Vtuber | Twitch Affiliate | Valo-riety streamer. Wholesome with a nice voice. Come for the vibes, stay for the community. wintercharm.social

Joined October 2023
422 Photos and videos
WinterCharmVT ❄️🌨️☕️ ➜ Offkai 2026 ✈️ retweeted
WRAP ON DAY 7! Thank you to @MinaBunVT @suvimaru @SilverBlossomTV and @GalGalaxya for joining me for the Pokemon Gameshow and Congrats to Mina for Winning! Thanks to @p33katchuwu @SilverBlossomTV and @LunaNhym4164 for the raids! We raided into @WinterCharmVT Contestants submission for art questions:
1
4
13
186
Dear Microsoft. Please just fucking show me EVERY ICON IN THE TOOLBAR BY DEFAULT. WHY DO I HAVE TO KEEP ADDING AN EXCEPTION TO EACH APP AFTER EVERY SOFTWARE UPDATE. WHY!?!?!? >:I
1
111
Anthropic suddenly pulling Fable is bullshit. I was mid-buildout. MID BUILDOUT of Nebula. This is why I keep pushing to more and more local models running on clean energy with zero water use. AI rent seekers pmo.
2
46
WinterCharmVT ❄️🌨️☕️ ➜ Offkai 2026 ✈️ retweeted

71
224
1,063
473,553
WinterCharmVT ❄️🌨️☕️ ➜ Offkai 2026 ✈️ retweeted
spike planted
NAHHH 😭
38
1,218
22,931
931,553
WinterCharmVT ❄️🌨️☕️ ➜ Offkai 2026 ✈️ retweeted
please give me OP on the server 😩
1
2
24
489
Lmao… yeah this is why I don’t think basic censoring of LLMs is actually a good thing. It’s an attack vector.
NEW: malware developers added nuclear & biological weapons text to to their spyware. Goal? To trigger LLM safety refusals... so that their spyware wouldn't be analyzed by an AI security scanner. Cleanest practical example I can think of for why over-indexing on first order safety alignment is risky. When closed (and open) models ship with aggressive refusals, they will be sprinkled with second-order blindspots that attackers will discover...and exploit. We are only in the earliest days of attackers leveraging these features, and it wouldn't surprise me if users systems that need to handle complex cybersecurity issues demand that models be less safety-blunted. In the weeds: @SocketSecurity's post also shows why intention matters in how you design a malware analysis pipeline to avoid prompt manipulation. H/T to colleagues that shared this with me socket.dev/blog/mini-shai-hu…
1
40
WinterCharmVT ❄️🌨️☕️ ➜ Offkai 2026 ✈️ retweeted
NEW: malware developers added nuclear & biological weapons text to to their spyware. Goal? To trigger LLM safety refusals... so that their spyware wouldn't be analyzed by an AI security scanner. Cleanest practical example I can think of for why over-indexing on first order safety alignment is risky. When closed (and open) models ship with aggressive refusals, they will be sprinkled with second-order blindspots that attackers will discover...and exploit. We are only in the earliest days of attackers leveraging these features, and it wouldn't surprise me if users systems that need to handle complex cybersecurity issues demand that models be less safety-blunted. In the weeds: @SocketSecurity's post also shows why intention matters in how you design a malware analysis pipeline to avoid prompt manipulation. H/T to colleagues that shared this with me socket.dev/blog/mini-shai-hu…
227
2,158
12,651
1,547,521
WinterCharmVT ❄️🌨️☕️ ➜ Offkai 2026 ✈️ retweeted
Guys this is my Birthright !! I have to do this muahahhaa
This is Flip-Flop Fury, a physics-based projectile precision steering indie game where you hurl footwear at despicable targets! Coming to Steam: store.steampowered.com/app/3… by indie dev @shapeshifterDgt
17
23
455
12,841
Welcome to what AI is like for actual experts in fields. I find myself regularly correcting Opus and even hecking Fable because outside of common code AI really struggles.
