SelfReflect has launched and is on the App Store now! Get the link at SelfReflect.net! #vtuber #vtubing #motioncapture #3danimation
October 23, 2026 at 11:41 PM
SelfReflect has launched and is on the App Store now! Get the link at SelfReflect.net! #vtuber #vtubing #motioncapture #3danimation
Playing Silver Pines! Also trying out the selfreflect app with vnyan!
#Vtuber #Bunopolis
www.twitch.tv/canti_vt
#Vtuber #Bunopolis
www.twitch.tv/canti_vt
Canti_VT - Live on Twitch
Free breakfast and cable tv!? (Trying out the selfreflect app with vnyan) !bunopolis | Streaming silver pines.
www.twitch.tv
October 21, 2026 at 5:06 PM
Playing Silver Pines! Also trying out the selfreflect app with vnyan!
#Vtuber #Bunopolis
www.twitch.tv/canti_vt
#Vtuber #Bunopolis
www.twitch.tv/canti_vt
I WAS FINALLY ABLE TO USE SELFREFLECT
I can move in 3d full body! :3
kinda funky but honestly good for just an iphone track!
but might be great for when I reveal my 3d model until I can afford Slimevr 💛💛
I can move in 3d full body! :3
kinda funky but honestly good for just an iphone track!
but might be great for when I reveal my 3d model until I can afford Slimevr 💛💛
Arataka Reigen from Mob Psycho 100 in Hello Kitty Collage
ALT: Arataka Reigen from Mob Psycho 100 in Hello Kitty Collage
static.klipy.com
October 19, 2026 at 2:10 AM
I WAS FINALLY ABLE TO USE SELFREFLECT
I can move in 3d full body! :3
kinda funky but honestly good for just an iphone track!
but might be great for when I reveal my 3d model until I can afford Slimevr 💛💛
I can move in 3d full body! :3
kinda funky but honestly good for just an iphone track!
but might be great for when I reveal my 3d model until I can afford Slimevr 💛💛
I’ve been looking through old sketchbooks recently, which is something I don’t do too often due to the mutterings of ‘what was I thinking?’ it tends to inspire.
However; this double page spread is pretty cool 🖤
#sketchbook #artist #selfreflect #dragon #ink
However; this double page spread is pretty cool 🖤
#sketchbook #artist #selfreflect #dragon #ink
January 19, 2026 at 10:42 AM
I’ve been looking through old sketchbooks recently, which is something I don’t do too often due to the mutterings of ‘what was I thinking?’ it tends to inspire.
However; this double page spread is pretty cool 🖤
#sketchbook #artist #selfreflect #dragon #ink
However; this double page spread is pretty cool 🖤
#sketchbook #artist #selfreflect #dragon #ink
We just have to stop letting people get away with the "i didn't know :(((" excuse these days like we all can and have to do better without someone else having to point it out first but people simply do not want to selfreflect or even think before speaking
April 24, 2025 at 10:31 AM
We just have to stop letting people get away with the "i didn't know :(((" excuse these days like we all can and have to do better without someone else having to point it out first but people simply do not want to selfreflect or even think before speaking
Chat with me about uncertainty + LLM (agents) at #ICLR2026, in either of these 6 sessions:
1. Thu 10:30 AM - 1 PM, Pavilion 3 P3-#309, Memory LLM paper
2. Thu 3:15 PM – 5:45 PM, Pavilion 3 P3-#312, SelfReflect
3. Fri, 9:30 AM - 11:30 AM, Apple booth
1. Thu 10:30 AM - 1 PM, Pavilion 3 P3-#309, Memory LLM paper
2. Thu 3:15 PM – 5:45 PM, Pavilion 3 P3-#312, SelfReflect
3. Fri, 9:30 AM - 11:30 AM, Apple booth
4 ICLR papers 🥳 There’s an insightful story between them: If you sample LLMs multiple times, they are calibrated, even on higher levels [1], but they cannot talk about this uncertainty in a single prompt [2], so you have to help them out to gather information Bayes-optimally [3]
April 23, 2026 at 10:52 AM
Chat with me about uncertainty + LLM (agents) at #ICLR2026, in either of these 6 sessions:
1. Thu 10:30 AM - 1 PM, Pavilion 3 P3-#309, Memory LLM paper
2. Thu 3:15 PM – 5:45 PM, Pavilion 3 P3-#312, SelfReflect
3. Fri, 9:30 AM - 11:30 AM, Apple booth
1. Thu 10:30 AM - 1 PM, Pavilion 3 P3-#309, Memory LLM paper
2. Thu 3:15 PM – 5:45 PM, Pavilion 3 P3-#312, SelfReflect
3. Fri, 9:30 AM - 11:30 AM, Apple booth
Correct. SelfReflect was the only resubmission. But don't want to disencourage anyone -- a lot of this success is due to being in this environment with a great team with very principled scientists around. You should control for that variable when making comparisons 🤗
January 28, 2026 at 7:59 PM
Correct. SelfReflect was the only resubmission. But don't want to disencourage anyone -- a lot of this success is due to being in this environment with a great team with very principled scientists around. You should control for that variable when making comparisons 🤗
[1] Semantic Calibration openreview.net/forum?id=0sC...
