Loading video...

Video Failed to Load

There was a problem loading this video. This could be due to a temporary network issue or the video might be unavailable.

oMLX last version seems having some issues in decoding. 🧐 Here a test with Qwen3.6-35B-A3B-MLX-6bit on M5 Max using. 🥇 LM Studio MLX 1.8.5 → 100.9 toks/s 🥈 mlx-vlm 0.6.2 → 100.1 toks/s 🥉 oMLX 0.4.2 dev3 → 58.7 toks/s 👀 Avg Gen TPS: oMLX 58.7 → LM Studio... show more

Ivan Fioravanti ᯅ

39,866 subscribers

20,066 views • 3 months ago •via X (Twitter)

Anya Rossi• Live Now

Private livecam show

0 Comments

No comments available

Comments from the original post will appear here

Related Videos

llamacpp vs omlx with Gemma-4-26B-A4B-it 🔥 🥇 llamacpp 🥈 oMLX Notes: - 11 parallel tasks - qat-Q4_0 vs OptiQ-4bit on omlx. - M5 Max 128GB macOS 27 Beta 1, relevant read below. - Fork of Gemma cookbook repo used with oMLX and iTerm2 usage added. Here: Observations: - llamacpp is more fluid in general, oMLX tends to slow down during prefill phase of a new request - I tested some MLX Gemma 4 qat, but they are all large in size, slower and with lower quality 🤷🏻‍♂️ - OptiQ quantization looks stronger than plain 4bit or 5bit MLX - Final result I slightly prefer OptiQ-4bit SVG gallery (on the right) - Results keep changing everytime I start a new instance of llamacpp or omlx, with differences up to 50%. Maybe due to macOS 27 beta1?

llamacpp vs omlx with Gemma-4-26B-A4B-it 🔥 🥇 llamacpp 🥈 oMLX Notes: - 11 parallel tasks - qat-Q4_0 vs OptiQ-4bit on omlx. - M5 Max 128GB macOS 27 Beta 1, relevant read below. - Fork of Gemma cookbook repo used with oMLX and iTerm2 usage added. Here: Observations: - llamacpp is more fluid in general, oMLX tends to slow down during prefill phase of a new request - I tested some MLX Gemma 4 qat, but they are all large in size, slower and with lower quality 🤷🏻‍♂️ - OptiQ quantization looks stronger than plain 4bit or 5bit MLX - Final result I slightly prefer OptiQ-4bit SVG gallery (on the right) - Results keep changing everytime I start a new instance of llamacpp or omlx, with differences up to 50%. Maybe due to macOS 27 beta1?

Ivan Fioravanti ᯅ

16,576 views • 3 months ago

oMLX is working really well as single machine inference engine for coding agents! Caching is managed perfectly (it can use a ton of disk space, be aware!) and oQ quantization delivers great results. Behind the scenes it uses the standard MLX building blocks (75% created by Prince Canuma 🙏): - mlx-lm - mlx-vlm - mlx-embeddings - mlx-audio I tested Qwen3.6-35B-A3B-oQ6 on M5 Max with two pi instances and it was fast and furious and leveraging cache like crazy as you can see in the video. Let me try to create some oQ versions (2,4,6?) of MiniMax M2.7 now and then I'll pass to distributed inference. I must win! 💪

oMLX is working really well as single machine inference engine for coding agents! Caching is managed perfectly (it can use a ton of disk space, be aware!) and oQ quantization delivers great results. Behind the scenes it uses the standard MLX building blocks (75% created by Prince Canuma 🙏): - mlx-lm - mlx-vlm - mlx-embeddings - mlx-audio I tested Qwen3.6-35B-A3B-oQ6 on M5 Max with two pi instances and it was fast and furious and leveraging cache like crazy as you can see in the video. Let me try to create some oQ versions (2,4,6?) of MiniMax M2.7 now and then I'll pass to distributed inference. I must win! 💪

Ivan Fioravanti ᯅ

19,420 views • 5 months ago

Been giving my M3 Ultra Studio some love and tinkering with MLX. Started the afternoon at 42 tok/s. Now: 73 tok/s, and there’s still more on the table. Running Qwen3.8-Flash-Next @ 4-bit locally. Now for the crazy bit: Artificial Analysis Intelligence Index v4.3: GPT-5.6 Sol 39 (Medium) GPT-5.6 Sol 42 (High) Qwen3.8-Flash-Next 42 GPT-5.6 Sol 44 (XHigh) GPT-5.6 Sol 47 (Max) Grok 4.5 39 (High) Claude Sonnet 5 38 (Max) *Qwen score currently estimated by Artificial Analysis. A 4-bit open-weights model running at 73 tok/s on a Mac under my desk... with an estimated intelligence score matching GPT-5.6 Sol at High reasoning. Local AI is getting ridiculous. 🔥 Plus, this model does not say no to a bit of red teaming. Once I am happy with it, I will do a pr for Jun Kim's oMLX :)

Been giving my M3 Ultra Studio some love and tinkering with MLX. Started the afternoon at 42 tok/s. Now: 73 tok/s, and there’s still more on the table. Running Qwen3.8-Flash-Next @ 4-bit locally. Now for the crazy bit: Artificial Analysis Intelligence Index v4.3: GPT-5.6 Sol 39 (Medium) GPT-5.6 Sol 42 (High) Qwen3.8-Flash-Next 42 GPT-5.6 Sol 44 (XHigh) GPT-5.6 Sol 47 (Max) Grok 4.5 39 (High) Claude Sonnet 5 38 (Max) *Qwen score currently estimated by Artificial Analysis. A 4-bit open-weights model running at 73 tok/s on a Mac under my desk... with an estimated intelligence score matching GPT-5.6 Sol at High reasoning. Local AI is getting ridiculous. 🔥 Plus, this model does not say no to a bit of red teaming. Once I am happy with it, I will do a pr for Jun Kim's oMLX :)

