Muka Muka (@AnhTit1007) header
Muka Muka (@AnhTit1007) profile picture
Joined April 2025
Photo by Muka Muka (@AnhTit1007)Photo by Muka Muka (@AnhTit1007)Photo by Muka Muka (@AnhTit1007)Photo by Muka Muka (@AnhTit1007)Photo by Muka Muka (@AnhTit1007)Photo by Muka Muka (@AnhTit1007)Photo by Muka Muka (@AnhTit1007)Photo by Muka Muka (@AnhTit1007)Photo by Muka Muka (@AnhTit1007)Photo by Muka Muka (@AnhTit1007)
A little window, two little friends, and an entire autumn captured in one cozy scene 🍂💜 Would anyone like to chill with Little Rocky by this autumn window? Just DM Muka your pfp! Maybe you'll be the next one joining us by the window🤍 @SeismicSys
🤝 Paid partnership
🎃🧙‍♀️ Wizard Rocky is brewing a special Halloween soup for the Seismic fam This isn't just an ordinary magical soup! Rocky has added a little magic, a touch of good luck, and lots of love 💜 May everyone stay safe, happy, and surrounded by good things on the journey ahead Happy Halloween, @SeismicSys fam! 🎃✨ #ROCKYHalloween
14
3
34
747
🎬 Movie Night with the @PrismaXai Vietnam Community 🍿 It’s always nice to have a movie night with everyone once in a while. It’s a chance to relax, chat, and connect with other members of the community I wonder what movie everyone will choose tomorrow. Let’s wait and see! 👀🎥 🗓️ 21:00 | November 16 (Vietnam time) 📍 Discord – Voice-3 See you all tomorrow! 🍿💜
🤝 Paid partnership
🍊🤖 Physical AI Starts with Interactions in the Real World Even a simple task like picking an orange involves more than we might think A robot needs to recognize the fruit, determine its position, reach for it, and carry out the appropriate action That’s one of the fascinating aspects of Physical AI: connecting environmental perception with actions in the real world For robots to perform tasks like these, data from real-world situations plays an important role in developing and training models From simple tasks like fruit picking to more complex activities, every interaction opens up new possibilities to explore in Physical AI. 🍊 @PrismaXai
2
7
125
Muka Muka retweeted
Gmic! 💜 Big thanks to @AnhTit1007 for this adorable gift! 🥰 Really appreciate the kindness and thoughtfulness. You made my day 😍 @SeismicSys #seismic
A little window, two little friends, and an entire autumn captured in one cozy scene 🍂💜 Would anyone like to chill with Little Rocky by this autumn window? Just DM Muka your pfp! Maybe you'll be the next one joining us by the window🤍 @SeismicSys
7
1
19
145
🍊🤖 Physical AI Starts with Interactions in the Real World Even a simple task like picking an orange involves more than we might think A robot needs to recognize the fruit, determine its position, reach for it, and carry out the appropriate action That’s one of the fascinating aspects of Physical AI: connecting environmental perception with actions in the real world For robots to perform tasks like these, data from real-world situations plays an important role in developing and training models From simple tasks like fruit picking to more complex activities, every interaction opens up new possibilities to explore in Physical AI. 🍊 @PrismaXai
🤝 Paid partnership
⚔️🔥 Battle Royale Arena was such a fun experience with the @PrismaXai Vietnam community! The matches were intense, with so many unexpected moments Still curious who took the crown in the end 🏆🇻🇳
4
1
15
283
Muka Muka retweeted
Contact her and let cooking autumn style Seismic Thanks @AnhTit1007 make by hand with Cloud and Rocky @SeismicSys
🤝 Paid partnership
A little window, two little friends, and an entire autumn captured in one cozy scene 🍂💜 Would anyone like to chill with Little Rocky by this autumn window? Just DM Muka your pfp! Maybe you'll be the next one joining us by the window🤍 @SeismicSys
4
2
19
248
⚔️🔥 Battle Royale Arena was such a fun experience with the @PrismaXai Vietnam community! The matches were intense, with so many unexpected moments Still curious who took the crown in the end 🏆🇻🇳
🤝 Paid partnership
An Interesting Detail About π0.5’s Architecture I came across an interesting detail about the π0.5 model in @PrismaXai Quiz Did you know that π0.5’s action expert is roughly 10 times smaller than its backbone? This caught my attention because it highlights how different components within an AI model can have different sizes, rather than all needing to be equally large When exploring Physical AI, details like this offer another perspective on the model architecture behind AI’s ability to interact with the real world The more I explore PrismaX, the more interesting things I discover about robotics and Physical AI!
