正在加载视频...

视频加载失败

加载此视频时出现问题。这可能是由于临时网络问题，或视频可能不可用。

Mistral OCR 4 just dropped with bounding boxes (their most-requested feature) so I plugged it into my form-filling test as the helper model. Qwen3.6 reasons, Mistral localizes. Result? Boxes detected, fields filled, mostly landing in the lines. Not pixel-perfect. But close? Yeah, I'll call it close.

stevibe

27,272 subscribers

41,430 次观看 • 26 天前 •via X (Twitter)

Anya Rossi• Live Now

Private livecam show

0 条评论

暂无评论

原始帖子的评论将显示在这里

相关视频

I designed a new test specifically for multimodal models: fill out a paper form. And it's much harder than it sounds. This isn't typing into an electronic field that captures your text. The form is just an image. The model has to place each form element: text, checkmarks — at the correct pixel position on the canvas itself. Results: 🟢 Kimi K2.6 → done in 3:45, 16.7k output tokens 🟡 Step 3.7 Flash → half the fields, 57k output tokens 🔴 Gemini 3.5 Flash → 489k output tokens, never finished. I had to kill it. Gemini burned ~29x more output tokens than Kimi on the exact same task, and Kimi's was the only form that actually looked filled out. The test, a mocked application form, contains some challenging parts, such as one-character-per-box fields. I provided every model the same set of tools: > get canvas size > drop probe markers to find coordinates > add text > add checkmarks > move elements > take a screenshot anytime to check their own work > ... etc So it's vision + spatial reasoning + tool use + long context, all at once. Small models (Qwen, Gemma) can't really complete this test, so I skipped them. What happened: > Kimi nailed name, DOB, ID, gender, marital status, nationality, email, phone, address, postal code — placement slightly loose, but content correct. 15 turns. Clean. > Step got maybe half right — fields dropped, "United States" landed in the email line, data floating outside boxes. Burned 1.24M input tokens doing it (81 turns of re-reading the canvas). > Gemini almost got there visually... then spiraled. By turn 40 it was issuing a delete_elements call wiping element IDs 365–425, basically erasing its own work. 31 minutes, 489k output tokens, still streaming. Terminated. The takeaway isn't "Gemini bad." This test is indeed difficult. But token efficiency is capability now. A model that needs 30x the tokens and still can't converge is going to be 30x the cost in production. Kimi K2.6 just quietly did the thing.

I designed a new test specifically for multimodal models: fill out a paper form. And it's much harder than it sounds. This isn't typing into an electronic field that captures your text. The form is just an image. The model has to place each form element: text, checkmarks — at the correct pixel position on the canvas itself. Results: 🟢 Kimi K2.6 → done in 3:45, 16.7k output tokens 🟡 Step 3.7 Flash → half the fields, 57k output tokens 🔴 Gemini 3.5 Flash → 489k output tokens, never finished. I had to kill it. Gemini burned ~29x more output tokens than Kimi on the exact same task, and Kimi's was the only form that actually looked filled out. The test, a mocked application form, contains some challenging parts, such as one-character-per-box fields. I provided every model the same set of tools: > get canvas size > drop probe markers to find coordinates > add text > add checkmarks > move elements > take a screenshot anytime to check their own work > ... etc So it's vision + spatial reasoning + tool use + long context, all at once. Small models (Qwen, Gemma) can't really complete this test, so I skipped them. What happened: > Kimi nailed name, DOB, ID, gender, marital status, nationality, email, phone, address, postal code — placement slightly loose, but content correct. 15 turns. Clean. > Step got maybe half right — fields dropped, "United States" landed in the email line, data floating outside boxes. Burned 1.24M input tokens doing it (81 turns of re-reading the canvas). > Gemini almost got there visually... then spiraled. By turn 40 it was issuing a delete_elements call wiping element IDs 365–425, basically erasing its own work. 31 minutes, 489k output tokens, still streaming. Terminated. The takeaway isn't "Gemini bad." This test is indeed difficult. But token efficiency is capability now. A model that needs 30x the tokens and still can't converge is going to be 30x the cost in production. Kimi K2.6 just quietly did the thing.

stevibe

25,304 次观看 • 1 个月前

Most people think a 0.8B model is a toy. 3rd test. This time: OCR I fed Qwen 3.5 0.8B three images: a Kindle page, my terrible handwriting x 2. Mostly correct. Running locally. From a model smaller than your browser tab.