[2] SelfReflect openreview.net/forum?id=hOE...
[3] Bayesian Experimental Design for LLMs openreview.net/forum?id=qyy...
[4] Pretraining memories openreview.net/forum?id=XOu...
[2] SelfReflect openreview.net/forum?id=hOE...
[3] Bayesian Experimental Design for LLMs openreview.net/forum?id=qyy...
[4] Pretraining memories openreview.net/forum?id=XOu...
January 27, 2026 at 12:49 PM
[1] Semantic Calibration openreview.net/forum?id=0sC...
[2] SelfReflect openreview.net/forum?id=hOE...
[3] Bayesian Experimental Design for LLMs openreview.net/forum?id=qyy...
[4] Pretraining memories openreview.net/forum?id=XOu...
[2] SelfReflect openreview.net/forum?id=hOE...
[3] Bayesian Experimental Design for LLMs openreview.net/forum?id=qyy...
[4] Pretraining memories openreview.net/forum?id=XOu...
Since its initial release, we didn’t stop cooking: First, we continued validating whether the scores that the SelfReflect benchmarks assigns are robust signals of quality. Across more LLMs and datasets, it works. I have more confidence in the benchmark than ever.
October 1, 2025 at 9:53 AM
Since its initial release, we didn’t stop cooking: First, we continued validating whether the scores that the SelfReflect benchmarks assigns are robust signals of quality. Across more LLMs and datasets, it works. I have more confidence in the benchmark than ever.
Many treat uncertainty = a number. At Apple, we're rethinking this: LLMs should output strings that reveal all information of their internal distributions. We find that Reasoning, SFT, CoT can't do it - yet. To get there, we introduce the SelfReflect benchmark.
arxiv.org/pdf/2505.20295
arxiv.org/pdf/2505.20295
October 1, 2025 at 9:53 AM
Many treat uncertainty = a number. At Apple, we're rethinking this: LLMs should output strings that reveal all information of their internal distributions. We find that Reasoning, SFT, CoT can't do it - yet. To get there, we introduce the SelfReflect benchmark.
arxiv.org/pdf/2505.20295
arxiv.org/pdf/2505.20295
If the LLM could do that, it would give users absolute honesty about subjective uncertainties. It could even use it itself to gain clarity about which follow-up questions to ask. We invite you to develop such strategies. Here’s a repo to get started: github.com/apple/ml-sel... 🧵9/9
GitHub - apple/ml-selfreflect
Contribute to apple/ml-selfreflect development by creating an account on GitHub.
github.com
July 3, 2025 at 9:08 AM
If the LLM could do that, it would give users absolute honesty about subjective uncertainties. It could even use it itself to gain clarity about which follow-up questions to ask. We invite you to develop such strategies. Here’s a repo to get started: github.com/apple/ml-sel... 🧵9/9
We put SelfReflect to the test (plus a whole landscape of other metrics we’ve explored). Our final metric is able to distinguish even almost-good from good summaries and pick up implicit distributional statements. We find it aligns with human judgements. So let’s use it: 🧵6/9
July 3, 2025 at 9:08 AM
We put SelfReflect to the test (plus a whole landscape of other metrics we’ve explored). Our final metric is able to distinguish even almost-good from good summaries and pick up implicit distributional statements. We find it aligns with human judgements. So let’s use it: 🧵6/9
RQ 1: How can we measure if a string does that? The idea of our SelfReflect metric is that the summary string should imply the same follow-up responses as the original distribution; theoretically speaking it should be a predictively sufficient statistic of the distribution. 🧵5/9
July 3, 2025 at 9:08 AM
RQ 1: How can we measure if a string does that? The idea of our SelfReflect metric is that the summary string should imply the same follow-up responses as the original distribution; theoretically speaking it should be a predictively sufficient statistic of the distribution. 🧵5/9
SelfReflect can now hook up to Unreal LiveLink for easier integration in your existing Unreal workflow. No extra plugins either. It pairs great with Metahuman Animator!
June 12, 2025 at 12:11 AM
SelfReflect can now hook up to Unreal LiveLink for easier integration in your existing Unreal workflow. No extra plugins either. It pairs great with Metahuman Animator!
SelfReflect works on multiple devices. Here I'm using one on the helmet for face, and one for body. Here's some unedited footage of it driving a Metahuman in UE5.