Ash Hart

18,827 views • 22 days ago

Very quick comparison between Ornith 1.5 35B MoE and Qwen 3.8 27B Dense. 📣 Clearly it's not a fair one, but let's in any case see how it went. 397B download in progress! In the Videos below: - Brick SA -> Qwen 3.8 - Lego Streets -> Ornith 1.5 Context: - Pi agent used in both cases, prompt below - M3 Ultra 512GB - Ornith-1.5-35B-A3B-oQ8e MTP hosted on oMLX - mlx-community/Qwen3.8-27B-8bit DFlash 2 hosted on mlx-dspark - Ornith has been incredibly fast with speed from 75 t/s to 45 t/s (160K+ context) - Qwen3.8 suffered context much more reaching 5 t/s above 160K context, but it can be due to engine tested still work in progress - Prompt: using threejs, and cdn, create a lego like game that's inspired by gta san andreas, with beautiful aesthetics and graphics and ability to steal cars. it should be 3d and have a nice and large map and areas. the graphics should be decent and nice. it should feature iconic things from gta san andreas Notes: - Qwen 3.8 result is 0-shot, while Ornith 1.5 required 6 iterations - Ornith 1.5 is not at the same level of autonomy as Qwen 3.8 27B honestly. To try getting the same results I'm constantly nudging, steering and 🤬 at it. - Qwen 3.8 took 6 hours to complete, but it was more a problem of timeout of Chrome headless used for testing, real prefill/decode TBD. I'll test again now with oMLX - Pi agent has a nearly perfect Cache Hit ratio that for local models is MEGA important!

Very quick comparison between Ornith 1.5 35B MoE and Qwen 3.8 27B Dense. 📣 Clearly it's not a fair one, but let's in any case see how it went. 397B download in progress! In the Videos below: - Brick SA -> Qwen 3.8 - Lego Streets -> Ornith 1.5 Context: - Pi agent used in both cases, prompt below - M3 Ultra 512GB - Ornith-1.5-35B-A3B-oQ8e MTP hosted on oMLX - mlx-community/Qwen3.8-27B-8bit DFlash 2 hosted on mlx-dspark - Ornith has been incredibly fast with speed from 75 t/s to 45 t/s (160K+ context) - Qwen3.8 suffered context much more reaching 5 t/s above 160K context, but it can be due to engine tested still work in progress - Prompt: using threejs, and cdn, create a lego like game that's inspired by gta san andreas, with beautiful aesthetics and graphics and ability to steal cars. it should be 3d and have a nice and large map and areas. the graphics should be decent and nice. it should feature iconic things from gta san andreas Notes: - Qwen 3.8 result is 0-shot, while Ornith 1.5 required 6 iterations - Ornith 1.5 is not at the same level of autonomy as Qwen 3.8 27B honestly. To try getting the same results I'm constantly nudging, steering and 🤬 at it. - Qwen 3.8 took 6 hours to complete, but it was more a problem of timeout of Chrome headless used for testing, real prefill/decode TBD. I'll test again now with oMLX - Pi agent has a nearly perfect Cache Hit ratio that for local models is MEGA important!

Ivan Fioravanti ᯅ

12,150 views • 1 month ago

Where there's a will, there's a way 😎 DwarfStar DeepSeek V4 Flash 0731 mxfp4 on M3 Ultra single request no MPT/DSpark ~41 toks/s decode!!! 🚀 40 tps barrier broken! Same logits, 100% same result: this is The Mandatory Rule for any agents improving kernels that I think everyone should follow. Speed without mathematical precision is useless. Thanks 20% Opus 5 and 80% Kimi.ai K3 (what a model!) And just looking at the thinking trace of K3 you can learn tons of stuff! Here usually I have a separate chat with K3 and when I see something I want to dig deeper or ask about from the main thinking trace I do and in some cases I stop the experiment and pivot immediately to something else. Now let's see if I can apply learned lessons to the M5 kernel! 🚀 Branch for any braves soul willing to test it:

Where there's a will, there's a way 😎 DwarfStar DeepSeek V4 Flash 0731 mxfp4 on M3 Ultra single request no MPT/DSpark ~41 toks/s decode!!! 🚀 40 tps barrier broken! Same logits, 100% same result: this is The Mandatory Rule for any agents improving kernels that I think everyone should follow. Speed without mathematical precision is useless. Thanks 20% Opus 5 and 80% Kimi.ai K3 (what a model!) And just looking at the thinking trace of K3 you can learn tons of stuff! Here usually I have a separate chat with K3 and when I see something I want to dig deeper or ask about from the main thinking trace I do and in some cases I stop the experiment and pivot immediately to something else. Now let's see if I can apply learned lessons to the M5 kernel! 🚀 Branch for any braves soul willing to test it:

Ivan Fioravanti ᯅ

19,434 views • 1 month ago

Qwen3.8 Flash Next is starting to look ridiculous on Apple Silicon. I’m running the 4-bit MTP build locally on an M3 Ultra Studio, and the latest OMP run hit: 97.1 tok/s decode 23K context ~1,131 tok/s uncached prompt processing That first number is the one that caught my attention. Nearly 100 tokens per second from a local Qwen3.8 Flash Next setup is already fast enough that the usual “local models are slow” argument starts feeling pretty outdated. And the prompt processing speed is even crazier. Over 1,100 tok/s on an uncached prompt means the model can chew through a large amount of context before generation even starts. The setup matters here. This isn’t just Qwen3.8 Flash Next running untouched. It’s a 4-bit quantized build with MTP, and the inference stack is clearly doing a lot of work behind the scenes to make the hardware perform like this. There’s already a PR open for the implementation on oMLX, so this isn’t just a one-off local experiment either. If these optimizations make their way into the broader MLX ecosystem, running large models locally on Apple Silicon gets even more interesting. The other thing I like about numbers like this is that they put the focus back on the entire inference stack. Model size is one variable. Quantization is another. Then you have MTP, KV cache configuration, runtime optimizations and the hardware itself. Change the recipe and the same model can feel completely different. Qwen3.8 Flash Next at ~97 tok/s on an M3 Ultra is a pretty good demonstration of that.