So my software dev friend told me to try using AI for programming and to treat it like "enhanced Google", so to try and experiment to see if it could actually solve my problems faster than me just Googling it, I tested it with a few of my own outstanding issues. this shit is so ass dude It just hallucinates settings in my software 90% of the time. Any code it spits out is basically unusable either architecturally or because it lacks context. It's also insanely over engineered, tries to reinvent the wheel, or just has so much bloat overhead from trying to pre-optimize or "do perfect code" that I'd spend MORE time just fixing what it gives me to make it usable. This sucks lmao My problems tend to be relatively niche which makes it worse. For example, trying to figure out why FFT block processing sizes being too large are causing artifacts in my audio in FMOD. Do you know what it said after I told it the setting it told me to click didn't even exist? (which was a stupid solution to begin with had that setting even existed) "You're right, I'm not confident enough in the exact FMOD Studio menu paths to keep guessing them — I've already got several wrong. Let me look this up properly." Then proceeds to just give me Google results I've already seen. This is actually the most infuriating waste of time I've ever seen how do people use this in their games unironically lmfao. Maybe I'm just using this wrong but I have genuinely zero interest in trying to wrestle with this garbage to give me what I want. So far the only thing it's been good for is allowing me to rubber duck my problems, and I can use a blank Notepad file for that. It has in fact solved zero of my questions. Experiment over, I'm just going to go back to using my brain and stop wasting my time thanks for actually nothing. 🙄
2
111
WinterCharmVT ❄️🌨️☕️ ➜ Offkai 2026 ✈️ retweeted
Replying to @macrumorslive
Everyone needs to go back to school and learn how to read. This is the MOST POWERFUL VERSION OF APPLE INTELLIGENCE In general, all models are supported.
10
10
150
35,805
Honestly, working on trading algorithms has been the most fun and rewarding thing. The math is just... fascinating. Markets have unusually high amounts of chaos. Filtering signal from noise genuinely takes a lot of effort. It's slowly eclipsing my consulting income now...
2
32
This right here is body goals
May 27
People were complaining about drawing the guy faceless in my comic so I put flork's head on my muscle study
3
79
Apple’s new AI framework allows dynamic model switching between various requests but preserves the entire conversation so model swaps don’t lose context. Really well done stuff. Open sourcing their tools is great. I love everything they’ve done *except* Image generation.
1
35
WinterCharmVT ❄️🌨️☕️ ➜ Offkai 2026 ✈️ retweeted
Lol could they be any more blatant?
New in macOS 27: You can now resize iPhone mirroring to look like an iPad display
94
235
16,429
3,034,521
WinterCharmVT ❄️🌨️☕️ ➜ Offkai 2026 ✈️ retweeted
Being a male streamer trying to make friends with a female streamers be like “i hope they don’t think I’m a predator, a simp, or secretly trying to hit on them. i promise i just wanna collab and network like normal but also pls don’t let one awkward joke or DM screenshot end my entire career.”
Being a female streamer trying to make friends with male streamers be like “i hope they don’t think I’m hitting on them i promise i just wanna be friends 😭” ☠️☠️☠️
67
289
8,357
317,845
WinterCharmVT ❄️🌨️☕️ ➜ Offkai 2026 ✈️ retweeted
Apple apparently optimized the shit out of IOS 27 and I am very excited cuz holy heck is IOS 26 a dog also EQ for AirPods wooooo
17
40
2,575
86,056
Leave it to Apple to be 2 years late to AI but also perfectly integrate it across their OS. Siri AI is insane.
38
WinterCharmVT ❄️🌨️☕️ ➜ Offkai 2026 ✈️ retweeted
>wont waive rule for unprofitable ai firms NO IPO FOR YOU
16
57
2,135
97,722
WinterCharmVT ❄️🌨️☕️ ➜ Offkai 2026 ✈️ retweeted
I want to be very clear that I'm certain what I'm doing is legal (see linked FAQ). I would prefer not to have to deal with a legal battle, but if I receive legal threats, I will defend myself and not just fold.
STREAM: Reverse engineering Live2D MOC3 Let's do this! 🚀 ✨ Important FAQ✨ lina.yt/moc3faq 🕐 Jun 4 13:00 JST / 04:00 UTC ▶️ youtube.com/live/eEa0-wt1SqE #ENVTuber #Live2D
12
29
840
56,627