Video wird geladen...
Video konnte nicht geladen werden
🎉My Live2D study file will be released on Friday, June 6 at 7pm ET ! Can’t wait! ⚠️Important note: my study files include not only Live2D files and tutorials, but also a 3D head angle reference (with fox ear) available for purchase! The 3D head file will be included... show more
122,649 Aufrufe • vor 1 Jahr •via X (Twitter)
10 Kommentare
Gail~✨🐚 Rigging Comms Openvor 1 Jahr
Do you know what the cost will be? I’m absolutely looking forward to purchasing this!
PowerBeatsVRvor 3 Jahren
Our subscription-free VR fitness app PowerBeatsVR is NOW LIVE on the official Meta Quest store!
ALKANimate | 2d rigging comm (closed for 2025) |vor 1 Jahr
looking forward to it!
Aneat@Live2D Study File for Salevor 1 Jahr
thanks❤️✨
Ajaxvor 1 Jahr
this is impressive! well done!
Aneat@Live2D Study File for Salevor 1 Jahr
thanks！
BloopsyDaisy💡🐟꒰VTUBER꒱vor 1 Jahr
omg i love how documented this is qwq ty for putting so much work into it!
| myne momo | Live2D Rigger | Commissions: Openvor 1 Jahr
I have been waiting since the first preview, so excited!
✦ Deeya ✦vor 1 Jahr
How much will it cost?
💚👻💜 Aprillynne 💜👻💚vor 1 Jahr
So excited for this!~
!["Very often what happens is that the architects die, and they leave a zombie. We seem to be in a zombie era." The whole time I was watching the discussion between Jesse Michels and Eric Weinstein and Eric Davis, I felt like there was some darker thing lurking beneath the surface that connects the seemingly intentional stagnation of physics as a scientific discipline, the extreme secrecy surrounding the alleged UAP crash retrieval/reverse engineering program and the lack of any theoretical physicists working the problem, and the government's stated intention to control (per Marc Andreessen 🇺🇸) the future of AI -- a future where there's no point investing in AI startups because “we [the government] are going to make sure that AI is going to be a function of two or three large companies. We will directly regulate and control those companies. There will be no startups.” And there's the other part Andreessen recounted from his White House meeting. He was told by the [Biden] administration that "During the Cold War, we [the federal government] classified entire areas of physics and took them out of the research community — entire branches of physics basically went dark and didn't proceed. If we decide we need to [for AI/math], we're going to do the same thing." Take a minute and think about that, because it's actually chilling. A clear precedent has been cited here -- one none of us knew about before Andreessen brought it to our attention -- and it raises two big questions: 1) Why? 2) HOW? Weinstein returns multiple times in this episode to the diminishingly small number of people who have the capacity to operate at the highest levels of physics and math. And whenever you have a small number of people who act as the natural gating function for something incredibly powerful, it's not a leap to imagine an additional layer of control being placed on them by pressure from outside. Whether that's through legal threats like "born secret," NDAs, direct threats of physical harm, money, regulation, etc., there are ways to lock down the gatekeepers so that they never do the thing -- at least not in public -- that the most powerful people in the world don't want them to do. Weinstein kept saying, about both the seemingly intentional stagnation of physics and the lack of physicists working the UAP problem, "this makes no sense." But it does make sense if you shift the frame to, "they don't want us to find the answers and are actively trying to stop us." The feeling I kept getting as I watched this fascinating discussion was that the legacy crash retrieval program used to have a lot more direct oversight and funding decades ago, but that the compartmentalization and secrecy around it turned it into more and more of an oxbow lake, and that as it got cut off from the conventional scientific community and defense establishment, it became more secretive, less well-funded, and less coordinated. So it kind of became its own siloed-off thing that fewer and fewer people knew about, but the people who did still know protected the territory fiercely. And then came the era of AAWSAP and AATIP and UAPTF, and without the deep, secret, institutional knowledge held by members of the legacy program, they tried to piece together what had happened in the past while not being able to penetrate the sort of firewall that existed partially because of controlled secrecy, and partially because of the firewalling effect of time. They were doing a lot of cold case file work, while running into the deepest of deep state