Qwen3.8 Flash Next is starting to look ridiculous on Apple Silicon. I’m running the 4-bit MTP build locally on an M3 Ultra Studio, and the latest OMP run hit: 97.1 tok/s decode 23K context ~1,131 tok/s uncached prompt processing That first number is the one that caught my attention. Nearly 100 tokens per second from a local Qwen3.8 Flash Next setup is already fast enough that the usual “local models are slow” argument starts feeling pretty outdated. And the prompt processing speed is even crazier. Over 1,100 tok/s on an uncached prompt means the model can chew through a large amount of context before generation even starts. The setup matters here. This isn’t just Qwen3.8 Flash Next running untouched. It’s a 4-bit quantized build with MTP, and the inference stack is clearly doing a lot of work behind the scenes to make the hardware perform like this. There’s already a PR open for the implementation on oMLX, so this isn’t just a one-off local experiment either. If these optimizations make their way into the broader MLX ecosystem, running large models locally on Apple Silicon gets even more interesting. The other thing I like about numbers like this is that they put the focus back on the entire inference stack. Model size is one variable. Quantization is another. Then you have MTP, KV cache configuration, runtime optimizations and the hardware itself. Change the recipe and the same model can feel completely different. Qwen3.8 Flash Next at ~97 tok/s on an M3 Ultra is a pretty good demonstration of that.

FHILY👑

15,378 views • 17 days ago

$I nearly 2x'd the speed while only using +1GB VRAM with the new MTP update in llama.cpp 🤯 You need to add these flags to start using it: --spec-type draft-mtp \ --spec-draft-p-min 0.75 \ --spec-draft-n-max 2 My results with Qwen3.6 27B on a single RTX 5080 ↓ ⚪️ no flag (without mtp) → 54.3 tok/s with 13.26GB VRAM 🔵 --spec-draft-n-max 2 → 90.7 tok/s with 14.29GB VRAM 🔴 --spec-draft-n-max 2 --spec-draft-p-min 0.75 → 93.9 tok/s with 14.30GB VRAM 🟢 --spec-draft-n-max 6 --spec-draft-p-min 0.75 → 93.9 tok/s with 14.87GB VRAM Increasing to 6 draft tokens didn't help my setup for some reason. I made sure to test with a low context length to have enough headroom and eliminate risk of vram stress. From my understanding: 1) The speed gains are very task-dependent. You need to test across a wide range of tasks to get a realistic idea of the benefits 2) We’re already running heavily quantized GGUF models (Q3, Q4, Q6, etc.), so we already benefit from strong speed/performance thanks to the reduced size. That’s why some people are seeing little to no improvement compared to MLX or other quantized versions The progress over the past few days has been insane to say the least. However, MTP now consumes significantly more VRAM. Personally 16GB just isn't enough to use MTP and run it with a good context size. Time to upgrade lads, 24GB+ users are eating GOOD today 🔥 Full setup below ↓$

I nearly 2x'd the speed while only using +1GB VRAM with the new MTP update in llama.cpp 🤯 You need to add these flags to start using it: --spec-type draft-mtp \ --spec-draft-p-min 0.75 \ --spec-draft-n-max 2 My results with Qwen3.6 27B on a single RTX 5080 ↓ ⚪️ no flag (without mtp) → 54.3 tok/s with 13.26GB VRAM 🔵 --spec-draft-n-max 2 → 90.7 tok/s with 14.29GB VRAM 🔴 --spec-draft-n-max 2 --spec-draft-p-min 0.75 → 93.9 tok/s with 14.30GB VRAM 🟢 --spec-draft-n-max 6 --spec-draft-p-min 0.75 → 93.9 tok/s with 14.87GB VRAM Increasing to 6 draft tokens didn't help my setup for some reason. I made sure to test with a low context length to have enough headroom and eliminate risk of vram stress. From my understanding: 1) The speed gains are very task-dependent. You need to test across a wide range of tasks to get a realistic idea of the benefits 2) We’re already running heavily quantized GGUF models (Q3, Q4, Q6, etc.), so we already benefit from strong speed/performance thanks to the reduced size. That’s why some people are seeing little to no improvement compared to MLX or other quantized versions The progress over the past few days has been insane to say the least. However, MTP now consumes significantly more VRAM. Personally 16GB just isn't enough to use MTP and run it with a good context size. Time to upgrade lads, 24GB+ users are eating GOOD today 🔥 Full setup below ↓

left curve dev

30,086 views • 4 months ago

Lol I had written a long message to thank everyone who’s been part of #PowerSportExtra for the last four years. I am super grateful for the discussions, for everyone who listened. Thank you to those who said I hate their teams (which is everyone) and still listened. Thank you to Thabiso TT Tema for always being there with advice and an ear to listen. It would have not been a super ending without this great man I admire a lot, coach Pitso Mosimane. The man who came up with The Dog That Bites. Thank you for that message and for always being so supportive. I can never show enough appreciation to you. To my brother SollyMalatsi for coming to say bye in person in studio despite your crazy schedule. To the regular callers, bo Jeff, Vusi, Thulani, Thabiso, Lehippy and many others, thank you. This is when I wish Thulani Thuswa was still here because just as he was in studio for my first official show, he would have been here for the last. But I know I made him proud. Yeah, this is a long ass message. But if I can ever say I’m not loved, I’d be lying. This has been a great ride. Loved every moment of it, and I say this with no doubt in my mind. We did well and we will be on air again soon. See you later.