Most people think a 0.8B model is a toy. 3rd test. This time: OCR I fed Qwen 3.5 0.8B three images: a Kindle page, my terrible handwriting x 2. Mostly correct. Running locally. From a model smaller than your browser tab.

stevibe

125,136 次观看 • 4 个月前

Qwen3.6 35B-A3B dropped yesterday, so I ran it on 4 GPUs to see how it performs: 🟣 RTX 3090 — 49.78 tok/s, TTFT 852ms 🟡 RTX 4090 — 118.93 tok/s, TTFT 686ms 🟢 RTX 5090 — 160.37 tok/s, TTFT 409ms 🔵 DGX Spark — 59.98 tok/s, TTFT 228ms I went with ollama as the backend because honestly, it's the easiest way for most people to get started. One command, model pulled, done. I used Q4_K_M (24GB) across all four cards. The reason is the 3090 and 4090 don't support NVFP4 (only the 5090 and DGX Spark could use it). Keeping the same quant everywhere felt like the fairest way to compare. And yes, you can absolutely squeeze more performance out of every card with vLLM, SGLang, or TensorRT-LLM. But that's not what this test is about. This is just the out-of-the-box experience for folks who own a GPU and want to try the new model tonight.

Qwen3.6 35B-A3B dropped yesterday, so I ran it on 4 GPUs to see how it performs: 🟣 RTX 3090 — 49.78 tok/s, TTFT 852ms 🟡 RTX 4090 — 118.93 tok/s, TTFT 686ms 🟢 RTX 5090 — 160.37 tok/s, TTFT 409ms 🔵 DGX Spark — 59.98 tok/s, TTFT 228ms I went with ollama as the backend because honestly, it's the easiest way for most people to get started. One command, model pulled, done. I used Q4_K_M (24GB) across all four cards. The reason is the 3090 and 4090 don't support NVFP4 (only the 5090 and DGX Spark could use it). Keeping the same quant everywhere felt like the fairest way to compare. And yes, you can absolutely squeeze more performance out of every card with vLLM, SGLang, or TensorRT-LLM. But that's not what this test is about. This is just the out-of-the-box experience for folks who own a GPU and want to try the new model tonight.

stevibe

398,902 次观看 • 3 个月前

🚨| Sir Lewis Hamilton's post race interview. "Good result, amazing result for the team. Third and fourth, good points." "I went forwards even after a bad start, so lots and lots of positives to take from it. I think there's, yeah, naturally I want to be further up, so just got to keep working at it." "Considering we've not been upgrading the car for us to be kind of in the fight, it shows that the team's done a really, really great job this weekend. I hope everyone back at the factory is happy. And yeah, we're definitely on the right path. I'm so close, still not with that podium, but I'll keep trying." - F1 2025 USGP -

🚨| Sir Lewis Hamilton's post race interview. "Good result, amazing result for the team. Third and fourth, good points." "I went forwards even after a bad start, so lots and lots of positives to take from it. I think there's, yeah, naturally I want to be further up, so just got to keep working at it." "Considering we've not been upgrading the car for us to be kind of in the fight, it shows that the team's done a really, really great job this weekend. I hope everyone back at the factory is happy. And yeah, we're definitely on the right path. I'm so close, still not with that podium, but I'll keep trying." - F1 2025 USGP -

sim

23,376 次观看 • 9 个月前

"Its (Sonnet 5) performance is close to Opus 4.8, at lower prices." So I ran 4 canvas test through both. > Opus 4.8, 4/4 actually animating. > Sonnet 5, 2/4 came back as static images. And "lower price"? On the paper shredder task, Sonnet 5 spent $0.36 for a static image. Opus 4.8 spent $0.18 and it actually animated. The 4 tests: > Win 98 drag-to-BSOD > Self-typing keyboard + CRT > Letter burning > Paper shredder