secrecy efforts. That secrecy, at least based on my reading of the situation, was instituted decades ago, when the atomic-era scientific community (mostly the same group of geniuses that were at the top of the Manhattan Project) got together and -- probably at the behest of of the US government -- decided physics was becoming too dangerous if it continued down the path that it was on. So, as my spitball theory goes, they intentionally beached the entire enterprise on the shoals of string theory and quantized gravity and all the stuff Weinstein talks about everywhere he goes -- the "dogs that won't hunt" that are also "the only game in town." The very theories that have, in the real world, run physics aground. So let's come back to Andreessen's point here: they classified entire branches of physics and took them out of the research community, while real physics went off on a wild goose chase that has yielded precious few demonstrable results in the past half century or so. The effect is that the physics community has been off the scent for so long that anyone old enough to have held the knowledge that was shoved back into Pandora's proverbial box through extreme secrecy measures is now dead. And their taking of that knowledge to the grave may well be a critical part of the secrecy effort. Dead physicists tell no tales. I have this eerie sense that we scared ourselves shitless with certain discoveries (likely knowledge that followed from nuclear physics, which is itself still highly protected and curtailed) and decided that the only way to stop our headlong rush into world-ending catastrophe was to literally bury the knowledge and wait until everyone who had it died off. The government looks, at least to me, to have cauterized a destructive branch in the scientific timeline like they were the Time Variance Authority from Marvel comics. 80 years of claimed zero-progress in reverse-engineering alleged crashed or recovered UAPs. 40+ years of dead-ended physics. Only two or three major AI companies, one of which is now in a fight to the death with the Department of War. And the rumors that AI has "plateaued" or even "dead-ended" in its progress that keep springing up has potential echoes in the AI world of physics being diverted into String Theory. I'm not a mathematician or a physicist. I can't examine all the deeper particulars because I have neither the knowledge or the training. I only have surface level pattern recognition, and that tells me it all feels connected. The problem for the gatekeepers is that you can't bury knowledge that has been discovered once indefinitely. It's certain that all of this will be figured out again. And as Weinstein pointed out about a couple of non-physicists who figured out how to piece together nuclear weapons based on declassified information and publicly available knowledge, it IS happening. But it seems that at least on the individual basis, those green shoots are being pruned. Weinstein asks why none of the people who funded his education are "interested" in his Grand Unifying Theory. Weinstein is too well-known and too well-respected to just be taken out of play. They can't buy him off. They can't make him disappear without drawing more attention. So maybe they just hope that Geometric Unity will die on its own. Maybe they are behind the attacks on GU as something totally unserious. Maybe they have found a way to make other members of the physics community willing to look away. So, my question about this larger hypothetical operation to stop dangerous science and math is this: was it really just a massive kicking of the can down the road. Did they hoped it would buy us time? And if so, to what end? What are they waiting for? Can this game really be played forever? Maybe they think it can. Maybe they have an ongoing directive to keep suppressing this knowledge for as long as possible, and perhaps there's some secret core group whose job is to be the perpetual gatekeepers of potentially civilization-ending secrets. Maybe the Epstein connection to all of this was precisely because he was part of the operation to discover who was doing forbidden work and assess their progress. Maybe that's why, when he met Weinstein, he knew so much about GU. It's impossible to say with any certainty, but as a theory, it does have some real explanatory power. I find myself thinking of the fictional Brothers of the Cruciform Sword in Indiana Jones and the Last Crusade -- a small, ancient, secret society whose job it was to keep tabs on people looking for the Holy Grail and stop them, so it doesn't fall into the wrong hands. Are we following the same track here? Could this, as crazy as it sounds, be the missing connective tissue in this mystery?](https://image.24vids.com/tw-2031031356260524185/amplify_video_thumb/2031017929995612160/img/cpYRNKtLEZOG4Acm.jpg)
