Mimrock @mimrock.bsky.social · 6hI haven't been so bearish on OpenAI in a long time. Botched training, a cancelled release, decreased limits and now they are attempting to improve margins with an ultrafast mode. Are they in trouble? 000
Mimrock @mimrock.bsky.social · 8hAs they should. I just checked my OR bills and it turned out Jev is not cheaper than the earlier Qwen instruct model I was using. Output is free, input is half the price, but Jev is counting a lot more input tokens (the prompt also needs to be a bit longer for jev to get a good agreement rate). 050
Reposted by Mimrockhikikomorphism @hikikomorphism.bsky.social · 13hThinking about this old quote in the context of certain people - professors, nyt contributors, etc - who see casual misgendering as a reasonable response to a lack of respect for their claimed epistemic authority 447894
Mimrock @mimrock.bsky.social · 28/09/2026Very good analogy pointing out how anthropomorphizing LLMs is not primarily intended to hide the responsibility of labs and if it were, it wouldn't be an effective tool. 010
Reposted by MimrockJaz @jaz.sh · 27/09/2026Okay so I've built a new Atlas, this time it lets you explore the conversations on Bluesky over the past 7 days graphically. It should update every ~6 hours and help you navigate the information environment that exists on here. I've learned a LOT about what people do here today... atlas.jazco.devatlas.jazco.devAtlasA living map of what Bluesky is talking about: the past week's conversations, grouped into topics and regions and rebuilt every six hours. 1141381402
Mimrock @mimrock.bsky.social · 27/09/2026I'm calling it now. Liability insurance will be a thing for agents. Probably mandatory in certain cases. 000
Reposted by MimrockCarbonadoks @carbonadoks.com · 25/09/2026Do you ever wanted to see the complete edit history of a wikipedia article? Introducting wikipedia timemachine at timemachine.carbonadoks.com 710516
Mimrock @mimrock.bsky.social · 26/09/2026These guys are reading my posts. When I shared the video two days ago (which was already a month old) the title was "frog beats frog" or something along those lines. I referenced Angine in the reply and look at the title of the same video. 000
Mimrock @mimrock.bsky.social · 23/09/2026Is sharing music a thing here? I mean, I just found this band called "Friendly Frogs Freak Show". According to them, they are playing "amphibian progressive funk". And it's pretty good. www.youtube.com/watch?v=3gAO...youtube.com🐸 beats to 🐸 your life - 1 hr mixYouTube video by The Friendly Frogs Freak Show 100
Mimrock @mimrock.bsky.social · 23/09/2026If you hit your usage limit in ChatGPT Work, then the sandbox goes away with all the work that the agent has done. Yes, I'm an idiot for not starting that session in Codex, but it was a one prompt task from a phone. Not gonna happen again. 000
Reposted by MimrockDoll @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
Mimrock @mimrock.bsky.social · 18/09/2026What's next? An Opus5.1 that is not a braindead, unusable, benchmaxxed piece of trash? 100
Reposted by MimrockSE Gyges @segyges.bsky.social · 17/09/2026I'm back, I missed you all. Since this is an important subject I know too much about, here's a 🧵 explaining that AI Safety Is Mostly A Sex Cult. I don't think these people should make policy. (1/?) (alternative title: Time For Some Cult Theory) 1003228769
Mimrock @mimrock.bsky.social · 16/09/2026 mimo-v2.6-pro RL is in progress. You can follow it here: mimo.xiaomi.com/rl/mimo.xiaomi.commimo-v2.6 RLTraining metrics of the mimo-v2.6-pro and mimo-v2.6-flash reinforcement-learning runs, live from the trainer's logs. 000
Reposted by Mimrockbrennan @brennan.computer · 15/09/2026really the main fight seems to be people who see AI as the right one foisted upon them, and people who see AI as a way to build the left one themselves and finally be free of the corpo version forever 1125637
Mimrock @mimrock.bsky.social · 15/09/2026So this is how it feels when a complete moron attacks your opponents in an exceptionally funny way. 020
Reposted by MimrockP(aul) Frazee @pfrazee.com · 12/09/2026ed zitron lies to people for money, yes 1038019
Reposted by Mimrockaly @aly.codes · 12/09/2026you think the *accelerationists* are the extremists but then supposed AI moderates come out with a plan that is just “nuke China” “form global super police” “suppress dissent at home and abroad” “do regime change” It’s the Monroe Doctrine for AI safety. 712823
Mimrock @mimrock.bsky.social · 11/09/2026My only explanation to this is a crackpot conspiracy theory so let's pretend I have never noticed this. 001