"Its (Sonnet 5) performance is close to Opus 4.8, at lower prices." So I ran 4 canvas test through both. > Opus 4.8, 4/4 actually animating. > Sonnet 5, 2/4 came back as static images. And "lower price"? On the paper shredder task, Sonnet 5 spent $0.36 for a static image. Opus 4.8 spent $0.18 and it actually animated. The 4 tests: > Win 98 drag-to-BSOD > Self-typing keyboard + CRT > Letter burning > Paper shredder

stevibe

60,293 次观看 • 18 天前

I unfortunately did not win, but I came very close to making it into the finals also bonus clip with my terrible mic because I was playing from a PC cafe in Japan in the middle of the night (I’m the Dr. Strange) ggs everyone, I had so much fun

I unfortunately did not win, but I came very close to making it into the finals also bonus clip with my terrible mic because I was playing from a PC cafe in Japan in the middle of the night (I’m the Dr. Strange) ggs everyone, I had so much fun

Iron Pineapple

81,918 次观看 • 2 个月前

Afroman really said “y’all raided my house, so I dropped a music video with the footage AND monetized it.” 😭 Only in America does a police raid turn into a rollout strategy. The charges got dropped… but the streams went UP. That’s not just beating the case—that’s turning it into passive income 💀📈🙌🏾 #thursdaynews

Afroman really said “y’all raided my house, so I dropped a music video with the footage AND monetized it.” 😭 Only in America does a police raid turn into a rollout strategy. The charges got dropped… but the streams went UP. That’s not just beating the case—that’s turning it into passive income 💀📈🙌🏾 #thursdaynews

SwayLaughs

27,921 次观看 • 4 个月前

Something in the Soil. I got early beta access to Seedance 2.0 from ByteDance's Dreamina AI on Saturday morning. What started as a simple motion and consistency test turned into a full-on, campy creature-feature short inspired by the 1950s B-movies I’ve always loved. I basically spent all Saturday experimenting, and this is what came out of it. We’re not at TV or feature-quality yet, but we're SO CLOSE.

Something in the Soil. I got early beta access to Seedance 2.0 from ByteDance's Dreamina AI on Saturday morning. What started as a simple motion and consistency test turned into a full-on, campy creature-feature short inspired by the 1950s B-movies I’ve always loved. I basically spent all Saturday experimenting, and this is what came out of it. We’re not at TV or feature-quality yet, but we're SO CLOSE.

Jason Zada

68,806 次观看 • 5 个月前

I am so tired of listening to this uninformed slop purely from people who never bothered engaging with the original Sonic X, or even just paying attention to the show in general. “Tragic” isn’t the word to describe it, but Chris did not have perfect loving parents. Not even close

I am so tired of listening to this uninformed slop purely from people who never bothered engaging with the original Sonic X, or even just paying attention to the show in general. “Tragic” isn’t the word to describe it, but Chris did not have perfect loving parents. Not even close

Caramel Smoothie

50,242 次观看 • 1 年前

🇩🇪 TESLA’S SELF-DRIVING JUST FLEXED IN GERMANY In a city where delivery vans park like they’re playing Tetris, Tesla’s Full Self-Driving didn’t even flinch. On the crowded streets of Düsseldorf, the car constantly had to swerve left to dodge cars poking out into its lane. And it did it with the kind of smooth confidence most humans fake during their driving test. It didn’t hesitate, didn’t jerk the wheel, and definitely didn’t clip a bumper. It just… drove. Calmly. Safely. The grand finale? It parked itself. Not “close enough” or “kind of between the lines,” but right into a real parking spot, like it meant to be there. Source: Esther Rebers