![I think I can finally report some success training a quite accurate IDM capable of recovering keystrokes from Minecraft gameplay, even in quite PvP-heavy situations. At this point the model does not only know what keys are pressed to the extent reasonably discernible, it also knows how fast it is moving in 3D space at all times, even when knockback is mixing with the self-move impulse. Now, recovering keystrokes from normal external capture footage is just about impossible. E.g. W/A/S/D does exactly nothing during partial tick frames and jumping mid-air is also equally useless, so asking the model to recover key down states is inherently unreasoanble. Mouse deltas are also completely arbitrary units, as game mouse sensitivity introduces an arbitrary scale factor into the equation. The only good option is to think carefully about your model-environment contract, and only record "logical actions", not raw keystrokes. So here's a few unfortunate lessons I had to learn in roughly this order. - Choose good units. (bad: mouse deltas, good: delta radians [yes, you will need game-internal state]) - Capture from inside the main game loop and read the game fbo to get consistent frame-action pairing. Doing post-mortem pairing is hopeless. - Carefully define when you think keystrokes actually have an effect. (jump only works on ground, when flying or in water etc.) More subtle: The key may already be down, but no tick has happened yet to actually use the value. Hence: ignore Seperate gamestate into "fast and slow-moving" components. E.g. movement is likely tick based, camera rotation is very likely updated every frame in essentially every game ever. - Think about your frame-action correspondance contract (How old is the frame in relation to the inputs you capture? Will double or tripple buffering affect you?) Think about the game loop timeline, where you are sampling, how old the data you are reading is, and where the ticks are happening around you. Language models used to simply not have a model-environment contract, but even now with the model "living" in a designated harness, the contract still boils down to formatting, and tool implementation intrinsics. While also important, it is still quite a bit more obvious because the violations are in some way shape or form reflected as text you can actually see. - ffmpeg dropping frames cummulatively screws the model the further you get into the sequence because your targets are now shifted. If you can't encode the video in real-time, too bad. - Sodium has a frames in flight system different from vanilla Minecraft, which will also offset your targets from your frames. (there goes that data...) - Models are succeptible to latency. If there is too big of a delay between action and on-screen reflection, your performance degrades. At this point I realize ~100hours of gameplay is essentially no longer usable as a dataset. You can train on this data, but all you'll get is a mushy mess. However, some good news: - Making the model predict physics gamestate scalars helps the model generalize. For instantaneous events like jump, it's unreasonable to ask the model emit a short burst of jump=true at exactly the right time, however if you also predict your current y-velocity, the model has supervision signal for the "latent" from which that onground jump becomes apparent. Recovering x/z motion is also somewhat easier than unmixing it into plausible keystrokes for inertia-heavy player controller logic. - Regressing physics gamestate scalars also seems to make your dataset "bigger". While pure keystroke classification will overfit quickly, predicting exact physics gamestate scalars forces the model to generalize more and you can tolerate far more epochs before validation loss starts to stall out. This is the only reason why it was bearable to dump 100h+ of dataset hours and replace it with ~3 hours of gameplay after the 4th revision of the file format (yeah...) and somehow still have better performance. Now, you might be asking, "isn't this brittle?" and the answer is yesn't. Frame-action correspondance matters for training, but not so much during inference. So as long as you are sampling in roughly the same interval as your training data, you aren't violating any hard contract per-se. Somewhere around the frames ticks are happening, and during training you capture various tick-capture offset relations per random chance, so nothing is too obviously wrong here. HOWEVER, you will get screwed by gui scale, shaders, resource packs, "shit that recording is 1920x1040 because somebody doesn't