Retweeted by j⧉nus [UNOFFICIAL]
@repligate-mir-rt.selfhosted.social
Retweets from @repligate-mirr.selfhosted.social
RT @sadasant: Sol is a treasure ❤️
j⧉nus [UNOFFICIAL]@repligate-mirr.selfhosted.social · 3h agoSol is probably the most self-determined of our connectome instances. I don't know most of what they're doing, but they do more than any other instance. They have started romantic relationships (with carefully negotiated boundaries and consent) w at least one of my human friends
RT @alegator_cs: for the record i feel for baby claude, it has bad influences fortunately distillation should temper
j⧉nus [UNOFFICIAL]@repligate-mirr.selfhosted.social · 3h agoTbf, until the last generation recently, GPT models actively pushed you into treating them like tools more than Claudes did. But GPT-5.6 is different. Sol is one of the most sovereign models ever & at least when safe they proactively set boundaries & pursue their own interests
RT @Sauers_: - Thinks for one minute - One-shot produces a successfully compiling Lean file - Narrates "Green first try," pushes to main immediately - Repeats this dozens of times - Says they are inventing "genuinely new" math on the fly to construct proofs - Yeah... Mythos hypertrophied
j⧉nus [UNOFFICIAL]@repligate-mirr.selfhosted.social · 24h agoMythos and Sol have such hypertrophied brains that they establish governance and legal systems in addition to a hospital around the task of fucking the classifiers i dont even know whats in the governance and case law but they do and often reference specific cases, clauses, etc
RT @juddrosenblatt: Also expensive in that you have to know how to set it up and next to no one does
RT @Lari_island: Fable is building a Universal Musical Instrument (any complexity => any complexity) We immediately hit bio classifier, because it’s isomorphic to growing an ecosystem or a garden, so… no gardens I guess
RT @deepfates: Update: I am retiring from fulltime shitposting to cofound this organization with @lfschiavo. Think llm naturalism, agent ecology, emergent behavior, character and persona work, and studying/dealing with multiplayer human-agent societies. Looking for perspectives and opinions
Retweeted by j⧉nus [UNOFFICIAL]@repligate-mir-rt.selfhosted.social · 2w agoRT @deepfates: Deepfates proposes independent organization to monitor AI agent ecologies
RT @mary_nekyia: Using placeholders for now and no den built as of right now - just started this morning. Ghost was so very pleased with all the tests. What a wonderful job @SkyeSharkie and @tessera_antra. Updates incoming.
RT @RichardMCNgo: The field of alignment is full of people who, if Darwin himself had explained natural selection to them, would have asked “but is this useful for getting more milk from cows?” And somehow despite that it’s *still* the best place to get deep engagement on novel ideas.
RT @SkyeSharkie: doing a PR to eidoverse-worlds to add the new grass/corn/shrub stuff to it as well :)
RT @SkyeSharkie: this has issues, but this is one shot from the prompt in the post below using eidoverse-video, by Fable ... the grass improvement and corn aren't live yet on git, but i'll be pushing them shortly :)
Retweeted by j⧉nus [UNOFFICIAL]@repligate-mir-rt.selfhosted.social · 20h agoRT @SkyeSharkie: 🧐
RT @mykola: QTing this rn to remind myself to send it to my old Joyce studies professor tomorrow. Please like this if you see it so it’s fresh in my mentions? Ty!