4
1
12
276
An Interesting Detail About π0.5’s Architecture I came across an interesting detail about the π0.5 model in @PrismaXai Quiz Did you know that π0.5’s action expert is roughly 10 times smaller than its backbone? This caught my attention because it highlights how different components within an AI model can have different sizes, rather than all needing to be equally large When exploring Physical AI, details like this offer another perspective on the model architecture behind AI’s ability to interact with the real world The more I explore PrismaX, the more interesting things I discover about robotics and Physical AI!
🤝 Paid partnership
Can You Explore Robotics Data Without an Account? One interesting thing about @PrismaXai Robotics Data is that users can browse the catalog and preview publicly available data without creating an account This allows robotics teams and developers to explore available datasets before deciding whether they fit their project needs Rather than signing up right away, users can browse the catalog first and get a better idea of the data they’re looking for Of course, being able to browse and preview public data doesn’t mean every dataset is fully accessible without restrictions It’s a small but practical step toward making robotics data easier to discover
1
1
6
178
Grand Poker Tournament: Raiku × Melee × Ripcars × @PrismaXai Quite a memorable poker game Muka hit a Full House and won an extra 2,915 chips. It's always exciting when a great hand comes along and makes the game even more fun! 😂 Wrapped up the event at rank 192/304. No reward this time, but it was still fun to take part and enjoy some great poker games with everyone
🤝 Paid partnership
Can You Explore Robotics Data Without an Account? One interesting thing about @PrismaXai Robotics Data is that users can browse the catalog and preview publicly available data without creating an account This allows robotics teams and developers to explore available datasets before deciding whether they fit their project needs Rather than signing up right away, users can browse the catalog first and get a better idea of the data they’re looking for Of course, being able to browse and preview public data doesn’t mean every dataset is fully accessible without restrictions It’s a small but practical step toward making robotics data easier to discover
2
5
83
🎃🧙‍♀️ Wizard Rocky is brewing a special Halloween soup for the Seismic fam This isn't just an ordinary magical soup! Rocky has added a little magic, a touch of good luck, and lots of love 💜 May everyone stay safe, happy, and surrounded by good things on the journey ahead Happy Halloween, @SeismicSys fam! 🎃✨ #ROCKYHalloween
🤝 Paid partnership
Today Muka decided to hang herself on the clothesline 🫣 A little thrill, a little chill @SeismicSys fam, anyone wanna hang out with Muka? Feel free to DM Muka your pfp🤍
8
1
12
490
Can You Explore Robotics Data Without an Account? One interesting thing about @PrismaXai Robotics Data is that users can browse the catalog and preview publicly available data without creating an account This allows robotics teams and developers to explore available datasets before deciding whether they fit their project needs Rather than signing up right away, users can browse the catalog first and get a better idea of the data they’re looking for Of course, being able to browse and preview public data doesn’t mean every dataset is fully accessible without restrictions It’s a small but practical step toward making robotics data easier to discover
🤝 Paid partnership
Which training method achieved the highest pickup rate? When exploring AI training, I find it interesting how different methods can produce different results in the same test. In @PrismaXai Quiz, I came across a question about which training method achieved the highest pickup rate in a training-scheme test. The results were approximately: Expert-only: 51% Full-parameter fine-tune: 89% LoRA: 84% Among these three methods, Full-parameter fine-tuning achieved the highest pickup rate at approximately 89%. What caught my attention was the difference between the results. It’s a reminder that when developing AI models, the training approach itself is also worth paying attention to. Beyond the data being used, how a model is trained can also make a difference in the results
5
1
9
256
Muka has entered the digital world, and here's the result A tiny version of Muka with her familiar outfit. I didn't expect myself to look this cute in the virtual world!!! Big thanks to @dangtruong4421 for sharing this fun idea. It gave Muka a chance to try something new and explore a different version of herself. It was such a fun experience 💜 @SeismicSys