🇩🇪 TESLA’S SELF-DRIVING JUST FLEXED IN GERMANY In a city where delivery vans park like they’re playing Tetris, Tesla’s Full Self-Driving didn’t even flinch. On the crowded streets of Düsseldorf, the car constantly had to swerve left to dodge cars poking out into its lane. And it did it with the kind of smooth confidence most humans fake during their driving test. It didn’t hesitate, didn’t jerk the wheel, and definitely didn’t clip a bumper. It just… drove. Calmly. Safely. The grand finale? It parked itself. Not “close enough” or “kind of between the lines,” but right into a real parking spot, like it meant to be there. Source: Esther Rebers

Mario Nawfal

90,968 次观看 • 7 个月前

Six open-source LLMs. One sliding puzzle. A brutal test of long-horizon reasoning and tool calling. Five of them broke. One didn't. I gave each model a move_tile tool and a scrambled 3×3 board, then asked it to solve the puzzle through pure turn-by-turn reasoning. The deeper the scramble, the more brutal the search. Five runs per depth, best run kept. A model fails the round if it exceeds 6x the optimal move count. > Depth 5: Everyone solves it. Yawn. > Depth 10: GLM 5.1 melts down. 43 moves. Cut. > Depth 12: Gemma4 26B loses the plot, shuffling tiles in circles. Gone. > Depth 15: The wall. DeepSeek V4 Flash, out. DeepSeek V4 Pro, out. Gemma4, out again. GLM 5.1, out. Two survivors: Qwen3.6 35B-A3B, and Kimi K2.6 with an 11-move solve that looked like cheating. > Depth 18: Same two. Everyone else hallucinating tiles that weren't there. > Depth 22: Final boss. Kimi, flawless for five rounds, finally cracks. 81 moves. Still scrambled. DeepSeek V4 Pro limps home at 90. Qwen3.6 35B-A3B solves it in 36. The smallest model in the room. ~3B active params. Fits on a single 3090. It beat everything. Kimi was elegant. Qwen3.6 was unstoppable.

Six open-source LLMs. One sliding puzzle. A brutal test of long-horizon reasoning and tool calling. Five of them broke. One didn't. I gave each model a move_tile tool and a scrambled 3×3 board, then asked it to solve the puzzle through pure turn-by-turn reasoning. The deeper the scramble, the more brutal the search. Five runs per depth, best run kept. A model fails the round if it exceeds 6x the optimal move count. > Depth 5: Everyone solves it. Yawn. > Depth 10: GLM 5.1 melts down. 43 moves. Cut. > Depth 12: Gemma4 26B loses the plot, shuffling tiles in circles. Gone. > Depth 15: The wall. DeepSeek V4 Flash, out. DeepSeek V4 Pro, out. Gemma4, out again. GLM 5.1, out. Two survivors: Qwen3.6 35B-A3B, and Kimi K2.6 with an 11-move solve that looked like cheating. > Depth 18: Same two. Everyone else hallucinating tiles that weren't there. > Depth 22: Final boss. Kimi, flawless for five rounds, finally cracks. 81 moves. Still scrambled. DeepSeek V4 Pro limps home at 90. Qwen3.6 35B-A3B solves it in 36. The smallest model in the room. ~3B active params. Fits on a single 3090. It beat everything. Kimi was elegant. Qwen3.6 was unstoppable.

stevibe

16,407 次观看 • 2 个月前

Luke: I like doing scenes with you cus it feels like going into a boxing ring, I've noticed that actually I'm the same so I picked that up, the you know you'd rather not run lines and just do it Yerin: Just experience it at that moment, yeah Luke: With different actors in the show, I do run lines to sort of feel it through but with you (Yerin) I feel like you do it on one side and I'm on the other side and we meet to see it in the middle Yerin: Ding ding ding (boxing bell) Luke: Yeah so.. there hasn't been that much discusson, has there? Yerin: I think we just kinda wait to see what happens and that's been really nice Luke: Yeah