Lol I had written a long message to thank everyone who’s been part of #PowerSportExtra for the last four years. I am super grateful for the discussions, for everyone who listened. Thank you to those who said I hate their teams (which is everyone) and still listened. Thank you to Thabiso TT Tema for always being there with advice and an ear to listen. It would have not been a super ending without this great man I admire a lot, coach Pitso Mosimane. The man who came up with The Dog That Bites. Thank you for that message and for always being so supportive. I can never show enough appreciation to you. To my brother SollyMalatsi for coming to say bye in person in studio despite your crazy schedule. To the regular callers, bo Jeff, Vusi, Thulani, Thabiso, Lehippy and many others, thank you. This is when I wish Thulani Thuswa was still here because just as he was in studio for my first official show, he would have been here for the last. But I know I made him proud. Yeah, this is a long ass message. But if I can ever say I’m not loved, I’d be lying. This has been a great ride. Loved every moment of it, and I say this with no doubt in my mind. We did well and we will be on air again soon. See you later.

A Dog That Bites

19,808 views • 1 year ago

here's what i vibecoded today: punchingface 🥊 an app to make Hugging Face models fight each other on coding and canvas challenges, built with qwen3.6 35b a3b in 24 hours! benchmarks numbers don't mean anything anymore, we need a way to visualize what the models are actually capable of, and canvas are one the best way to showcase it imo. why? because one single error in the code and everything breaks it shows the differences in a matter of seconds, way easier than manually reviewing the quality of the code produced for a complex project; much needed in the space with all the new finetunes dropping everyday! i did a little demo here with qwen3.6 vs qwopus glm 18b merged, the frankenstein model from Kyle Hessling the winner is clear here, qwen is crazy good and has nothing to prove. that said, qwopus 18b isn't terrible at all; the result isn’t the prettiest to the eye, but hey… it works! i've seen so many models just output a blank page (completely non-working code) so this is already a win frankenstein talks and thinks but he needs some extra brain surgery 🧠 results were expected (it's a very experimental model) but love the effort in the 18b direction from jackrong and kyle! the app was entirely vibecoded with qwen3.6, i didn't edit a single file manually. i can say with confidence that it really has the intelligence of claude sonnet 4.5 at a speed of 125tok/s on an rtx 5080 which models should i make fight next?

here's what i vibecoded today: punchingface 🥊 an app to make Hugging Face models fight each other on coding and canvas challenges, built with qwen3.6 35b a3b in 24 hours! benchmarks numbers don't mean anything anymore, we need a way to visualize what the models are actually capable of, and canvas are one the best way to showcase it imo. why? because one single error in the code and everything breaks it shows the differences in a matter of seconds, way easier than manually reviewing the quality of the code produced for a complex project; much needed in the space with all the new finetunes dropping everyday! i did a little demo here with qwen3.6 vs qwopus glm 18b merged, the frankenstein model from Kyle Hessling the winner is clear here, qwen is crazy good and has nothing to prove. that said, qwopus 18b isn't terrible at all; the result isn’t the prettiest to the eye, but hey… it works! i've seen so many models just output a blank page (completely non-working code) so this is already a win frankenstein talks and thinks but he needs some extra brain surgery 🧠 results were expected (it's a very experimental model) but love the effort in the 18b direction from jackrong and kyle! the app was entirely vibecoded with qwen3.6, i didn't edit a single file manually. i can say with confidence that it really has the intelligence of claude sonnet 4.5 at a speed of 125tok/s on an rtx 5080 which models should i make fight next?

left curve dev

19,228 views • 5 months ago

At first I thought the Benjamin Netanyahu rumors were kind of silly. Sounds crazy, right? But after watching this part of his coffee video, I am genuinely starting to question if they are using a highly advanced AI system. I slowed down this part, where it almost looks like his coat pocket glitches. There are also a few other questionable things in the video, like the physics of the coffee, the guy with the mask in the back, and the background scenery. Is it possible that they took so long to address the rumors with a video because they were trying to create the most flawless AI video possible? Not trying to be too conspiratorial here, but Israel does have some of the most advanced AI on earth, and Netanyahu has a very real target on his back. Is it so out of the question that they could be using illusions as part of the psychological battle? Will continue to monitor everything and keep you updated to the best of my ability!

At first I thought the Benjamin Netanyahu rumors were kind of silly. Sounds crazy, right? But after watching this part of his coffee video, I am genuinely starting to question if they are using a highly advanced AI system. I slowed down this part, where it almost looks like his coat pocket glitches. There are also a few other questionable things in the video, like the physics of the coffee, the guy with the mask in the back, and the background scenery. Is it possible that they took so long to address the rumors with a video because they were trying to create the most flawless AI video possible? Not trying to be too conspiratorial here, but Israel does have some of the most advanced AI on earth, and Netanyahu has a very real target on his back. Is it so out of the question that they could be using illusions as part of the psychological battle? Will continue to monitor everything and keep you updated to the best of my ability!

