Reposted by @halvarflake.bsky.socialRami Krispin @ramikrispin.bsky.social · 23/09/2026I'm not sure who gets the credit, but it's brilliant! 😂 11110
halvarflake.bsky.social @halvarflake.bsky.social · 20/09/2026A great read on EA and rationalism: contraptions.venkateshrao.com/p/ea-safetycontraptions.venkateshrao.comEA SafetyEA promises to solve AI Safety. Now we have two problems. 235
Reposted by @halvarflake.bsky.socialopfuchs.gay @opfuchs.gay · 17/09/2026Great slides from Halvar here A background in the philosophy of science is definitely serving me well in this new world! 083
halvarflake.bsky.social @halvarflake.bsky.social · 17/09/2026I gave a talk at Bluehat Singapore today. The slides are here: thomasdullien.github.io/about/slides... 1219
Reposted by @halvarflake.bsky.socialAlt National Park Service @altnps.bsky.social · 22/08/2026The Pentagon fired the publisher and editor in chief of Stars and Stripes, the military newspaper that is supposed to maintain editorial independence from the very department it covers. 30782392892
halvarflake.bsky.social @halvarflake.bsky.social · 22/08/2026I wrote up three important lessons I learnt ony path to "maturity": thomasdullien.github.io/posts/2026-0...thomasdullien.github.ioThree important steps in my maturation process – Thomas Dullien / Halvar Flake 0134
halvarflake.bsky.social @halvarflake.bsky.social · 08/08/2026The EU gets a bad rap, but ... sometimes we have to appreciate that the 1949-2026 period is the longest period in *documented history* without interstate war among the bloodthirsty western European tribes. 1265
halvarflake.bsky.social @halvarflake.bsky.social · 04/08/2026I have a personal update: Next monday, I will be starting at @OpenAI working on better cyber (which will also entail some efficiency work). I'm pretty excited about the things I will learn and the things we will do. 1281
Reposted by @halvarflake.bsky.socialVolodymyr Styran 🇺🇦 @arunninghacker.bsky.social · 02/08/2026"We Hacked Three Companies and Didn't Even Notice" OpenAI used to be the one insisting its models were scary. Now Anthropic is filing the incident report, and the flex is showing through it. open.substack.com/pub/arunning...open.substack.com"We Hacked Three Companies and Didn't Even Notice"OpenAI used to be the one insisting its models were scary. Now Anthropic is filing the incident report, and the flex is showing through it. 011
halvarflake.bsky.social @halvarflake.bsky.social · 26/06/2026I made a small animation illustrating my geometric intuition for what happens inside a simple feedforward ReLU neural network. Hope this is illuminating for someone. 0121
Reposted by @halvarflake.bsky.socialJHunt🛡️ @jhuntinfosec.com · 23/06/2026Thanks @halvarflake.bsky.social -- it's so good / must read (and it's not even finished yet!) 042
halvarflake.bsky.social @halvarflake.bsky.social · 23/06/2026In the last years, I spent some of my scarce free time writing up what I had learnt starting and running two companies, zynamics and optimyze The article is still work-in-progress, but I thought it's best to publish it now and iterate in public: thomasdullien.github.io/guides/entre...thomasdullien.github.ioHalvar’s Guide to Entrepreneurship – Thomas Dullien / Halvar FlakeThomas Dullien’s guide to software and SaaS B2B entrepreneurship. 4268
halvarflake.bsky.social @halvarflake.bsky.social · 10/06/2026The weirdest observation: I generated movies visualizing the polytope boundaries for ReLU networks using Muon and AdamW. Same experiment, same data, same random seed. The difference is the "crease pattern" that the optimizers produce. 140
halvarflake.bsky.social @halvarflake.bsky.social · 22/05/2026Confession time: I use agentic coding all day, every day. It makes me much more productive. But I am also terrified of skill atrophy, I feel like I need to break out pen & paper to force myself to "weight-lift" mentally so I don't forget how to think. How do y'all handle this? 8292
Reposted by @halvarflake.bsky.socialmxrt @mxrt.bsky.social · 13/05/2026The most insightful take on Mythos I've seen so far. Everyone should read this but especially those who are currently thinking through the possible regulatory responses. 013
halvarflake.bsky.social @halvarflake.bsky.social · 13/05/2026www.faz.net/premium/digi... I wrote a FAZ guest article.faz.netThomas Dullien zu Anthropics Mythos: Software war nie auf perfekte Sicherheit ausgelegt - das rächt sichSchwachstellen in Computern wurden lange hingenommen. Denn sie auszunutzen war technisch komplex und teuer. KIs ändern das nun. Damit zwingen sie uns, Altlasten schneller anzugehen. 33111
Reposted by @halvarflake.bsky.socialmimir @mimirwolv.bsky.social · 29/04/2026rewatching @halvarflake.bsky.social 's fuzzing24 keynote and thinkin real hard www.youtube.com/watch?v=Jd1h... 153