Luke: I like doing scenes with you cus it feels like going into a boxing ring, I've noticed that actually I'm the same so I picked that up, the you know you'd rather not run lines and just do it Yerin: Just experience it at that moment, yeah Luke: With different actors in the show, I do run lines to sort of feel it through but with you (Yerin) I feel like you do it on one side and I'm on the other side and we meet to see it in the middle Yerin: Ding ding ding (boxing bell) Luke: Yeah so.. there hasn't been that much discusson, has there? Yerin: I think we just kinda wait to see what happens and that's been really nice Luke: Yeah

lukerin archive

16,564 次观看 • 4 个月前

A 3B model just cleared a puzzle that a 1.6 TRILLION param model couldn't. You've seen this benchmark before: my sliding-puzzle test. Same Kimi & DeepSeek runs as last time. The only new thing: I dropped VibeThinker-3B in for a side-by-side. > VibeThinker → 3B > DeepSeek V4 Flash → 284B > Kimi K2.6 → 1T > DeepSeek V4 Pro → 1.6T Shuffle depths 5, 10, 12, 15, 18, 22. One wrong move scrambles the whole board, so it's pure long-chain reasoning. ✅ VibeThinker-3B: solved all six. Never lost the thread. ⚠️ The giants started cracking at depth 15: Flash, Pro, and even Kimi each blew a run, scrambling the board past the move cap. As VibeThinker was not trained for tool calling, I had it emit X and ran the move. Bigger generalist ≠ smarter.

A 3B model just cleared a puzzle that a 1.6 TRILLION param model couldn't. You've seen this benchmark before: my sliding-puzzle test. Same Kimi & DeepSeek runs as last time. The only new thing: I dropped VibeThinker-3B in for a side-by-side. > VibeThinker → 3B > DeepSeek V4 Flash → 284B > Kimi K2.6 → 1T > DeepSeek V4 Pro → 1.6T Shuffle depths 5, 10, 12, 15, 18, 22. One wrong move scrambles the whole board, so it's pure long-chain reasoning. ✅ VibeThinker-3B: solved all six. Never lost the thread. ⚠️ The giants started cracking at depth 15: Flash, Pro, and even Kimi each blew a run, scrambling the board past the move cap. As VibeThinker was not trained for tool calling, I had it emit X and ran the move. Bigger generalist ≠ smarter.

stevibe

32,215 次观看 • 25 天前

Qwen3.6 27B landed yesterday, so I ran it on 4 setups side-by-side to see how they stack up: 🔴 RTX 4090 — 45.59 tok/s, TTFT 525ms 🟢 RTX 5090 — 51.83 tok/s, TTFT 752ms ⚫️ M2 Ultra — 22.30 tok/s, TTFT 216ms 🟣 DGX Spark — 11.08 tok/s, TTFT 319ms This is a standard test: no tuning, just the out-of-the-box experience. For the NVIDIA cards I used llama.cpp with Unsloth's UD-Q4_K_XL quant. For the M2 Ultra I used MLX with Unsloth's UD-MLX-4bit quant, since MLX is the native path on Apple Silicon. Please consider this as the baseline, you can definitely squeeze more out of every one of these with fine-tuned settings.

Qwen3.6 27B landed yesterday, so I ran it on 4 setups side-by-side to see how they stack up: 🔴 RTX 4090 — 45.59 tok/s, TTFT 525ms 🟢 RTX 5090 — 51.83 tok/s, TTFT 752ms ⚫️ M2 Ultra — 22.30 tok/s, TTFT 216ms 🟣 DGX Spark — 11.08 tok/s, TTFT 319ms This is a standard test: no tuning, just the out-of-the-box experience. For the NVIDIA cards I used llama.cpp with Unsloth's UD-Q4_K_XL quant. For the M2 Ultra I used MLX with Unsloth's UD-MLX-4bit quant, since MLX is the native path on Apple Silicon. Please consider this as the baseline, you can definitely squeeze more out of every one of these with fine-tuned settings.

stevibe

104,345 次观看 • 2 个月前

Many have asked, so here is my test project showing the ~1fps lag with image tracking in visionOS's ARKit. I just anchored a simple sphere onto a well-embellished image. This red envelope 🧧 tracks 100% smoothly on iOS, but as you can see, it's not even close on the Vision Pro.