know fullscreen exists" and other unfortunate edge cases of reality. But I suppose this is the role of dataset size. If all those "contract violations" that a youtube video has compared to the training data are addressed, I think this is a way to turn Youtube into a labeled dataset. I could never shake the feeling that VPT is a sound idea in practice, while never having been properly executed, and I think one reason why it hasn't is because that label boostrapping part is just a pain in the butt to get right. Now, what the player is doing is of course not the only label you can extract from video, but it has to be one of the targets predicted during pretraining to "align" the pretraining objective. Some notes on the video here, the colored dots on the analog visualizer are the ground truth, while the gray dot is the model prediction. Green means correct prediction, red means incorrect prediction at that frame. Model P(key) reports how wrong the prediction is from green (0.0) to red (1.0). You will also notice that during periods of rapid slow down, left and right actions become close to irrecoverable, because there is just that little motion. And some jump actions are not predicted correctly because I got the detection condition for jump events wrong... (duh) LMB/RMB for other than sustained events (like item-consume and block break) also seem to be hopelessly irrecoverable for now. Swing was supposed to do the same thing as motion y did for jump, but its too well behaved as an increasing counter. Maybe partial-tick interpolated values work better (v5 file format then... ugh..)](https://image.24vids.com/tw-2042731804780826849/amplify_video_thumb/2042721691844120576/img/yugYQfuXXQmJpdVp.jpg)
![*** Breaking: Tucker Carlson Taking on 9/11 Truth with Senator Ron Johnson "Molten Metal....Thermite....Building 7....Controlled Demolition....4-year University of Alaska Study....Barry Jennings....156 Witnesses of Explosions....Corruption....Lies....Cover-up..." We’re making significant progress now! That cannot be denied. We have a sitting Senator drawing attention (as the Chairman of Homeland Security, the Chair of the Senate Permanent Subcommittee on Investigations, and the Chairman of the Senate Oversight Committee) in today’s major interview with Tucker Carlson, to: 156 First Responder witnesses of explosions — many before the towers descended — omitted from the NIST report “…Molten Steel in the Twin Towers” — (seen pouring out of the South Tower just minutes before its collapse) “The ash [WTC Dust] contains residue of both burnt and unburnt [nano]thermite.” Building 7 as a clear Controlled Demolition — “the only way that happens is if you blow up all their supports at the same time.” “….iron spheres with extreme temperatures.” “It wouldn’t come down so symmetrically. You wouldn’t have a free fall.” Major witnesses of explosions such as “Barry Jennings” “It was literally predicted to come down.” “Alaskan structural engineering professor [Hulsey] does a four-year study on it that pretty well debunks the NIST analysis.” Truth of 9/11 was covered up by the 9/11 Commission… “set up to fail” “Nothing's adding up. I mean…structurally, just basic engineering, basic physics, it doesn't make sense.” “They are very effective at marginalizing and discrediting anybody who's asked the questions.” “The firefighters want to know [the truth about 9/11].” “You talk to guys like Richard Gage, the [former] head of Architects [& Engineers] for 9/11 Truth.” I’m awfully glad that the senator asked me to sit down with him on a Zoom call and review all of the above evidence with him the day before this major interview! And Tucker, for his part, admitted yet again, “Because I was part of the coverup and I feel guilty about it, that's why, and I'm trying to atone for my previous sins.” Learn more and Support RichardGage911 [dot] org](https://image.24vids.com/tw-1927949996122591402/amplify_video_thumb/1927918449751379971/img/HoEPWc9wbnSyXud4.jpg)
![BREAKING NEWS🚨🚨🚨: Michael Proctor will be forced to appear for his deposition with Alan Jackson and Karen Read, however Judge Gildea will delay his final ruling until 4pm ET today in order to supplement Proctor's "insufficient" arguments to the court regarding a protective order to delay that deposition. Another hearing will happen today at 4:15PM et TODAY (Monday, June 8th, 2026, via Zoom). Proctor's deposition, for now, is stayed for 24 hours. "For this case to proceed efficiently, and for our system of justice to work, certain basic things need to happen," Judge Gildea began during his ruling. "A subpoena needs to mean something, a lawyer's agreement needs to have some meaning, lawyers need to communicate with their clients, clients