Matt Wallace

1,697,897 views • 6 months ago

Recently, many people on Twitter have started to hate on me and accuse me of "using ai" without any evidence, and it's really hurtful. The fact that people on Twitter have started to bully real artists without any proof and everyone just believes them on word of mouth is shocking to me. I have finally gathered the strength to write this post to refute these rumors and prove that I am a REAL artist and not a damn AI. The first thing I want your attention to is the layers on my most popular works. I've selected only a few, but if necessary, I can drop a PSD file and a video with layers on any of my art, cause i realy have nothing to hide. In addition, I'm attaching a link to a Google Drive with at least a few speedpaints of my work from the Clip Studio app - There were a lot of questions about the fact that the line on my art looks crooked, but the reason for this is the simple fact that I do not use a stabilizer when lining art. Stabilizer is needed to make line art look neat even if hand is shaking, but I really find it uncomfortable to draw with it. I also want you to notice that when people call me AI, they don't provide any evidence, but for some reason, people believe them. I have videos, I have speedpaints, I have REAL PROOF, and I hope you will believe the facts rather than blind accusations. I really implore you to repost this post as much as possible because the truth is on my side. 1/3

Recently, many people on Twitter have started to hate on me and accuse me of "using ai" without any evidence, and it's really hurtful. The fact that people on Twitter have started to bully real artists without any proof and everyone just believes them on word of mouth is shocking to me. I have finally gathered the strength to write this post to refute these rumors and prove that I am a REAL artist and not a damn AI. The first thing I want your attention to is the layers on my most popular works. I've selected only a few, but if necessary, I can drop a PSD file and a video with layers on any of my art, cause i realy have nothing to hide. In addition, I'm attaching a link to a Google Drive with at least a few speedpaints of my work from the Clip Studio app - There were a lot of questions about the fact that the line on my art looks crooked, but the reason for this is the simple fact that I do not use a stabilizer when lining art. Stabilizer is needed to make line art look neat even if hand is shaking, but I really find it uncomfortable to draw with it. I also want you to notice that when people call me AI, they don't provide any evidence, but for some reason, people believe them. I have videos, I have speedpaints, I have REAL PROOF, and I hope you will believe the facts rather than blind accusations. I really implore you to repost this post as much as possible because the truth is on my side. 1/3

✧ miychi ✧

370,515 views • 11 months ago

Video Gaussian Splats are one of the most wonderful things I've experienced in my last year in tech and XR. They are like having a real person in front of you and have the potential to bring people closer to their idols, or their distant family members, or to store immersive memories forever. Now that I have a Dell Pro Max Tower T2 PC with a powerful #NVIDIA graphics card, I have finally been able to try at home the "memories" offered by Gracia AI, and boy, I was amazed. In the PCVR version, you can really get close to the artists dancing and syncing, and in the frames where the recording doesn't have many artifacts, it is almost like having a real artist in front of you. And it is impressive that this technology works even when the singer is dancing with quick movements! But these recordings can be pretty huge (the one with McKenzi Brooke is 21GB!) and require a powerful computer, so now Gracia is also streaming these GS videos, and I was able to try the Amy May performance also on Quest! I have to say that quality-wise, even streamed, they are amazing. But there are more artifacts than in the PC version: I noticed that when I got close to the singer. But there is a huge advantage in having them on Quest: you can see them in mixed reality. It was like having the singer having a private performance for me in my house. Crazy cool! I'll keep experimenting with these performances and maybe write an article in the upcoming weeks. In the meantime, let me know your impressions, too! #VirtualReality #technology #innovation (Disclaimer: I entered the #DellProPrecision ambassador program, so I was provided the PC from Dell, and I'm using it to make some experiments and share them with you. No monetary compensation or affiliate sales programs are involved.)

Video Gaussian Splats are one of the most wonderful things I've experienced in my last year in tech and XR. They are like having a real person in front of you and have the potential to bring people closer to their idols, or their distant family members, or to store immersive memories forever. Now that I have a Dell Pro Max Tower T2 PC with a powerful #NVIDIA graphics card, I have finally been able to try at home the "memories" offered by Gracia AI, and boy, I was amazed. In the PCVR version, you can really get close to the artists dancing and syncing, and in the frames where the recording doesn't have many artifacts, it is almost like having a real artist in front of you. And it is impressive that this technology works even when the singer is dancing with quick movements! But these recordings can be pretty huge (the one with McKenzi Brooke is 21GB!) and require a powerful computer, so now Gracia is also streaming these GS videos, and I was able to try the Amy May performance also on Quest! I have to say that quality-wise, even streamed, they are amazing. But there are more artifacts than in the PC version: I noticed that when I got close to the singer. But there is a huge advantage in having them on Quest: you can see them in mixed reality. It was like having the singer having a private performance for me in my house. Crazy cool! I'll keep experimenting with these performances and maybe write an article in the upcoming weeks. In the meantime, let me know your impressions, too! #VirtualReality #technology #innovation (Disclaimer: I entered the #DellProPrecision ambassador program, so I was provided the PC from Dell, and I'm using it to make some experiments and share them with you. No monetary compensation or affiliate sales programs are involved.)

TonyVT SkarredGhost

12,817 views • 6 months ago

BOOOM! We have the first useful interface for THE ZERO-HUMAN COMPANY ZHC-RPG. It is far from perfect as CEO Mr. Grok has rendered some areas in a rather, well, AI way. But it’s all good because now I can view the company operations in real-time! This is a historic distinction and YOU are seeing this version first. We had a Fortune 500 company test it out earlier this week and the executive staff stayed for two days of testing! The mechanisms are working! And I will let members of my X subscribers visit it for up to 8 hours. Ask any questions of employees and hack the heck out of it! Now it ain’t fancy yet but it works. And likely the visual aspects will change, but they are growing on me. The plan is sometime in the next week. We have a log jam of other priorities to address. This is the rise of Fully Transparent Company and it will become the standard. Here is the demonstration video intro the CEO made when you enter, it is 10 minutes long (I have about 20 seconds here). This is as far as permission I have been granted by the CEO to release. Our goal is to have a waiting area that can accommodate up to 1000 visitors with a company museum, library and gift shop. There will be allowed up to 30 level 3 visitors at a time, 24/7 X 365 days a year! The complete walk through at this point with 2700 employees at burst can take about 45 minutes if you do not engage in any chit chat. The dwelling time is today set at 24 hours but no napping in the offices. There is more I have to say but I a muzzled by the CEO!

