for the exchange though! 020
gribeill.bsky.social @gribeill.bsky.social · 05/08/2026Hard to find good open source information, but I am going to bet the limitation is subKs and propulsion providers, not tech transfer as is. 030
gribeill.bsky.social @gribeill.bsky.social · 05/08/2026Inevitable is maybe a strong word, but I believe that the lack of constraints is one of the main drivers of our inability to formulate coherent strategic objectives over the last 35 years. As for (2) I personally think it's at least worth the attempt. It would at least be productive in failure. 100
gribeill.bsky.social @gribeill.bsky.social · 05/08/2026Didn't MBDA license Aster production to Ukr recently? Obviously far away but not obviously less so than PAC-2/3 260
gribeill.bsky.social @gribeill.bsky.social · 05/08/2026Sure, but the argument isn't the nature of the threat, it is the appropriate size of the response. Expeditionary power projection and FONOPs aren't in any case orthogonal, so hard to say, but it's notable that the current threat to freedom of navigation came about because we can project power... 020
gribeill.bsky.social @gribeill.bsky.social · 05/08/2026Is freedom of navigation actually ensured by carriers, though? They seem to me to be sized as strike platforms, not necessarily for anti piracy... 111
gribeill.bsky.social @gribeill.bsky.social · 05/08/2026LLM : coders :: Excel : accountants is my view of the future 020
gribeill.bsky.social @gribeill.bsky.social · 05/08/2026Problem is going to be taste, which is what distinguishes good management from bad imo. And I'm skeptical you can develop taste without putting in the reps (at least not yet) 250
gribeill.bsky.social @gribeill.bsky.social · 05/08/2026Or said another way, until national security is brought back under true democratic control, strategy (viewed as the total power of state and society applied to ends) will always be incoherent, forever confusing ends, ways and means. 020
gribeill.bsky.social @gribeill.bsky.social · 05/08/2026How about instead: this was the inevitable consequence of hegemonic power, and the only way out is through the discipline imposed by constraint. Either Cold War 2.0 or an actual reconstitution of international institutions with the US binding itself to them. 2120
gribeill.bsky.social @gribeill.bsky.social · 27/07/2026This typically refers to 193nm from an excimer laser. Nikon and Canon are the other common suppliers. Introduced in 2003-2004 so pretty much inevitable a Chinese company would get there. 020
gribeill.bsky.social @gribeill.bsky.social · 27/07/2026Unfortunately that would involve agreement on what the purpose of state power is (and probably democratic legibility of the national security state). Not sure how we get there! 031
gribeill.bsky.social @gribeill.bsky.social · 23/07/2026Kind of interesting how you don't see much in Anglo media about the French Sahel missions. But this is just wanting to bounce rocks to look like you're doing something. 010
gribeill.bsky.social @gribeill.bsky.social · 27/06/2026Yeah. I'm working with an HVAC contractor in France right now and they are absolutely overwhelmed. Like in the US there's a huge gap in the trades. There are real structural issues and tradeoffs that would be much more interesting to talk about... 1160
gribeill.bsky.social @gribeill.bsky.social · 27/06/2026Sure, just trying to explain why the online discourse is so rotten on this. An in my experience in real life, no one I know is really anti-AC in France. 1160
gribeill.bsky.social @gribeill.bsky.social · 27/06/2026I mean just taking the example of France one issue is that the far right keeps proposing to zero out all green energy/climate programs in order to fund AC purchases (while having previously voted to slash funding for heat pump and insulation subsides). 1170
gribeill.bsky.social @gribeill.bsky.social · 27/06/2026Frankly it's mostly just Online derangement... at least from my experience in France people in real life are much less weird about it. 280
gribeill.bsky.social @gribeill.bsky.social · 27/06/2026Pretty interesting if a bit abstract for my primitive experimentalist brain. Look forward to the next post. 010
gribeill.bsky.social @gribeill.bsky.social · 27/06/2026Yes, I didn't mean just mech int, although that's the context I'm most familiar with Ising analogies in aside from people doing oscillator computing. Will read, thanks. 110
