Reposted by Erica "digifox" KovacChrononaut @chrononaut.bsky.social · 19hAre the AI-generated design docs my colleagues give me to review too long and convoluted? Yes. But do they contain fatal errors, that you find only after wasting hours to carefully review them word by word? Also yes. bsky.app/profile/chro... 131
Erica "digifox" Kovac @digifox.binaryden.net · 19hThis is also the case with technical design docs and even API documentation, by the way. The number of times I had an idea for something and it wasn't until I wrote the design doc out that I realized it had major flaws that may have gone unnoticed once I was heads-down in code is... fairly high. 2111
Reposted by Erica "digifox" KovacWilliam B. Fuckley @opinionhaver.bsky.social · 19hHowever, that is not the point of other types of writing. For the theory and literature review sections, the work that writing has done is not just conveying information, it is assuring you, the reader, that someone has written this and been forced to think about it. I think this is very bad to LLM! 1662
Reposted by Erica "digifox" KovacSisyphus, but for reducing tail latency @bluespacecanary.bsky.social · 21hevery line of memory-unsafe code needs to be treated as a critical security vulnerability by default 1153
Reposted by Erica "digifox" Kovacrev. howard arson @theophite.bsky.social · 28/09/2026in the real world, synthetic resampling with anisotropic noise just disappears any evidence beneath the noise floor while making your conclusions increasingly radical and disjoint from the signsl 2743
Reposted by Erica "digifox" Kovacrev. howard arson @theophite.bsky.social · 28/09/2026right, yes. one of the basic problems is that they think that thinking about something really hard is a type of sampling, and that you actually don't need to sample anything to get information, and that there is no difference between thinking about something and doing it. 1645643
Reposted by Erica "digifox" Kovacponder @ponder.ooo · 29/09/2026the rationalist scene has always been inhospitable to people with *actual* epistemic standards 1242
Reposted by Erica "digifox" Kovacrev. howard arson @theophite.bsky.social · 28/09/2026no, "rationalist" isn't ironic. they specifically contrast themselves with empiricism -- and they literally believe that they can derive everything in their light cone from first principles to the disappearance of the last neutron. 1843438
Reposted by Erica "digifox" Kovacrev. howard arson @theophite.bsky.social · 28/09/2026i think RFK Jr.'s colon-destroying steak and sauerkraut diet, tradwife culture, and Silicon Valley CNC parties are the same thing: a desperate attempt among powerful men to convince each other that gluttony, rapaciousness, and raw domination are not just their birthright but their responsibility 191491238
Reposted by Erica "digifox" KovacMax Kennerly @maxkennerly.bsky.social · 28/09/2026It's fine if people don't want to have liberal arts education. Go ahead, have large gaps in understanding the natural sciences, social sciences, and humanities. Live your life ignorant, incomplete, undeveloped, inadequate. You have the right to be a worse citizen, a shallow dullard, a lesser person. 1859067
Reposted by Erica "digifox" Kovacponder @ponder.ooo · 28/09/2026the main discovery behind llms imo was a new class of architecture that does a better job of approximating a kind of function that people had been interested in approximating for a long time (causal language modeling) 1282
Reposted by Erica "digifox" KovacJustin Wolfers @justinwolfers.bsky.social · 28/09/2026My belief: This election isn’t about the economy in 2027. It’s about the economy in 2060. Prosperity comes from institutions like the rule of law, markets, a free press, and capitalism (not crony capitalism) newsletter.platypuseconomics.com/p/…newsletter.platypuseconomics.comWe Spend So Much Time Talking About Gas PricesBut is that what we really want to focus on? Also: Is the U.S. economy really the envy of the world? 29522129
Reposted by Erica "digifox" KovacEugene Vinitsky 🍒 @eugenevinitsky.bsky.social · 28/09/2026Agents calling agents is fine. Humans calling humans is fine. I do not ever want to be called by an AI agent 101165
Reposted by Erica "digifox" Kovacponder @ponder.ooo · 28/09/2026meanwhile the main *empirical* results they gesture at to back up the theory on that forum are instances of their own companies' unusual incompetence at responsibly managing ai systems compared to other ai companies 0171
Reposted by Erica "digifox" Kovacponder @ponder.ooo · 28/09/2026their main theoretical argument for why ai is a uniquely dangerous field only they should be allowed to practice is that that's the consensus that has formed on their niche online forum where agreeing with that point has from the inception of the forum been considered table stakes for participation 1313
