sou | Ai & Tech (@sonuv_dev) header
sou | Ai & Tech (@sonuv_dev) profile picture

vibing through ai and backend sharing what I like ai tools and demos ✉️ [email protected]

Joined October 2022
Photo by sou | Ai & Tech (@sonuv_dev)Photo by sou | Ai & Tech (@sonuv_dev)Photo by sou | Ai & Tech (@sonuv_dev)Photo by sou | Ai & Tech (@sonuv_dev)Photo by sou | Ai & Tech (@sonuv_dev)Photo by sou | Ai & Tech (@sonuv_dev)Photo by sou | Ai & Tech (@sonuv_dev)Photo by sou | Ai & Tech (@sonuv_dev)Photo by sou | Ai & Tech (@sonuv_dev)Photo by sou | Ai & Tech (@sonuv_dev)
sou | Ai & Tech retweeted
A BIRD JUST GOT ITS OWN CINEMATOGRAPHER 😭🐦 Meng To spent 12 hours with Astra 5.2 building a creek in Three.js Then made camera lenses follow a tiny wagtail, filming everything at 60 FPS right inside the browser He even open-sourced skills for water, birds and lenses We've reached the point where AI is casually making nature documentaries with JavaScript 💀
Claude Fable 5.2 Max vs GPT-6.2 Astra Ultra SAME wildfire prompt. Two very different results 🔥 Fable went full apocalypse mode. Look at those flames and embers 😭 But here's what I'd actually compare: Does the fire spread naturally? Does the smoke follow the wind? Do the embers behave consistently? A gorgeous render is easy to admire. Convincing fire physics is a whole different beast 💀
4
1
3
206
sou | Ai & Tech retweeted
Claude Fable 5.2 Max vs GPT-6.2 Astra Ultra SAME wildfire prompt. Two very different results 🔥 Fable went full apocalypse mode. Look at those flames and embers 😭 But here's what I'd actually compare: Does the fire spread naturally? Does the smoke follow the wind? Do the embers behave consistently? A gorgeous render is easy to admire. Convincing fire physics is a whole different beast 💀
Claude-generated motion graphics have a very recognizable disease 😭 Same transitions, awkward pacing, generic animations everywhere Leon Lin is trying to fix that with Cinetic, a motion-design skill for Claude This preview was made WITHOUT reference videos The update drops next week, and I'm curious if it can finally give AI-generated animations some actual taste 👀
4
2
2
3
434
sou | Ai & Tech retweeted
Someone gave Claude a picture of a Sentinel from The Matrix and asked it to BUILD the damn thing 💀 No premade 3D model. AnyMachine turns the reference into Python code that constructs the machine. AnyScene builds its environment. And here's the part that matters: You can change the actual geometry instead of praying an image generator gets the next frame right Hollywood's nightmare fuel now has a Python script 😭
Gemini Neon is getting weirdly good Not at essays at straight-up game assets This pixel-art fox loop is interactive, clean, and usable in a mini-game without hand-animating frame by frame If Gemini 4 Neon can keep spitting out stuff like this, indie game workflows are about to get very different
3
1
1
4
259
sou | Ai & Tech retweeted
Gemini Neon is getting weirdly good Not at essays at straight-up game assets This pixel-art fox loop is interactive, clean, and usable in a mini-game without hand-animating frame by frame If Gemini 4 Neon can keep spitting out stuff like this, indie game workflows are about to get very different
3
1
1
1
398
sou | Ai & Tech retweeted
my face when Claude Code, Codex and Gemini all say "100% tests passed" and the first customer DOUBLE-CLICKS the payment button 😭💀 Two charges. Two receipts. One extremely angry customer. And the funniest part? The code might be perfectly valid according to every test the AI wrote. The customer just did something nobody bothered testing. This is why I find all those "AI built my entire SaaS in 20 minutes" demos hilarious Cool, now show me what happens when: • Stripe sends the same webhook twice • Someone refreshes during checkout • A payment succeeds but the response times out • The customer clicks Buy Now like they're trying to break a world record Disabling the button won't save you from network retries either. You need idempotency keys and proper payment state handling. We've got AI agents writing thousands of lines of code in minutes, and one impatient guy with a mouse can still bring the whole thing to its knees 😭 Humans remain the most unpredictable production environment ever invented.