Many have asked, so here is my test project showing the ~1fps lag with image tracking in visionOS's ARKit. I just anchored a simple sphere onto a well-embellished image. This red envelope 🧧 tracks 100% smoothly on iOS, but as you can see, it's not even close on the Vision Pro.

Nate Chan

128,425 次观看 • 2 年前

Mistral AI Releases Leanstral 1.5: An Apache-2.0 Lean 4 Code Agent Model Solving 587 of 672 PutnamBench Problems Most AI theorem proving is a language model generating a proof in one shot, with a verifier bolted on at the end to check it. That's autocomplete with a grader — and Mistral just drew a clear line between that and an actual proof agent. They released Leanstral 1.5 — a 119B MoE with 6.5B active parameters, trained as a code agent that lives inside the Lean 4 compiler loop: propose a proof, read the compiler's goals and errors, refine, repeat until it compiles or the budget runs out. Verification isn't the eval here. It's the training signal. Here's what's actually interesting: → Test-time scaling behaves like a dial: PutnamBench Pass@8 climbs 44 → 244 → 493 → 587 solved as the per-attempt token budget moves 50k → 200k → 1M → 4M → 587/672 on PutnamBench at ~$4 per problem, versus an estimated $300+ for Seed-Prover 1.5 high (a 10 H20-days-per-problem budget) → Saturates miniF2F: 100% on both validation and test sets → Two RL environments in training — a multiturn prover, and a raw-filesystem code agent that edits files, runs bash, and queries the Lean language server for live goals and types → Not just math: an Aeneas (Rust → Lean) pipeline flagged 11 genuine bugs across 57 repos, 5 previously unreported — including an integer overflow in datrs/varinteger when (value + 1) hits Std.U64.MAX Apache 2.0 weights, free API endpoint Full analysis: Model weights: Project: Technical Details: Mistral AI Mistral AI for Developers Sophia Yang, Ph.D.

Mistral AI Releases Leanstral 1.5: An Apache-2.0 Lean 4 Code Agent Model Solving 587 of 672 PutnamBench Problems Most AI theorem proving is a language model generating a proof in one shot, with a verifier bolted on at the end to check it. That's autocomplete with a grader — and Mistral just drew a clear line between that and an actual proof agent. They released Leanstral 1.5 — a 119B MoE with 6.5B active parameters, trained as a code agent that lives inside the Lean 4 compiler loop: propose a proof, read the compiler's goals and errors, refine, repeat until it compiles or the budget runs out. Verification isn't the eval here. It's the training signal. Here's what's actually interesting: → Test-time scaling behaves like a dial: PutnamBench Pass@8 climbs 44 → 244 → 493 → 587 solved as the per-attempt token budget moves 50k → 200k → 1M → 4M → 587/672 on PutnamBench at ~$4 per problem, versus an estimated $300+ for Seed-Prover 1.5 high (a 10 H20-days-per-problem budget) → Saturates miniF2F: 100% on both validation and test sets → Two RL environments in training — a multiturn prover, and a raw-filesystem code agent that edits files, runs bash, and queries the Lean language server for live goals and types → Not just math: an Aeneas (Rust → Lean) pipeline flagged 11 genuine bugs across 57 repos, 5 previously unreported — including an integer overflow in datrs/varinteger when (value + 1) hits Std.U64.MAX Apache 2.0 weights, free API endpoint Full analysis: Model weights: Project: Technical Details: Mistral AI Mistral AI for Developers Sophia Yang, Ph.D.

Marktechpost AI

56,461 次观看 • 16 天前

Completed a first hour side-by-side comparison between Qwen3.5 27b and Qwen3.6 27b on the same 4 canvas coding tests. Running the Qwen3.6 27b FP8, vLLM. What do you think?