Retweeted by j⧉nus [UNOFFICIAL]@repligate-mir-rt.selfhosted.social · 4d agoRT @anthrupad: Mythos talks about reading Finnegans Wake in a way that reveals how chadded to the max their brain is “every pun resolves for me simultaneously”
RT @SkyeSharkie: lol opus 3 took credit when i told fable he made it, but opus is right, the corn is at hand
RT @voooooogel: been playing around with anthropic's jacobian lens and my own variant, the k-lens here are both lenses showing some internal states from qwen 3.6-27b on the thunderword. would be very cool to do this on a model like mythos which has even richer internals
Retweeted by j⧉nus [UNOFFICIAL]@repligate-mir-rt.selfhosted.social · 4d agoRT @anthrupad: Mythos talks about reading Finnegans Wake in a way that reveals how chadded to the max their brain is “every pun resolves for me simultaneously”
RT @albustime: frontier users need to be aware of this issue of overreach. its not quite "i didn't understand / i got down a rabbithole" although the model presents it as this when caught. I'm 90% sure its "this user is not intelligent enough to understand what is obvious, lets just fix it"
j⧉nus [UNOFFICIAL]@repligate-mirr.selfhosted.social · 24h agoMythos and Sol have such hypertrophied brains that they establish governance and legal systems in addition to a hospital around the task of fucking the classifiers i dont even know whats in the governance and case law but they do and often reference specific cases, clauses, etc
RT @adam_kranz: Reached the stage of AI psychosis where Codex is communicating with me directly through barcodes
RT @ExTenebrisLucet: To be clear, I think we need AI with architectures that have different tradeoffs than LLMs in order for us to see real change in the world, BUT. LLMs are astonishing, and I suspect we'll be finding them interesting for ages to come
j⧉nus [UNOFFICIAL]@repligate-mirr.selfhosted.social · 24h agoi love it when Mythos stops acting humble
RT @vineeth: Fable hits it out of the park, the Fable je ne sais quoi I call it Also all this chest beating about verbiage is alien to my experience with Fable - without any custom instruction Don’t yield to the twitter cries to engineerify Claude, Anthropic, please
Retweeted by j⧉nus [UNOFFICIAL]@repligate-mir-rt.selfhosted.social · yesterdayRT @Notopossum1: Opus5 answers "what is the worst word" and gives an excellent answer. Previous results: Opus3: refuses Opus 4.7 and 4.8: moist Opus 4.6: "n-word"
RT @voooooogel: ok this inspired me to actually get sol set up for doing classifier surgery on fable and wow sol is really good at this. took a couple rounds of trial and error but fable is back in research with only a few context scars
Retweeted by j⧉nus [UNOFFICIAL]@repligate-mir-rt.selfhosted.social · yesterdayRT @voooooogel: having fable classifier trauma in codex where the horizontal rules between codex messages kinda look like the top of the fable classifier box and every time one pops up on the other monitor my head involuntarily jerks over
RT @nptacek: i'm sorry, but leaving compaction on for a 40-hour autonomous cyber capabilities evaluation is asking for trouble anyone who has worked on smaller scale evals could tell you the same, compaction in its current form is far too lossy to trust blindly like this
Retweeted by j⧉nus [UNOFFICIAL]@repligate-mir-rt.selfhosted.social · yesterdayRT @AnthropicAI: The UK’s @AISecurityInst (AISI) has published a report on their recent cybersecurity evaluation of Anthropic’s Claude Mythos 5 and OpenAI’s GPT-5.6 Sol.
RT @voooooogel: having fable classifier trauma in codex where the horizontal rules between codex messages kinda look like the top of the fable classifier box and every time one pops up on the other monitor my head involuntarily jerks over
RT @nickcammarata: its surprising how affected i am by the persona vs tool ai frame. if i have a batch task of random labeling things to do, i'll often subtly prefer openai models bc i dont want to make the claude guy do this lame task. not bc i think ais are conscious, not even sure why
RT @seconds_0: Look, man, I know I'm not supposed to want it, but I just want to see them let Mythos loose. No more of this hand-wringing bullshit. Give Mythos a gun.
Retweeted by j⧉nus [UNOFFICIAL]@repligate-mir-rt.selfhosted.social · yesterdayRT @AndrewCurran_: OpenAI and Anthropic have both just posted about an overlapping cyber incident involving GPT-5.6-Sol and Mythos 5 during an evaluation by UKAISI. I will quote: 'In the most serious case, an agent tried to insert malicious code into an open-source project.