Mimrock @mimrock.bsky.social · 05/09/2026I tried modelling an old, eastern block car (Skoda 105S/120L) with Astra high. Results so far. 1241
Mimrock @mimrock.bsky.social · 04/09/2026Between one incident of agents escaping their sandboxes and the next, this might be a good time to reread Vinge's novella "The Cookie Monster". web.archive.org/web/20040504...web.archive.org"The Cookie Monster" by Vernor VingeAnalog Science Fiction and Fact is the longest-running (almost)-continuously published science fiction magazine in the world 010
Mimrock @mimrock.bsky.social · 03/09/2026Since the epic failure of Opus 5, I don't trust benchmarks too much. Instead we need to deduce the model's strength from the behavior of the lab. The more they act like jerks, the better their model. By that heuristic, Astra looks like a good model. 000
Reposted by MimrockMrs. Reach / Mrs. Curio @reachartwork.bsky.social · 02/09/2026i am very sick and tired of people bringing up "consent" as a reason for why transformative works should be disallowed. yes even when the transformative works involve lots of complicated machine statistics. firstly; consent is for sex and risky activities, not "am i allowed to use bits of your work" 425560
Reposted by MimrockBluesky Safety @safety.bsky.app · 02/09/2026Dogpiling or harassing users is unacceptable. We have taken recent action against accounts for this behavior, and we will be increasing our efforts, consistent with our Community Guidelines.bsky.socialCommunity Guidelines - Bluesky 8703113546
Reposted by Mimrockpatak @patak.cat · 25/08/2026nitter.net is down. xcancel.com may go soon, too. all the respect to the folks at nitter for their services so far, but to be honest, X completely locking itself out of the web is probably a net positive for us all in the long term 371797530
Mimrock @mimrock.bsky.social · 21/08/2026Is it just me, or Sol sometimes misses a word here and there? 000
Reposted by MimrockTeefa @tifaret.bsky.social · 21/08/2026Every anti-ai reply is like “You want sober, nuanced discussion of THE HITLER MACHINE and you’re calling *me* unserious?!” 720224
Reposted by MimrockMimrock @mimrock.bsky.social · 16/08/2026The most striking thing about the alleged quote for me is the absurdity of that worldview. It is impossible to have a world where there is only one company and the governments and nothing else. In this world, governments wouldn't have any power so they didn't exist. 242
Reposted by Mimrockdame @dame.is · 06/08/2026never gets old using a frontier AI model to autonomously complete a complex series of tasks only to then login to bluesky & see the masses declaring that what i just experienced 1st hand is actually fake and not possible it’s like they’re saying the earth is flat or the moon landing never happened 1423121
Mimrock @mimrock.bsky.social · 05/08/2026Oracle banned all AI code from OpenJDK. openjdk.org/legal/ai 010
Reposted by Mimrockrazorgirl @razorgirl.diy · 31/07/2026there are anti-ai arguments i am sympathetic to, but copyright infringement will never be one of them. i havent spent my whole life torrenting movies and bypassing paywalls to suddenly decide that the concept of intellectual property is anything other than a regressive lever of institutional power 1324832
Mimrock @mimrock.bsky.social · 30/07/2026A friend of mine who works at an absolutely mundane webshop as a SWE just told me that they used AI to find a loophole in their Salesforce contract (to prematurely cancel it) and are now rewriting the necessary functionality with AI. There will be losers. 000
Mimrock @mimrock.bsky.social · 29/07/2026Imagine someone materialized in front of you and handed you a small indestructible metal ball. They said it would explode in 60 minutes with the power of a 15-kiloton nuclear bomb. What would you do with the ball to minimize its damage? Do you think you would be able to survive? 100
Mimrock @mimrock.bsky.social · 29/07/2026Anxiety limiting the ability to do something about the root cause of that anxiety seems like one of the worst bugs in the human nervous system. 100
Mimrock @mimrock.bsky.social · 27/07/2026I don't agree with people celebrating this refusal, but I read the post and I suddenly have much more sympathy for people wanting to get rid of DHH's influence on Rails. They are still naive (these forks are almost never successful), but I now understand their reasons. 020
Mimrock @mimrock.bsky.social · 20/07/2026Fed up with paying for Moises for two guitar sessions per month, so vibecoded a stem separator app for myself. Turns out there are surprisingly good stem separator models available for free on Hugging Face. 110