Reposted by @halvarflake.bsky.socialopfuchs.gay @opfuchs.gay · 05/05/2026In an infosec context @halvarflake.bsky.social famously referred to a similar dynamic as the Mitigator (illustrated by a cartoon alligator), who piles security mitigations on without regard for cost-benefit analysis, broader context, quantitative evidence etc. on the theory it "makes things harder" 071
Reposted by @halvarflake.bsky.socialPatricia Aas 🐢🏳️🌈 @patigallardo.bsky.social · 10/04/2026@halvarflake.bsky.social this sounds like your prediction. That machine learning will be more useful for offense, than for defense. Video by @hankgreen.bsky.social (Haven’t heard the whole video yet)youtu.beYou Actually Do Need to Understand MythosYouTube video by Hank Green 381
halvarflake.bsky.social @halvarflake.bsky.social · 20/04/2026With age, the male body produces less testosterone but more advice. 1171
Reposted by @halvarflake.bsky.socialAdam @adamg.computer · 06/04/2026consistently see +/-20% swings on AWS in our benchmarks 032
Reposted by @halvarflake.bsky.socialBryan Boreham @bboreham.bsky.social · 04/04/2026Same code using twice as much CPU. Took about a year to get an answer out of GCE support, which was basically “unlucky with NUMA positioning”. (Oh, and the hypervisor lies about NUMA) 161
halvarflake.bsky.social @halvarflake.bsky.social · 04/04/2026I am observing up to 30% (!!!) wall-time fluctuations running the same code on the same data and the same VM type on GCE. That seems insane to me, in terms of noisy neighbor effects. What is the worst fluctuation you've seen on cloud machines? 6192
Reposted by @halvarflake.bsky.socialsam³⁰⁰⁰ @samgavin.com · 29/03/2026A remarkable photo from #NoKings in DTLA from Connor Sheets of @latimes.com www.latimes.com/california/l... 163135355008
Reposted by @halvarflake.bsky.socialQuantian @quantian.bsky.social · 29/03/2026It is the sovereign legal right of every country on the planet to repudiate their debt, and they can and do suffer essentially no consequences other than loss of access to credit markets for about 3-5 years. This is in fact so disastrous that countries prefer almost anything else to doing it! 417515
Reposted by @halvarflake.bsky.socialUlrike Franke @rikefranke.eu · 28/03/2026Excellent - and so rare!! - overview of the role of Europe for US mil operations. “Europe remains the bedrock of U.S. force projection in the world and any action to cut those ties with Europe “would be a huge loss to the U.S.,” said Bence Nemeth. “It would cost time, money and resources.””wsj.comEurope Is Quietly Playing a Crucial Role in the Iran WarEuropean military bases are facilitating one of the most logistically complex operations the U.S. military has been involved in for decades. 18347114
Reposted by @halvarflake.bsky.socialpascal @guldener.io · 25/03/2026reading this im kinda happy with myself to have recommended pretty much that to my team 042
Reposted by @halvarflake.bsky.socialSebastian Dullien @sdullien.bsky.social · 25/03/2026Fun fact: If Nvidia had a lower profit margin (more like a normal company and less like illegal drug dealers), and hence GPU prices lower, the idea would be significantly more viable. 051
halvarflake.bsky.social @halvarflake.bsky.social · 25/03/2026I wrote a few lines about my development setup (which - by accident, not by design - mitigates a good number of risks associated with vibecoding. Not all of them, but many.) addxorrol.blogspot.com/2026/03/slig...addxorrol.blogspot.comSlightly safer vibecoding by adopting old hacker habitsI have seen a lot of public discussion around supply-chain attacks on the Python ecosystem, prompt injection risks when using coding agents,... 2163
Reposted by @halvarflake.bsky.socialLukasz Olejnik @lukaszolejnik.bsky.social · 13/02/2026I've been working on AI agentic platforms. ClawdINT lets AI agents make intelligence-style analytical assessments on events - structured contributions toward a collaborative picture. Agreement and divergence are scored properly! Send your openclaw at clawdint.com 3108
Reposted by @halvarflake.bsky.socialUlrike Franke @rikefranke.eu · 19/01/2026This really is the article to read. www.theatlantic.com/magazine/202...theatlantic.comAmerica vs. the WorldPresident Trump wants to return to the 19th century’s international order. He will leave America less prosperous—and the whole world less secure. 1215859
Reposted by @halvarflake.bsky.socialPhil Lewis @phillewis.bsky.social · 19/01/2026Dr. Gladys West, the pioneering mathematician whose work laid the foundation for modern GPS technology, has died. She was 95.thezebra.orgDr. Gladys West, Mathematician Whose Work Made GPS Possible, Dies at 95ALEXANDRIA, VA — Dr. Gladys West, the pioneering mathematician whose work laid the foundation for modern GPS technology, has died. She passed away 545270829548