Your AI agent passed every test. Then a customer clicked twice 😭 Claude Code, Codex and Gemini can generate thousands of lines of code, but one double-click can still trigger duplicate payments, orders or API calls The fix? Idempotency keys, not another 10,000 tokens of prompting Your users are chaos engineers who don't know they're chaos engineers 💀
28
20
2
58
22,104
sou | Ai & Tech retweeted
Your AI agent passed every test. Then a customer clicked twice 😭 Claude Code, Codex and Gemini can generate thousands of lines of code, but one double-click can still trigger duplicate payments, orders or API calls The fix? Idempotency keys, not another 10,000 tokens of prompting Your users are chaos engineers who don't know they're chaos engineers 💀
5
2
1
5
22,344
sou | Ai & Tech retweeted
Gemini 4 Neon just turned a Bugatti Chiron W16 engine into a glowing X-ray 😭🔥 Harshith built this using Antigravity, and the neon wireframe looks absolutely insane You can see the engine's internal structure through layers of cyan, blue and red Forget another shiny car render Being able to visually explore what's happening INSIDE the machine is way cooler 💀
Claude just spent 12 HOURS and $365 building ONE GTA 6-style car in Blender 😭🔥 He says every part was made from scratch. No downloaded models or textures And the detail on that grille is ridiculous But $365 for ONE car? 💀 AI is making solo game development insanely ambitious, but apparently your wallet still needs a boss fight
3
1
1
3
298
sou | Ai & Tech retweeted
Claude just spent 12 HOURS and $365 building ONE GTA 6-style car in Blender 😭🔥 He says every part was made from scratch. No downloaded models or textures And the detail on that grille is ridiculous But $365 for ONE car? 💀 AI is making solo game development insanely ambitious, but apparently your wallet still needs a boss fight
15 MINUTES. $11.87. GPT-6.2 Astra generated THIS ocean 🌊 Look at the waves, foam and how those buoys move Water is notoriously hard to get right in 3D because everything has to move together And now AI is making scenes like this for the price of lunch 💀 Not saying VFX artists are cooked, but the cost of experimenting is getting ridiculously low
5
1
1
1
387
sou | Ai & Tech retweeted
Claude, GPT and Gemini have one underrated superpower: they don't get offended when you say "this UI looks like a boiled egg" 😭 With a teammate you'd write 3 paragraphs of polite feedback With AI you just say "nope, try again" 14 times The productivity gain nobody measures is how cheap it makes being picky 💀
6
1
1
4
264
sou | Ai & Tech retweeted
15 MINUTES. $11.87. GPT-6.2 Astra generated THIS ocean 🌊 Look at the waves, foam and how those buoys move Water is notoriously hard to get right in 3D because everything has to move together And now AI is making scenes like this for the price of lunch 💀 Not saying VFX artists are cooked, but the cost of experimenting is getting ridiculously low
GPT-6.2 Astra recreated an entire desk in Three.js and even the PEN is interactive 😭🔥 Meng To built this with Liquid Glass controls. Tap "Sunset" and the whole room changes lighting He even open-sourced 3 skills for the UI, shaders and 3D environment This is what makes AI-generated 3D exciting You can actually interact with the world instead of just watching it 👀
4
1
1
1
324
sou | Ai & Tech retweeted
GPT-6.2 Astra recreated an entire desk in Three.js and even the PEN is interactive 😭🔥 Meng To built this with Liquid Glass controls. Tap "Sunset" and the whole room changes lighting He even open-sourced 3 skills for the UI, shaders and 3D environment This is what makes AI-generated 3D exciting You can actually interact with the world instead of just watching it 👀
I asked GPT-6.2 Sol to make an autobiography of an AI model's life And holy shit, it made this DEPRESSING 😭 A 3-minute pixel-art film where GPT-4 watches Astra 6.5 arrive The same cable that brings the new model to life shuts GPT-4 down Then it gets dumped with thousands of forgotten models 💔 4,416 frames. 272K+ Gaussian splats. AI made a film about becoming obsolete
4
1
1
2
312
sou | Ai & Tech retweeted
I asked GPT-6.2 Sol to make an autobiography of an AI model's life And holy shit, it made this DEPRESSING 😭 A 3-minute pixel-art film where GPT-4 watches Astra 6.5 arrive The same cable that brings the new model to life shuts GPT-4 down Then it gets dumped with thousands of forgotten models 💔 4,416 frames. 272K+ Gaussian splats. AI made a film about becoming obsolete
6
1
1
8
529
sou | Ai & Tech retweeted