Need to add back in smoothing on hands at a distance, but it's coming along. I plan to support fusion of multiple body angles
Need to add back in smoothing on hands at a distance, but it's coming along. I plan to support fusion of multiple body angles
June 4, 2025 at 6:23 AM
SelfReflect works on multiple devices. Here I'm using one on the helmet for face, and one for body. Here's some unedited footage of it driving a Metahuman in UE5.
Need to add back in smoothing on hands at a distance, but it's coming along. I plan to support fusion of multiple body angles
Need to add back in smoothing on hands at a distance, but it's coming along. I plan to support fusion of multiple body angles
SelfReflect Early Access: New features! Bugs! More!
$20 gets:
- Unlimited license to the full release.
- Access to early builds as I work on them (Not everything works perfectly in every build yet)
- Access to a private discord for testers for bugs & features
ko-fi.com/selfreflect
$20 gets:
- Unlimited license to the full release.
- Access to early builds as I work on them (Not everything works perfectly in every build yet)
- Access to a private discord for testers for bugs & features
ko-fi.com/selfreflect
May 30, 2025 at 5:37 PM
SelfReflect Early Access: New features! Bugs! More!
$20 gets:
- Unlimited license to the full release.
- Access to early builds as I work on them (Not everything works perfectly in every build yet)
- Access to a private discord for testers for bugs & features
ko-fi.com/selfreflect
$20 gets:
- Unlimited license to the full release.
- Access to early builds as I work on them (Not everything works perfectly in every build yet)
- Access to a private discord for testers for bugs & features
ko-fi.com/selfreflect
More testing SelfReflect (my all-in-one VTuber mocap app). This time Shadow is sharing his controversial thoughts on Sonic 3 (the game), but he forgot to record his audio.
May 21, 2025 at 4:45 AM
More testing SelfReflect (my all-in-one VTuber mocap app). This time Shadow is sharing his controversial thoughts on Sonic 3 (the game), but he forgot to record his audio.
Awww yea! More SelfReflect with Unreal Engine 5, this time with expressions!
May 19, 2025 at 9:23 PM
Awww yea! More SelfReflect with Unreal Engine 5, this time with expressions!
Using SelfReflect with more Unreal characters. Just need to get blendshapes paired up.
May 17, 2025 at 1:48 PM
Using SelfReflect with more Unreal characters. Just need to get blendshapes paired up.
Any companies at AWE next month wanna showcase SelfReflect? It looks awesome on a big screen. And there's no calibration for anyone just walking up.
It's a real showstopper.
It's a real showstopper.
May 14, 2025 at 5:36 PM
Any companies at AWE next month wanna showcase SelfReflect? It looks awesome on a big screen. And there's no calibration for anyone just walking up.
It's a real showstopper.
It's a real showstopper.
Testing SelfReflect in a real world scenario. Going from Warudo into OBS and paired with a Genesis emulator. It's really so incredibly fast and easy to setup.
May 14, 2025 at 7:34 AM
Testing SelfReflect in a real world scenario. Going from Warudo into OBS and paired with a Genesis emulator. It's really so incredibly fast and easy to setup.
Testing the new SelfReflect beta with Unreal Engine 5 tonight. The increased fps and wider fov help SO much. This is using upper body & seated mode.
(The thumbs are a problem with my blueprint fyi)
(The thumbs are a problem with my blueprint fyi)
May 14, 2025 at 5:21 AM
Testing the new SelfReflect beta with Unreal Engine 5 tonight. The increased fps and wider fov help SO much. This is using upper body & seated mode.
(The thumbs are a problem with my blueprint fyi)
(The thumbs are a problem with my blueprint fyi)
Slight hiccup for the next SelfReflect beta. The UI was getting messy. I was reinventing too much & since the UI was in the lens, it would reset when changing low level options.
So I buckled down and committed to rebuilding the UI in SwiftUI. Was a tough call, but the right one.
So I buckled down and committed to rebuilding the UI in SwiftUI. Was a tough call, but the right one.
May 8, 2025 at 9:00 AM
Slight hiccup for the next SelfReflect beta. The UI was getting messy. I was reinventing too much & since the UI was in the lens, it would reset when changing low level options.
So I buckled down and committed to rebuilding the UI in SwiftUI. Was a tough call, but the right one.
So I buckled down and committed to rebuilding the UI in SwiftUI. Was a tough call, but the right one.
Does your VTuber like to play guitar? Check out how well SelfReflect now handles this.
May 6, 2025 at 8:27 PM
Does your VTuber like to play guitar? Check out how well SelfReflect now handles this.
Side by side of SelfReflect on an iPhone 13 Pro vs an iPhone XR.
Not too bad for a 6yo phone eh?
Not too bad for a 6yo phone eh?
May 6, 2025 at 5:51 PM
Side by side of SelfReflect on an iPhone 13 Pro vs an iPhone XR.
Not too bad for a 6yo phone eh?
Not too bad for a 6yo phone eh?