Reposted by Erica "digifox" Kovaceevee @eev.ee · 24/09/2026"cancel nitro" seems like a tough sell when paying for nitro is very likely to make discord classify you as an adult automatically 5341
Reposted by Erica "digifox" KovacYael Grauer @yaelwrites.com · 24/09/2026I don't need to connect my fill in the blank to the internet arstechnica.com/gadgets/2026...arstechnica.comOwners mourn spoiled food after firmware update bricks Samsung smart fridgesSo absurd and embarrassing ..." 02410
Erica "digifox" Kovac @digifox.binaryden.net · 24/09/2026The good news is there's basically no reason to think this will make it out of committee 092
Reposted by Erica "digifox" Kovaceva (^_^)/ @eva.computer · 24/09/2026oh and losing a charter can result in your assets being seized and all of your work destroyed but it’s fine because the office of the president has never been used for vindictive purposes 412715
Reposted by Erica "digifox" Kovaceva (^_^)/ @eva.computer · 24/09/2026so dangerous that the government gets to set arbitrary floating point math speed limits UNLESS you get a special certificate then it’s fine :) 213411
Reposted by Erica "digifox" Kovaceva (^_^)/ @eva.computer · 24/09/2026and there it is: regulatory capture 527263
Reposted by Erica "digifox" Kovacae @aelkus.bsky.social · 24/09/2026People really need to go back to the original Jurassic Park and look at the rather disturbing ways in which InGen's management of it now looks prescient 330885
Reposted by Erica "digifox" KovacChatham Harrison dba TRUMP DELENDUS EST @chathamharrison.bsky.social · 24/09/2026Silicon Valley's damage becomes a lot easier to understand when you realize how much of their innovation involves the government not enforcing laws, or enforcing them too feebly for anyone to care 510819
Reposted by Erica "digifox" KovacHulex DJing Furpoc! Sun 11pm @hulex.bsky.social · 24/09/2026People are just getting banned en masse for falsely being flagged for violating “child safety” Fuck ai and fuck discord they’re so fuckin washed at this point 384
Reposted by Erica "digifox" Kovac☀️ Starshine @starshine.bsky.social · 23/09/2026The Expanse: gods are real, they noticed you, and they are PISSED 1220111
Reposted by Erica "digifox" KovacElixir of Progress @elucidating.extradimensional.space · 23/09/2026The entire edifice of science is currently under attack because it's been so radically successful at spreading public knowledge. It seems like we need to solve the plight of scientists today with or without the models. I'm much more worried about that angle. 061
Reposted by Erica "digifox" KovacTed Underwood @tedunderwood.com · 23/09/2026Something that worries me more every day: I suspect most fields have 5-25 breakthrough ideas latent in existing data and literature. Say you’re a midsize AI company looking for cred. Why not hire a *small* team of domain experts + 1000 agents … and promote your models as the ones that broke Econ? 12807
Reposted by Erica "digifox" KovacEugene Vinitsky 🍒 @eugenevinitsky.bsky.social · 23/09/2026If you view AI as filling in the convex hull of knowledge, this is great in the short term but bad in the long term if there's no one left to expand the hull after 1015715
Reposted by Erica "digifox" KovacDr. Samantha Hancox-Li @sjshancoxli.liberalcurrents.com · 22/09/2026over on musk-X they're trying to cancel us for telling the truth: democrats wouldn't piss on donald trump if he was on fire 33996120
Reposted by Erica "digifox" KovacDoll @dollspace.gay · 19/09/2026The pro-AI coalition only looks unified because everyone is shooting at the same anti AI barricade from wildly incompatible metaphysical positions. 421228
Reposted by Erica "digifox" KovacDoll @dollspace.gay · 19/09/2026Chaotic Good: Open-weight liberationists “Nobody should need corporate or state permission to think with a machine. Release the weights.” 1994
Erica "digifox" Kovac @digifox.binaryden.net · 22/09/2026This is frustrating but they do at least support the Play/App store age range APIs, which are a much better solution to this problem imho. 151
Reposted by Erica "digifox" KovacRona Altana @ronaaltana.bsky.social · 22/09/2026Discord finally confirmed for real when they're rolling out age verification service-wide (tomorrow) and I've not seen a single person make a peep about itdiscord.comSafer for Teens, Same Discord for AdultsWe’re launching a revamped, privacy-focused age assurance solution including methods that won’t require an ID or selfie. Here’s how it works. 6727332411