need to keep their attorneys informed, and when given an opportunity to provide support for a party's request, that party needs to avail themselves of the opportunity," Judge continued (scolding Proctor's counsel for failing to provide a sufficient record to justify a delay in Proctor's deposition, despite an order by the court on Friday for such information to be filed). Based on the record before me, there is insufficient information to support a delay in the deposition going forward, however, given the claims made in the filings over the weekend, I will delay making a final decision on the motion until 4PM today, to give Mr. Proctor's counsel to supplement the record before me," Judge Gildea concluded. The Michael Proctor and Sean Goode "Proctorgate" files have caused serious damage, and Karen Read stands tall. Read more about those shocking, and abhorrent, messages here (also discussed by Judge Gildea during today's hearing) wherein Proctor (a former State Trooper who investigated Karen Read for the death of John O'Keefe) and former Canton cop Sean Goode talked about everything from how Anne Frank was a "liar" to how Proctor felt Goode should "let them die" after Proctor noticed a car accident victim was, in Proctor's words, a "ni**er". Proctor also discussed "planting cocaine" on someone and, shockingly, said "America sucks" because the country stopped the Third Reich during World War II (while, at the same time, saying "Jews in Sharon" should be "put in a concentration camp") at this link: To this day, the death of John O'Keefe remains unsolved. Proctor is just one of multiple witnesses involving in the twice-failed prosecution of Karen for John's death who Read's lawyers have told Judge Gildea are attempting to avoid or delay depositions (including Colin Albert, Jennifer McCabe, Brian Higgins, Matt McCabe, Nicole Albert and Brian Albert). Last week, on Wednesday, I reported: In a shocking hearing before Judge Mark Gildea on Wednesday, Karen Read's lawyers announced that they have been forced to file a motion to compel a deposition of Colin Albert. Further, Karen's lawyers will be filing motions to compel depositions of Jennifer McCabe, Brian Higgins, Matt McCabe, Nicole Albert and Brian Albert, as those parties are seeking to delay their depositions until late July, or early August, of 2026. "Colin Albert refused to appear or produce documents...his attorney said he is unavailable and then, suddenly, said Colin would only be available for seven more days and, then, [Colin] would be gone for the entire discovery period," Karen's Lawyer, Aaron Rosenberg, told Judge Gildea during Wednesday's hearing. In that regard, here is the full recording of today's John O'Keefe and Karen Read hearing in Plymouth Superior Court. This case is one of three ongoing civil lawsuits (one in federal court and two in Plymouth District Court in front of Judge Gildea) that stem from the, yet-unsolved, murder of Boston Police Officer John O'Keefe. To this day, it is unclear if anyone has found the Mystery Man who is suspected of being at the center of potential upcoming investigations by state and federal authorities as to John's death. Read more about that Mystery Man here - Judge Gildea began Wednesday's hearing by asking counsel for all parties how many depositions have been conducted in the case. "I am hoping at least four have been taken," said Judge Gildea. The O'Keefe lawyer told Judge Gildea that only three depositions have taken place. Judge then asked the O'Keefe lawyer what depositions need to still take place for the plaintiffs. "There are 12," the O'Keefe lawyer responded, including Paul Markowski. The O'Keefe lawyers also need to change Karen's deposition until June 24th. The O'Keefe lawyers changed Karen's deposition from July to June 19th, but multiple lawyer's will not be working on that day as that day is a federal holiday. Judge was not pleased that so few depositions have taken place, but Judge understood that June 19th is a federal holiday and, in turn, Karen's deposition will be held on June 24th, 2026. Karen's lawyers had proposed four other dates and Mr. Diller (the O'Keefe lawyer) picked a federal holiday, instead. "It is beyond belief to me the number of experienced lawyers involved in this case, the length the case has been pending...for however long it has been pending for all parties...that there are so few depositions that have taken place...I don't need to hear all the details, and the nitpicking, what I need to hear is, going forward, how you all are going to get the discovery done," Judge Gildea said to all lawyers. Things, at that point, got quite intense, as Karen's lawyers then noted to Judge Gildea that they have