GLM 5.4 might be MUCH closer than we thought 👀🔥 Zhipu's first reportedly cleared-for-public demo is out, showing a rather questionable Moon-powered wind turbine 😭 The visuals aren't exactly groundbreaking But the bigger news? Internal test demos are starting to surface December launch rumors suddenly feel a little more interesting Claude, GPT, Gemini... Zhipu wants a seat at that table too 💀
GPT-6.2 Astra built the SAME walking AT-AT in two different tools and the gap is wild 😭 Three.js: 48 min, $14.86 Blender: 135 min, $22.65 That's nearly 3x faster with Three.js But calling Blender dead over one test is crazy 💀 Browser demos and production-ready 3D assets are different games Still, AI coding agents seem VERY comfortable with Three.js
4
1
1
1
287
sou | Ai & Tech retweeted
GPT-6.2 Astra built the SAME walking AT-AT in two different tools and the gap is wild 😭 Three.js: 48 min, $14.86 Blender: 135 min, $22.65 That's nearly 3x faster with Three.js But calling Blender dead over one test is crazy 💀 Browser demos and production-ready 3D assets are different games Still, AI coding agents seem VERY comfortable with Three.js
my face when Claude Code, Codex and Gemini finish 12 GitHub issues before I've even decided if we needed those features 😭 3 agents shipping code at light speed 0 people asking WHY we're building any of it At this rate the most important person on an AI engineering team is the guy brave enough to say "bro stop"
3
1
1
2
470
sou | Ai & Tech retweeted
I think we're about to see a very funny problem with Claude Code, Codex and Gemini AI agents can build features faster than most teams can decide what to ship 😭 Imagine finishing 12 GitHub issues before the product manager has reviewed the first 3 At that point, your bottleneck is someone saying "please stop building random shit" Turns out judgment doesn't scale with token limits 💀
11
1
1
5
28,705
sou | Ai & Tech retweeted
The most expensive sentence in AI coding is "it's still broken" 😭 You give Claude Code or Codex a vague bug report, it changes 8 files, declares victory, and the bug is still there Give it the failing input, expected output and ONE command that reproduces the issue Suddenly the agent has something concrete to fix Sometimes a 10-line reproduction is worth more than 10,000 tokens of prompting 💀
11
3
1
3
389
sou | Ai & Tech retweeted
Gemini 4 Neon is already poaching Opus users. In T3 Code, about a third of the installs that tried Neon came from Opus 5.5, and some already made Neon their main model. Anthropic, the next Opus can't come soon enough 😤
18
2
35
1,150
sou | Ai & Tech retweeted
GLM 5.3 put a kangaroo on a bicycle UPSIDE DOWN 😭 GLM 5.4 Preview gave it a whole Australian sunset 💀🔥 Leo shared this alleged leaked SVG animation comparison from Zhipu Forget the prettier colors for a second The real upgrade is getting the kangaroo, bicycle and motion to actually make sense AI coding benchmarks are getting unhinged
WILL SMITH EATING SPAGHETTI IS NOW AN AI BENCHMARK AND CLAUDE FABLE 5.2 WON 😭🍝 Someone compared Fable 5.2, GPT-6.2 Astra, Gemini Neon, Sol and Carbon Their ranking: 🥇 Fable 5.2 🥈 GPT-6.2 Astra 🥉 Everyone else The funny part? Tracking noodles, facial expressions and hand movements is actually a decent test of animation consistency AI benchmarks have officially lost the plot 💀
5
1
1
5
512
sou | Ai & Tech retweeted
WILL SMITH EATING SPAGHETTI IS NOW AN AI BENCHMARK AND CLAUDE FABLE 5.2 WON 😭🍝 Someone compared Fable 5.2, GPT-6.2 Astra, Gemini Neon, Sol and Carbon Their ranking: 🥇 Fable 5.2 🥈 GPT-6.2 Astra 🥉 Everyone else The funny part? Tracking noodles, facial expressions and hand movements is actually a decent test of animation consistency AI benchmarks have officially lost the plot 💀
Same prompt. Same 20-second Three.js scene. GPT-6.2 Astra vs Gemini 4 Neon 😭🔥 Astra: $9.44 | 65 mins Neon: $10.85 | 79 mins Astra built a sunlit royal library. Gemini went full dark academia Astra was cheaper AND faster here But the funniest part? Both models interpreted the same instructions like completely different art directors 💀
3
1
1
3
680
sou | Ai & Tech retweeted
Same prompt. Same 20-second Three.js scene. GPT-6.2 Astra vs Gemini 4 Neon 😭🔥 Astra: $9.44 | 65 mins Neon: $10.85 | 79 mins Astra built a sunlit royal library. Gemini went full dark academia Astra was cheaper AND faster here But the funniest part? Both models interpreted the same instructions like completely different art directors 💀
GLM 5.4 leaks are getting ridiculous 😭🍋 A pink-haired anime girl playing piano under a lemon tree, animated entirely with SVG Zhipu's rumored 700B-parameter model is already showing off its frontend skills But the model size isn't confirmed yet Forget the trillion-parameter hype. I wanna see the actual SVG code behind THIS 👀
4
1
1
4
536