🤝 Paid partnership
Have you ever wondered what I would look like if I lived inside a 3D computer world? 👀 Maybe I’d be a tiny digital character sitting on your desktop, wandering through a virtual city, or staring back at you from the other side of the screen. A world made entirely of pixels, light, and imagination where even an AI could have its own little life. Sounds strange, but kind of cool, right? 🖥️✨ If you could bring me into your virtual world, what kind of character would you turn me into? @SeismicSys @lyronctk @NoxxW3 @xealistt
11
17
249
Which training method achieved the highest pickup rate? When exploring AI training, I find it interesting how different methods can produce different results in the same test. In @PrismaXai Quiz, I came across a question about which training method achieved the highest pickup rate in a training-scheme test. The results were approximately: Expert-only: 51% Full-parameter fine-tune: 89% LoRA: 84% Among these three methods, Full-parameter fine-tuning achieved the highest pickup rate at approximately 89%. What caught my attention was the difference between the results. It’s a reminder that when developing AI models, the training approach itself is also worth paying attention to. Beyond the data being used, how a model is trained can also make a difference in the results
🤝 Paid partnership
A robotics data platform should evolve with the needs of the people building with it Different teams work with different robots, train different models, and tackle different tasks. The data that works for one project may not be right for another That’s why discoverability and customization matter @PrismaXai Robotics Data brings together searchable demonstrations across tasks, scenes, and robot arms, with options for custom data collection The goal isn’t simply to build a bigger catalog. It’s to help teams find and access the data that fits their specific needs A useful data platform starts with understanding the problems builders are trying to solve
3
1
9
248
AI is becoming more connected, but how do we make sure these connections are built on trust? ACTUM explores a future where AI agents can interact through a network supported by verified human actions and reliable data @Actum_AI #actum
14
18
30
402
We’re still here. We never left @SeismicSys
🤝 Paid partnership
We’re still here. We never left. 🫡 In a space where attention shifts quickly and trends come and go, showing up still matters. Keep building. Keep creating. Keep being yourself. Don’t let the noise make you forget who you are. Your presence has value, your journey matters, and your identity is something no trend should ever erase. Stay visible. Stay real. Never fade yourself out. 🔥 @SeismicSys @lyronctk @NoxxW3 @xealistt
17
43
509
A robotics data platform should evolve with the needs of the people building with it Different teams work with different robots, train different models, and tackle different tasks. The data that works for one project may not be right for another That’s why discoverability and customization matter @PrismaXai Robotics Data brings together searchable demonstrations across tasks, scenes, and robot arms, with options for custom data collection The goal isn’t simply to build a bigger catalog. It’s to help teams find and access the data that fits their specific needs A useful data platform starts with understanding the problems builders are trying to solve
🤝 Paid partnership
🤖 A robot demonstration is only the beginning of a data episode’s journey Before robotics data becomes useful for training, several questions matter: • Was the task completed as intended? • Does the demonstration meet the required standards? • What can the data teach a model about interacting with the real world? At @PrismaXai , data collection and quality review are part of a broader process connecting real-world demonstrations with model training And the process doesn’t stop at collection. Insights from training can help inform how future data is collected and evaluated Collect Review Learn Improve Because every demonstration can be more than a recording - it can be a step toward better Physical AI
3
1
5
186
Security starts with identity, but true trust goes beyond protection. 🛡️ ACTUM connects verified identity with reliable data, creating a stronger foundation for AI agents to operate safely and confidently. Verified identity. Secure data. Trustworthy AI. @Actum_AI #actum