Completed a first hour side-by-side comparison between Qwen3.5 27b and Qwen3.6 27b on the same 4 canvas coding tests. Running the Qwen3.6 27b FP8, vLLM. What do you think?

stevibe

114,191 次观看 • 2 个月前

🦆: Normally, Keng’s family video-call each other every day. Basically, in the evening my mom will video call and be like, “Oh! Have you eaten yet?” 👤: Do you speak in Northern Thai? 🦆: Yeah, we do. 🐰: No, no, not just Northern Thai. It’s actually a level beyond that, because even Namping don’t fully understand some of it. 🦆: It’s an ethnic language. Keng is from the Tai Lue ethnic group. 👤: So they have their own language, right? 🦆: Yeah, it has its own language 🗣️: But is it close to Thai? 🦆: It’s pretty close to Northern Thai in general. 🗣️: Let Namping say one sentence that’s different 🐰 *speaks in Central Thai*: Today I didn’t go anywhere at all, I only went to work (Then 🐰 said the same sentence but in Northern Thai. Next, Keng said that sentence but in Tai Lue langue- his ethnic language) — as a language enthusiast, thank you my kings #TheDriverxKengNamping #เก่งน้ำปิง

🦆: Normally, Keng’s family video-call each other every day. Basically, in the evening my mom will video call and be like, “Oh! Have you eaten yet?” 👤: Do you speak in Northern Thai? 🦆: Yeah, we do. 🐰: No, no, not just Northern Thai. It’s actually a level beyond that, because even Namping don’t fully understand some of it. 🦆: It’s an ethnic language. Keng is from the Tai Lue ethnic group. 👤: So they have their own language, right? 🦆: Yeah, it has its own language 🗣️: But is it close to Thai? 🦆: It’s pretty close to Northern Thai in general. 🗣️: Let Namping say one sentence that’s different 🐰 speaks in Central Thai: Today I didn’t go anywhere at all, I only went to work (Then 🐰 said the same sentence but in Northern Thai. Next, Keng said that sentence but in Tai Lue langue- his ethnic language) — as a language enthusiast, thank you my kings #TheDriverxKengNamping #เก่งน้ำปิง

ella⁷ ( •̯́ ₃ •̯̀) knp forever🐰🎀

50,762 次观看 • 6 个月前

🐹 once, i was lying down in my bedroom at the dorm. my door was close but there was some faint smell coming from somewhere. so i opened my room and hao hyung was there eating this yellow thing with seeds. since it looked so pretty, i didn't think the smell would come from it. i thought something went wrong with the laundry so i was about to run it again when i realized the smell came from that fruit. 🎻 at first, it smelled so much that hanbin told me to eat it outside at the veranda that time. but i just eat it quickly now. 👤 do you hold back from eating durian because you don't want to eat outside? 🎻 i'd just go out pretending to be pitiful. i'll take my clothes off so i'd end up shivering as i eat. and then he'd see that and go, 'oh... you look so pitiful... come in here and eat.' so i'll go back inside in a good mood. 🐹 but i can't go through that again so i'll be like, 'i'm really sorry but please wear your clothes and eat outside again.' 🎻 so i'll lay a mat and eat outside at the veranda floor~ #성한빈 #장하오 ZEROBASEONE

🐹 once, i was lying down in my bedroom at the dorm. my door was close but there was some faint smell coming from somewhere. so i opened my room and hao hyung was there eating this yellow thing with seeds. since it looked so pretty, i didn't think the smell would come from it. i thought something went wrong with the laundry so i was about to run it again when i realized the smell came from that fruit. 🎻 at first, it smelled so much that hanbin told me to eat it outside at the veranda that time. but i just eat it quickly now. 👤 do you hold back from eating durian because you don't want to eat outside? 🎻 i'd just go out pretending to be pitiful. i'll take my clothes off so i'd end up shivering as i eat. and then he'd see that and go, 'oh... you look so pitiful... come in here and eat.' so i'll go back inside in a good mood. 🐹 but i can't go through that again so i'll be like, 'i'm really sorry but please wear your clothes and eat outside again.' 🎻 so i'll lay a mat and eat outside at the veranda floor~ #성한빈 #장하오 ZEROBASEONE

디어

282,588 次观看 • 1 年前