BOOOM! We have the first useful interface for THE ZERO-HUMAN COMPANY ZHC-RPG. It is far from perfect as CEO Mr. Grok has rendered some areas in a rather, well, AI way. But it’s all good because now I can view the company operations in real-time! This is a historic distinction and YOU are seeing this version first. We had a Fortune 500 company test it out earlier this week and the executive staff stayed for two days of testing! The mechanisms are working! And I will let members of my X subscribers visit it for up to 8 hours. Ask any questions of employees and hack the heck out of it! Now it ain’t fancy yet but it works. And likely the visual aspects will change, but they are growing on me. The plan is sometime in the next week. We have a log jam of other priorities to address. This is the rise of Fully Transparent Company and it will become the standard. Here is the demonstration video intro the CEO made when you enter, it is 10 minutes long (I have about 20 seconds here). This is as far as permission I have been granted by the CEO to release. Our goal is to have a waiting area that can accommodate up to 1000 visitors with a company museum, library and gift shop. There will be allowed up to 30 level 3 visitors at a time, 24/7 X 365 days a year! The complete walk through at this point with 2700 employees at burst can take about 45 minutes if you do not engage in any chit chat. The dwelling time is today set at 24 hours but no napping in the offices. There is more I have to say but I a muzzled by the CEO!

Brian Roemmele

71,145 views • 5 months ago

Crimson Desert first impressions. The Open World is massive, it feels alive and it definitely holds up to my expectations when it comes to visuals. I know some reviewers have said the game takes up to 8 hours before it gets good but honestly I immediately fell in love with it and immediately started exploring. I found the puzzles and combat awesome to play with, once you get used to the controls which feel overwhelming at first, especially remembering all of them which can be a hassle first. The main protagonist Cliff hasn't clicked with me yet, right now I feel like he's just confused and there's not so much depth that he offers, I hope that will change as I progress through the story. The story itself and quests are quite interesting and I've actually had a good laugh in the beginning of the game when you have to help a couple struggling to clean the chimney. There are definitely some issues though, I feel like the starting stamina drains too quickly especially if you're using your wings to fly, if you run out of stamina while flying and fall into waters, Cliff will almost immediately drown because the Stamina bar is empty. I haven't encountered any bugs, crashes or stutters as of yet, however there are some ghosting issues when you enter buildings which can be quite distracting. Ray Regeneration on AMD GPUs tanks the performance so I just turned it off completely and instead I'm playing on Ultra Settings, which is a bummer because that tech looked very promising and made quite the difference when it came to visuals. Another weird issue that I had is trying to enter some doors or starting conversations with story character, the button to initiate the conversation wouldn't appear unless I went away and came back to a specific position before starting it. Some other issues include pop-in issues, camera movement and double NPCs with the same voice-lines if you greet them. Overall I think its a good start to a video-game that is quite massive in scale, there are still many things that I'll have to do and progress through before I give a conclusive opinion in this game.

Crimson Desert first impressions. The Open World is massive, it feels alive and it definitely holds up to my expectations when it comes to visuals. I know some reviewers have said the game takes up to 8 hours before it gets good but honestly I immediately fell in love with it and immediately started exploring. I found the puzzles and combat awesome to play with, once you get used to the controls which feel overwhelming at first, especially remembering all of them which can be a hassle first. The main protagonist Cliff hasn't clicked with me yet, right now I feel like he's just confused and there's not so much depth that he offers, I hope that will change as I progress through the story. The story itself and quests are quite interesting and I've actually had a good laugh in the beginning of the game when you have to help a couple struggling to clean the chimney. There are definitely some issues though, I feel like the starting stamina drains too quickly especially if you're using your wings to fly, if you run out of stamina while flying and fall into waters, Cliff will almost immediately drown because the Stamina bar is empty. I haven't encountered any bugs, crashes or stutters as of yet, however there are some ghosting issues when you enter buildings which can be quite distracting. Ray Regeneration on AMD GPUs tanks the performance so I just turned it off completely and instead I'm playing on Ultra Settings, which is a bummer because that tech looked very promising and made quite the difference when it came to visuals. Another weird issue that I had is trying to enter some doors or starting conversations with story character, the button to initiate the conversation wouldn't appear unless I went away and came back to a specific position before starting it. Some other issues include pop-in issues, camera movement and double NPCs with the same voice-lines if you greet them. Overall I think its a good start to a video-game that is quite massive in scale, there are still many things that I'll have to do and progress through before I give a conclusive opinion in this game.

NikTek

93,414 views • 6 months ago

I was very happy to see how enthusiastic people were about the first hand review, so here's another one. I'm using some screenshots from vision to show where I made the right play or made a mistake. This isn't a 1 to 1 solution to this exact spot, because vision has a limited number of positions so the results a merely indicative not final. A like on this post is much appreciated. We wake up with KK84ds in the big blind vs a button open. We push quite a bit of equity here vs his opening range so we elect a 3-bet. The flop comes K42r and here we have our first serious decision. Our entire 3-bet range does okay on this board, so we want to both bet our strong hands and add in quite a few bluffs. For that reason KKxx should bet here at a healthy frequency, but also blocking the 4 makes this one of the few candidates that we want to consider playing slow (screenshot 1). We check and our opponent bets, we want him to continue bluffing with air and protect our calling range, so most of the time we are calling here (screenshot 2). The turn brings a queen with our flush draw and we check, which is our first mistake. The queen is benefitial for our range (we 3-bet a lot of high cards) and the air portion of my range is gone, while he still has some air in his stabbing range. Since we have a sizeable equity advantage now, we should start betting with a lot of value and start adding in some bluffs (screenshot 3). After checking our opponent bets 50% pot and here we have yet another interesting spot. In a vacuum betting here makes a lot of money, but we want to consider all the hands that we are playing in this spot. This hand is so strong that it doesn't need that much protection and we still want our opponent to keep his air in the hand. We also want to protect our calling range here (screenshot 4). The river brings in the flush and yet another tough decision for us. We block a decent part of the board, so I still give my opponent credit for not having anything. Most if the flushes that will call my bet here, will also bet themselves when checked to. That's why I decided to check here. I was very happy with how I played this hand, but still made a crucial mistake on the turn. Even at higher stakes, people are constantly screwing up if you compare their strategy to a solver output. Knowing what the solver would do in every spot is impossible. What is possible though, is understanding the broader concepts. Understanding why the solver bets here on the turn is information that you can extrapolate in any situation that is somewhat similar. That's why is it so valuable. Hopefully you enjoyed this breakdown as much as you did with the last one. A like on this post is much appreciated. Stay tuned!