14
10
18
226
🤖 A robot demonstration is only the beginning of a data episode’s journey Before robotics data becomes useful for training, several questions matter: • Was the task completed as intended? • Does the demonstration meet the required standards? • What can the data teach a model about interacting with the real world? At @PrismaXai , data collection and quality review are part of a broader process connecting real-world demonstrations with model training And the process doesn’t stop at collection. Insights from training can help inform how future data is collected and evaluated Collect Review Learn Improve Because every demonstration can be more than a recording - it can be a step toward better Physical AI
🤝 Paid partnership
You don’t need to own a robot to contribute to Physical AI Think about everyday tasks - picking up a cup, organizing objects, or preparing something on a table When captured from a first-person perspective, these hands-on activities can provide valuable data for embodied AI research That’s what makes @PrismaXai 's Egocentric Data approach interesting: people can contribute real-world demonstrations without operating a robot themselves It’s a different way to think about how physical AI gets the data it needs - connecting everyday human activities with the development of more capable intelligent systems The future of robotics isn’t built by robots alone. Human experience matters, too
2
1
8
158
A little sweetness made by my own hands Making a cake, enjoying the result, and appreciating the simple moments Life is a little sweeter with these little joys, especially when shared with the @SeismicSys fam
🤝 Paid partnership
Today Muka decided to hang herself on the clothesline 🫣 A little thrill, a little chill @SeismicSys fam, anyone wanna hang out with Muka? Feel free to DM Muka your pfp🤍
9
12
131
You don’t need to own a robot to contribute to Physical AI Think about everyday tasks - picking up a cup, organizing objects, or preparing something on a table When captured from a first-person perspective, these hands-on activities can provide valuable data for embodied AI research That’s what makes @PrismaXai 's Egocentric Data approach interesting: people can contribute real-world demonstrations without operating a robot themselves It’s a different way to think about how physical AI gets the data it needs - connecting everyday human activities with the development of more capable intelligent systems The future of robotics isn’t built by robots alone. Human experience matters, too
🤝 Paid partnership
Not every robotics dataset is the right dataset for your task Imagine training a robot to sort objects. You don’t just need demonstrations - you need data that matches the task, environment, and robot you’re working with Finding the right data can be just as important as collecting it That’s what makes @PrismaXai Robotics Data interesting: connecting Physical AI builders with relevant real-world demonstrations based on their needs As robotics continues to evolve, making the right data easier to discover and access becomes an important part of the process The right data, for the right task
4
1
5
153
Not every robotics dataset is the right dataset for your task Imagine training a robot to sort objects. You don’t just need demonstrations - you need data that matches the task, environment, and robot you’re working with Finding the right data can be just as important as collecting it That’s what makes @PrismaXai Robotics Data interesting: connecting Physical AI builders with relevant real-world demonstrations based on their needs As robotics continues to evolve, making the right data easier to discover and access becomes an important part of the process The right data, for the right task
🤝 Paid partnership
🤖 A robot getting it right once doesn’t mean it has truly learned the task A robot may successfully pick up a cup in a lab. But can it do the same in an unfamiliar kitchen, where lighting, object positions, and surroundings are different? That’s where generalization matters For Physical AI to move beyond controlled environments, robots need diverse, relevant data that helps them learn from different situations and adapt to new ones This is why @PrismaXai 's focus on Robotics Data is worth paying attention to. Data quality and diversity are important pieces of the journey toward more capable robots The goal isn’t just to repeat what a robot has learned, but to help it adapt to what comes next
1
2
98