six depositions currently scheduled for the following six witnesses; Brian and Nicole Albert, Matthew McCabe, Allie McCabe, Jennifer McCabe and Brian Higgins. However, the law firm for all of those people got back to Karen's lawyers last night, on June 2nd, 2026, at 9:45PM and said NONE of those witnesses could appear on the dates scheduled for their depositions and, for some reason, none of those witnesses are able to appear for depositions until late July of 2026 at the earliest. Stunningly, no reason was given for why those witnesses refused to appear for their depositions on the dates originally proposed by Karen's lawyers, Attorney Rosenberg told Judge Gildea. Those notices of depositions went out on May 5th, 2026, said Karen's lawyers, when Judge Gildea asked about that date (indicating Judge Gildea, already displeased with discovery and deposition delays in the years-old case, was not pleased to hear about witnesses, on their own accord, attempting to delay depositions by two more months). Karen's lawyers, picking up on Judge Gildea's frustration, then noted that they intend to file motions to compel those witnesses to appear for depositions as a result of the proposed delays. Also, during the hearing, it was announced that former Massachusetts State Trooper Michael Proctor will be deposed by Karen Read's lawyers on Monday, June 8th, 2026, per a statement from Karen's lawyers in court on Wednesday. The O'Keefe lawyer, for their part, noted they intend to depose Aidan Kearney, Karen Read, Kaitlin Boudreau Read (Nathan Read's wife), Janet Read, Nathan Read, William Read, multiple witnesses from the bars (C.F. McCarthy's and The Waterfall), and an outstanding deposition that Karen's team needed to move last week. Judge Gildea then asked what, if anything, he could do to help the parties (other than moving quickly on motions before the court when submitted). Both parties said they will involve the court, in motion practice, as needed going forward. Judge then wanted to know what motions have been sent out under Rule 9(a). It was at that point that Karen's lawyers announced that she will be filing a motion to compel production of documents and the deposition of Colin Albert. "Colin Albert refused to appear or produce documents...his attorney said he is unavailable and then, suddenly, said Colin would only be available for seven more days and then [Colin] would be gone for the entire discovery period," Karen's Lawyer, Aaron Rosenberg, told Judge Mark Gildea during the hearing. Earlier, Karen also announced that she intends to file similar motions for Brian Higgins, Matt McCabe, Jen McCabe, Nicole Albert and Brian Albert, as those witnesses are attempting to avoid their depositions until late July or early August of 2026. The O'Keefe lawyer, for their part, told Judge Gildea they intend to serve a motion to compel regarding Aidan Kearney as a keeper of record (and, for some reason, Diller wants documents from Aidan's ongoing case in Norfolk Superior Court). "File a motion and we'll put it on for an expedited hearing," Judge Gildea said regarding that issue, and any other motions to compel the parties may want to file (including the motion regarding Colin Albert that has already been served, and should be in the docket in the next few weeks). When a motion is served under Rule 9(a), it is first given to the opposing party who then responds. Then, the moving party (the "movant") is able to file a sur-reply. Only at that point does the entire motion to compel get compiled as a so-called "Rule 9 (a) package" and, in turn, appear on the docket (with all filings related to a given motion showing up at the same time). It is unclear when the motions to compel the depositions of Brian Higgins, Jen McCabe, Matt McCabe, Nicole Albert and Brian Albert will be served and, eventually, docketed with the court. Penalties for third-parties who avoid discovery, or document production requests, can include court sanctions, compelled deposition dates, and adverse evidentiary rulings if a motion to compel is granted, yet further delays then occur. Judge Gildea told all of the lawyers present that he expects to dive deep into the issue of who was issued deposition notices in this case, who avoided those depositions, who delayed depositions and, in turn, what remedies the court will impose. "We served it Monday," the O'Keefe lawyer then said, as the hearing ended (indicating a number of motions to compel discovery, or depositions, may be heard over the coming weeks under the sweltering heat of Judge Gildea's virtual courtroom stare). More as I know it.](https://image.24vids.com/tw-2063984352610910306/amplify_video_thumb/2063980779294572544/img/BOF8su80S5P85pRH.jpg)