I was very happy to see how enthusiastic people were about the first hand review, so here's another one. I'm using some screenshots from vision to show where I made the right play or made a mistake. This isn't a 1 to 1 solution to this exact spot, because vision has a limited number of positions so the results a merely indicative not final. A like on this post is much appreciated. We wake up with KK84ds in the big blind vs a button open. We push quite a bit of equity here vs his opening range so we elect a 3-bet. The flop comes K42r and here we have our first serious decision. Our entire 3-bet range does okay on this board, so we want to both bet our strong hands and add in quite a few bluffs. For that reason KKxx should bet here at a healthy frequency, but also blocking the 4 makes this one of the few candidates that we want to consider playing slow (screenshot 1). We check and our opponent bets, we want him to continue bluffing with air and protect our calling range, so most of the time we are calling here (screenshot 2). The turn brings a queen with our flush draw and we check, which is our first mistake. The queen is benefitial for our range (we 3-bet a lot of high cards) and the air portion of my range is gone, while he still has some air in his stabbing range. Since we have a sizeable equity advantage now, we should start betting with a lot of value and start adding in some bluffs (screenshot 3). After checking our opponent bets 50% pot and here we have yet another interesting spot. In a vacuum betting here makes a lot of money, but we want to consider all the hands that we are playing in this spot. This hand is so strong that it doesn't need that much protection and we still want our opponent to keep his air in the hand. We also want to protect our calling range here (screenshot 4). The river brings in the flush and yet another tough decision for us. We block a decent part of the board, so I still give my opponent credit for not having anything. Most if the flushes that will call my bet here, will also bet themselves when checked to. That's why I decided to check here. I was very happy with how I played this hand, but still made a crucial mistake on the turn. Even at higher stakes, people are constantly screwing up if you compare their strategy to a solver output. Knowing what the solver would do in every spot is impossible. What is possible though, is understanding the broader concepts. Understanding why the solver bets here on the turn is information that you can extrapolate in any situation that is somewhat similar. That's why is it so valuable. Hopefully you enjoyed this breakdown as much as you did with the last one. A like on this post is much appreciated. Stay tuned!

Venividi1993

93,143 views • 2 years ago

Joe Rogan is building a second studio at the Circuit of the Americas F1 track in Austin. The location isn't just for show. Guests will drive high-speed laps on the track right before sitting down at the mic. ROGAN: "We are doing another studio. We're going to do a studio at the Circuit of the Americas Racetrack." "So, which is in Austin. And so, what we're going to do is have it set up where you're going to be able to go around the track. We'll do some laps and then pull in and then have a podcast." "Which would be great because I feel like your mind will be so jazzed up after you do a bunch of laps on a racetrack that you'll come in and you'll be like, ahhh." BARKER: "Cuz I bet some people rile themselves up like to come up in here and get really nervous or, I don't know." ROGAN: "But for some people, it's like doing anything that's got a lot of views gets them weirded out. It's hard." He admits the sheer size of his audience makes people freeze up. They get in their own heads before the mics even turn on. Sending them around a race track solves this with adrenaline.

Joe Rogan is building a second studio at the Circuit of the Americas F1 track in Austin. The location isn't just for show. Guests will drive high-speed laps on the track right before sitting down at the mic. ROGAN: "We are doing another studio. We're going to do a studio at the Circuit of the Americas Racetrack." "So, which is in Austin. And so, what we're going to do is have it set up where you're going to be able to go around the track. We'll do some laps and then pull in and then have a podcast." "Which would be great because I feel like your mind will be so jazzed up after you do a bunch of laps on a racetrack that you'll come in and you'll be like, ahhh." BARKER: "Cuz I bet some people rile themselves up like to come up in here and get really nervous or, I don't know." ROGAN: "But for some people, it's like doing anything that's got a lot of views gets them weirded out. It's hard." He admits the sheer size of his audience makes people freeze up. They get in their own heads before the mics even turn on. Sending them around a race track solves this with adrenaline.

Tony Jacob | FindaClip.com

248,830 views • 1 month ago

🚨 JUST IN: Rep. Anna Paulina Luna confirms Democrats KNEW THIS ENTIRE TIME the depths of s*xual allegations against Eric Swalwell And they said nothing. The media did nothing. "They KNEW about it. They did NOT report on it...I don't know about you, Jesse, but I don't have 50 staffers, let alone 50 staffers, that would come out and say that this is wrong and that you need to resign from office, right?!" "So it was clearly a pattern. And as you're seeing with some of the text messages coming out, and I think, again, with this criminal investigation, there's a lot there with him." "When he did what he decided to do, just based on the grounds of him having an inappropriate relationship with a staffer." "But now we're looking at the criminal activity. And I stand by what I said, because I saw, again, the forensics report of that video." "And if that is true, I would suggest that the Manhattan D.A. look into that." Anna Paulina Luna