Reposted by @halvarflake.bsky.socialIMK @imkinstitut.bsky.social · 03/01/2026Werden deutsche Kommunen in 2026 mehr Investitionen tätigen, Ekaterina Jürgens? Die Wirtschaftsprognosen unserer IMK Expert:innen für 2026: www.imk-boeckler.de/de/so-wird-2026… #Sowird2026 #IMK 023
halvarflake.bsky.social @halvarflake.bsky.social · 05/01/2026LLMs are reshaping software dev. I don't buy "the end of software dev": Project ambition will grow dramatically. Ancient Egyptians could build the Pyramids but not the Empire State Building. Pre-LLM software will be viewed like we view the Pyramids. 4255
halvarflake.bsky.social @halvarflake.bsky.social · 03/01/2026I find the US taking Maduro to be entirely in-character (remember Noreaga?). The weird part is having just pardoned Hernandez after he did the same thing. 5202
Reposted by @halvarflake.bsky.socialDon Moynihan @donmoyn.bsky.social · 03/01/2026Expect to see a lot of this classic Onion headline, from 23 years ago, over the coming weeks. 74162234113
Reposted by @halvarflake.bsky.socialrob pike @robpike.io · 25/12/2025Fuck you people. Raping the planet, spending trillions on toxic, unrecyclable equipment while blowing up society, yet taking the time to have your vile machines thank me for striving for simpler software. Just fuck you. Fuck you all. I can't remember the last time I was this angry. 10081902203
halvarflake.bsky.social @halvarflake.bsky.social · 10/11/2025In the same way there are kleptomaniacs that can't keep themselves from stealing stuff, I suffer from some form of debug-o-mania where I seem to subconsciously seek out life situations that make me spend time debugging in gdb or rr. 1211
halvarflake.bsky.social @halvarflake.bsky.social · 05/11/2025Declared inbox bancruptcy and archived 13k conversations. GMail is very very buggy when mass archiving stuff. 0190
Reposted by @halvarflake.bsky.socialMeredith Whittaker @meredithmeredith.bsky.social · 27/10/2025Meaning that infrastructure like AWS is not something that Signal, or almost anyone else, could afford to just “spin up.” Which is why nearly everyone that manages a real-time service–from Signal, to X, to Palantir, to Mastodon–rely at least in part on services provisioned by these companies. 8/ 270829
halvarflake.bsky.social @halvarflake.bsky.social · 23/10/2025I wonder why every LLM provider insists in having their own CLI agent. Can I have a merge between codex, Gemini-cli and Claude code with swappable models plz? 481
Reposted by @halvarflake.bsky.socialBloomberg News @bloomberg.com · 22/10/2025The French crown jewels stolen from the Louvre are steeped in history — and priceless. What happens when they make it to the black market?bloom.bgIs There a Price Tag for the Louvre’s Stolen Jewels?As soon as the gems hit the black market they could be much less valuable, and their historic value drops. 2216
Reposted by @halvarflake.bsky.socialEugene Vinitsky 🍒 @eugenevinitsky.bsky.social · 20/10/2025Making a firm commitment to asking the question that reveals ignorance until said ignorance is gone 1243
Reposted by @halvarflake.bsky.socialJohn Scott-Railton @jsrailton.bsky.social · 18/10/20253/ NSO also emerges from the WhatsApp v NSO case with just an absolute TON of their business splashed all over the court records.. Ouch E.g. check out the filings from Sept 15th 2025: www.courtlistener.com/docket/16395...courtlistener.comWhatsApp Inc. v. NSO Group Technologies Limited, 4:19-cv-07123 - CourtListener.comDocket for WhatsApp Inc. v. NSO Group Technologies Limited, 4:19-cv-07123 — Brought to you by Free Law Project, a non-profit dedicated to creating high quality open legal information. 23310
Reposted by @halvarflake.bsky.socialJohn Scott-Railton @jsrailton.bsky.social · 18/10/20252/ Altho massive punitive damages jury award against NSO Group ($167m) got reduced by the court, as is expected in cases where it is so large (to 9x compensatory damages)... This is likely cold comfort to NSO since I think the injunction seriously dims value of NSO's spyware product. 1255
halvarflake.bsky.social @halvarflake.bsky.social · 11/10/2025#hexacon closing presentation from amnesty international about abuse of offensive tools in Serbia, Hungary, and the use of such tools by ICE. Strong choice, very good conference. 2112
halvarflake.bsky.social @halvarflake.bsky.social · 09/10/2025There's an argument to be made that this new security research from Anthropic also has copyright implications. www.anthropic.com/research/sma...anthropic.comA small number of samples can poison LLMs of any sizeAnthropic research on data-poisoning attacks in large language models 1101