Erica "digifox" Kovac @digifox.binaryden.net · 21/09/2026It's a failure of our software engineering as a discipline that a bank website can load slowly, like, ever. 2352
Reposted by Erica "digifox" KovacTed Underwood @tedunderwood.com · 21/09/2026Another story about institutions deciding they can build on open models rather than pay premium to a frontier lab. h/t @ver.ooobloomberg.comOpenAI, Anthropic Costs Push More Startups to Build Off Cheaper Open ModelsThe $15.6 billion legal startup Harvey built its business around training AI models like OpenAI’s GPT-4 to do specialized work for lawyers. Recently, however, soaring artificial intelligence costs hav... 192
Reposted by Erica "digifox" KovacFreedom of the Press Foundation @freedom.press · 21/09/2026David Ellison should put his mouth where his money is. 44917
Reposted by Erica "digifox" KovacMaterialist Gnostic @walmsley.bsky.social · 21/09/2026it's sort of like a baby on a plane. I have a ton of sympathy for my friend who just got off a flight and wants to complain about the baby who was on it with them. I have no sympathy for the person who wants a policy fix for crying babies 41017
Reposted by Erica "digifox" KovacNatanael, Tech janitor @natanael.bsky.social · 21/09/2026Meanwhile, Facebook doesn't ban nazistheguardian.comMeta bans ads for Virginia Woolf play in Spain because they relate to feminismTheatre told that Facebook and Instagram filter out ads mentioning ‘social topics’ such as civil rights or feminism 03117
Reposted by Erica "digifox" KovacSharon 🪳🌹 @sharonk.bsky.social · 21/09/2026you want teen takeovers to end? provide youth with third spaces that are easily accessible and visible alongside redirecting them to such activities 624416
Reposted by Erica "digifox" Kovac#StupidSuez @diplomatofnight.com · 21/09/2026Chris Van Hollen issues the first ever call by a Democratic member of Congress for US officials to be sanctioned by other countries 201907417
Reposted by Erica "digifox" KovacInternet Archive @archive.org · 21/09/2026Local and state government records help communities understand their history and participate in public life. But digital records can be hard to preserve and access. The Internet Archive’s new ATLAS program will help address that. Learn more 👉 blog.archive.org/2026/09/10/a... 🧵 118546
Reposted by Erica "digifox" KovacSenator Ron Wyden @wyden.senate.gov · 21/09/2026I called out Flock a year ago. BUT, it is grassroots organizing that made it a national movement against mass surveillance. 1458292
Reposted by Erica "digifox" KovacChristo Silvia @christosilvia.bsky.social · 21/09/2026AI makes it easier to write software, and that's bad 1436
Reposted by Erica "digifox" KovacPavel @spavel.bsky.social · 21/09/2026"Higher velocity = more good" is a misunderstanding premised on the idea that every piece of work on your Jira board is there because it adds value in some way. Once you understand that the majority of developer tickets are actually negative-value, you will become enlightened. 48915
Reposted by Erica "digifox" KovacElixir of Progress @elucidating.extradimensional.space · 21/09/2026Some folks have used my post on Superintelligence and the assumption of value to be dismissive of the seriousness of AI alignment problems. Please do not! I think the thing we have today is much weirder and scarier than superintelligence. It's supertenaciousness. 1101
Reposted by Erica "digifox" KovacTommaso Sciortino @sciortino.bsky.social · 20/09/2026My reaction: Fine! Instead of saying LLMs "think" just say that they "think2" a new word I made up just now that means "to form ideas, make choices, or have opinions." 331
Erica "digifox" Kovac @digifox.binaryden.net · 20/09/2026No one has proposed a fix for this that isn't "we need to make gatekeeping infrastructure mandatory," which is clearly bad. I don't think it's a worthwhile angle to approach things from. 061
Erica "digifox" Kovac @digifox.binaryden.net · 20/09/2026Somewhere between Neutral-Neutral and Neutral-Lover. 0110
Reposted by Erica "digifox" KovacEli Mallon @iame.li · 19/09/2026So is the superintelligent AI's plan to kill all the humans and _then_ figure out how to automate mining rare earth metals and construct EUV lithography machines so it can print new chips without humans in the loop? To like... continue existing? Wow yeah that sounds really smart I'm so scared 10928
Reposted by Erica "digifox" KovacErica Windisch @ewindisch.ontological.observer · 19/09/2026In 2024, I found indicators that I could likely escape OpenAI's sandboxes. I did not demonstrate an escape because I felt there was significant enough information to make a responsible report. I was told that this was not of interest to their bounty program and they closed my report. 29315