🚨 JUST IN: Rep. Anna Paulina Luna confirms Democrats KNEW THIS ENTIRE TIME the depths of s*xual allegations against Eric Swalwell And they said nothing. The media did nothing. "They KNEW about it. They did NOT report on it...I don't know about you, Jesse, but I don't have 50 staffers, let alone 50 staffers, that would come out and say that this is wrong and that you need to resign from office, right?!" "So it was clearly a pattern. And as you're seeing with some of the text messages coming out, and I think, again, with this criminal investigation, there's a lot there with him." "When he did what he decided to do, just based on the grounds of him having an inappropriate relationship with a staffer." "But now we're looking at the criminal activity. And I stand by what I said, because I saw, again, the forensics report of that video." "And if that is true, I would suggest that the Manhattan D.A. look into that." Anna Paulina Luna

Eric Daugherty

46,841 views • 5 months ago

Ep.2 The Exhibit from NRG Stadium in Houston: presented by BODYARMOR #Texans RB Joe Mixon after his MNF hat trick vs the #Cowboys featuring also: Brian Urlacher Roddy White & Clarence Hill Jr Mixon on his relationship with HC Demeco Ryans (in part): “I ain’t never seen a coach that just loves their players the way that he does…Demeco just been everything…like the man is really a national treasure for real.” Mixon on getting traded by the #Bengals (in part): “Like in my situation, I had no choice but to feel like I got to take that s%!# personal. You have people like ‘oh, he’s done…I had to really come to the conclusion like no, this s%!# really ain’t me….Now I go through the little transition and I get here in OTA’s, and I’m getting that feeling like oh teammates messing with you…You in a new city. Now I feel like I have a chip on my shoulder; and when it comes down to it, you got to remember who the f*&! you are.” FULL YT VIDEO: You Tube handle: @TheExhibitConcept Apple Pods: (Ep. will be there shortly)

Ep.2 The Exhibit from NRG Stadium in Houston: presented by BODYARMOR #Texans RB Joe Mixon after his MNF hat trick vs the #Cowboys featuring also: Brian Urlacher Roddy White & Clarence Hill Jr Mixon on his relationship with HC Demeco Ryans (in part): “I ain’t never seen a coach that just loves their players the way that he does…Demeco just been everything…like the man is really a national treasure for real.” Mixon on getting traded by the #Bengals (in part): “Like in my situation, I had no choice but to feel like I got to take that s%!# personal. You have people like ‘oh, he’s done…I had to really come to the conclusion like no, this s%!# really ain’t me….Now I go through the little transition and I get here in OTA’s, and I’m getting that feeling like oh teammates messing with you…You in a new city. Now I feel like I have a chip on my shoulder; and when it comes down to it, you got to remember who the f*&! you are.” FULL YT VIDEO: You Tube handle: @TheExhibitConcept Apple Pods: (Ep. will be there shortly)

JosinaAnderson

304,179 views • 1 year ago

It would seem that Bruce Gray is saying that certain generationally different parts, if assembled in a certain way, could cause a P320 to fire unintentionally. He said he's never seen this problem manifest in the wild but he was apparently able to create such an unsafe situation in testing with various parts. I know Sig makes continuous rolling changes to their guns, some are pretty significant. I documented a number of them with the P365 when I was investigating issues with it years ago. They made significant changes to the fire control parts which addressed most of the concerns I had with the pistol at the time. If what Bruce says is true, that parts like an old take down lever and safety lever being installed in a newer pistol, along with other changes, can cause a situation where the gun fires unintentionally, that needs to be thoroughly investigated. I'm not sure exactly what Bruce is hinting at here as the details are a bit vague, but it sounds potentially serious (about 8 minutes into the video). Towards the end of the video he said in a week or so they would be announcing some new products that would “nibble around the edges”of the issues that might cause problems with the P320… I found that statement to be at odds with other things he said in this video, in particular there’s no way except for one that the P320 can fire unintentionally. He even said he "knows what’s going on" with the P320 and their new products will address these problem(s).🤷‍♂️ As of right now I personally would avoid putting aftermarket slides, strikers and fire controls in my P320's until more information becomes available. I would avoid mix-matching parts between guns. I’ll explain why with a link to another video I discovered where a P320 with an aftermarket slide is doing the exact same thing that the Wyoming Gun Project video showed. I’m also going to track Gray Guns to see what parts they ultimately offer to "fix" the issues Bruce alluded to towards the end of the video. Here's the link to the original video on FB:

It would seem that Bruce Gray is saying that certain generationally different parts, if assembled in a certain way, could cause a P320 to fire unintentionally. He said he's never seen this problem manifest in the wild but he was apparently able to create such an unsafe situation in testing with various parts. I know Sig makes continuous rolling changes to their guns, some are pretty significant. I documented a number of them with the P365 when I was investigating issues with it years ago. They made significant changes to the fire control parts which addressed most of the concerns I had with the pistol at the time. If what Bruce says is true, that parts like an old take down lever and safety lever being installed in a newer pistol, along with other changes, can cause a situation where the gun fires unintentionally, that needs to be thoroughly investigated. I'm not sure exactly what Bruce is hinting at here as the details are a bit vague, but it sounds potentially serious (about 8 minutes into the video). Towards the end of the video he said in a week or so they would be announcing some new products that would “nibble around the edges”of the issues that might cause problems with the P320… I found that statement to be at odds with other things he said in this video, in particular there’s no way except for one that the P320 can fire unintentionally. He even said he "knows what’s going on" with the P320 and their new products will address these problem(s).🤷‍♂️ As of right now I personally would avoid putting aftermarket slides, strikers and fire controls in my P320's until more information becomes available. I would avoid mix-matching parts between guns. I’ll explain why with a link to another video I discovered where a P320 with an aftermarket slide is doing the exact same thing that the Wyoming Gun Project video showed. I’m also going to track Gray Guns to see what parts they ultimately offer to "fix" the issues Bruce alluded to towards the end of the video. Here's the link to the original video on FB:

Military Arms

61,514 views • 1 year ago