@antononwaves

Just chill and serf.🏄‍♂️ Relax dude.

Dubai
Joined January 2025
Anton Oberhuber retweeted
Anduril Unveils Steelfin Family Of Modular Surface-To-Air Missiles Anduril aims to produce Steelfin interceptors at scale by leveraging common components that can be configured for different threats. twz.com/land/anduril-unveils…
10
80
4
560
80,939
Anton Oberhuber retweeted
Xiaomi released MiMo-V2.6 on Hugging Face An omni-modal family scaling RL compute toward self-improvement, with open training dynamics, RL environments, and RL framework.
1
7
36
2,433
Anton Oberhuber retweeted
🚨 CLAUDE vs CLAUDE CODE vs CLAUDE COWORK — WHICH ONE SHOULD YOU ACTUALLY USE? Most people think Claude is just an AI chatbot. But the Claude ecosystem offers different ways to write, reason, build software, and automate everyday work. The secret is knowing which tool to use, when to use it, and what it's actually designed to do. Here’s the complete breakdown. 👇
23
59
133
5,272
Anton Oberhuber retweeted
AI coding tools still write SwiftUI like it's 2021. Paul Hudson, the Hacking with Swift guy, released SwiftUI Pro, a free agent skill that fixes that. what it catches: • deprecated APIs → swapped for modern ones • accessibility → buttons VoiceOver can actually see • performance → hidden slowdowns flagged • design → layout, navigation, animations, state how to use it: → works with Claude Code, Codex, Gemini and Cursor → one command to install → run /swiftui-pro on your code Better iOS code, from the teacher half of iOS devs learned from. the repo: github.com/twostraws/SwiftUI…
23
13
105
53,144
Anton Oberhuber retweeted
Anthropic engineer just dropped a 2-hour workshop on "Graph Engineering" for agentic systems: "80% of our engineers are using self-improving loops. Now everyone is building agentic graphs." 04:43 - RAG and graphs, the basics 26:30 - nodes and edges explained 01:09:11 - how agents index data 01:30:50 - three graph layers for agents 01:48:15 - adaptive RAG that self-verifies 02:16:37 - where graph engineering goes next Prompts → Loops → Agents → Graphs most people add one more agent and call it a system she draws the graph once and every run after starts smarter same model, same tokens, completely different week watch it today then read how to become a graph engineer in the article below, save it while it is still early ↓
1
13
1
99
6,992
Anton Oberhuber retweeted
Google just shipped a meeting notetaker that runs 100% on your Mac. No cloud, no subscription It listens to your calls, expands your bullet points into full notes, and answers questions from your own files Catch: Google itself calls it experimental, and it's aimed at developers first Try it, but don't cancel your paid notetaker yet
This is INSANE. Google just dropped a free AI notetaker app for Mac. Every premium feature works without any subscription. 100% free and fully local, powered by Embedding Gemma 2.
4
2
17
887
Anton Oberhuber retweeted
🚨 MICROSOFT IS EXPERIMENTING WITH AI THAT BUILDS, TESTS, AND IMPROVES APPS. Most AI coding tools follow a familiar loop: Prompt → Code → Done. But what if the app isn't actually solving the user's problem? Microsoft Apeiron explores a different approach. It uses AI agents to build an app, simulate different user personas, observe their interactions, and feed the results back into another improvement cycle. Here's what makes it interesting: → AI agents synthesize application code → Simulated users interact with the app → An activity tracer connects behavior to code → Agents use that evidence to identify improvements → A locality controller limits how much code changes And the research gets interesting. The paper evaluates Apeiron across: • 300 app scenarios • 2,400 simulated personas • 46,338 user demands The reported results include a 64.7% improvement in task scores during its optimization process. It also reports recovering 96.9% of a previous mean simulated-user rating after a demand shift, with less than 30% of code changed. Think about the implication. The future of AI coding might not be about generating code faster. It might be about building software that gets better through feedback. One important caveat: Apeiron is a research preview, not a production-ready self-improving AI. Its agents improve the target application, not their own underlying system. Still, this is a fascinating direction for AI software engineering. Would you trust an AI to improve your app based on simulated users? #AI #Coding #AIAgents #Microsoft nitter.cf/vicky_grok/status/2107…
13
9
1
46
2,001
Anton Oberhuber retweeted
Anthropic engineer: "At Anthropic we don't write prompts anymore. We build loops and graphs." In 28 minutes she breaks down how Anthropic builds agents that remember, fix their own mistakes and get sharper with every run Prompts → Tools → Loops → Graphs a loop sends every mistake back to the step that made it a graph turns what worked into a rule the next run starts from most people stop at the loop and wonder why week two looks exactly like week one watch it today then save the full graph engineering guide below before everyone catches up ↓
7
26
1
144
22,306
Anton Oberhuber retweeted
Opus 5.5 is f*cking insane for motion design 20 tools to build your own motion design studio with Opus 5.5 product demos. launch videos. animated explainers. interactive UI. build the video, tune the movement, shape the scene, add sound, ship. 01 Remotion ▸ turn React components into editable videos remotion.dev/docs/ 02 Remotion Skills ▸ give your coding agent the official video workflow github.com/remotion-dev/skil… 03 Motion Canvas ▸ build animated explainers with TypeScript github.com/motion-canvas/mot… 04 Motion ▸ animate layouts, gestures and springs motion.dev/docs 05 GSAP ▸ coordinate complex sequences with timelines gsap.com/docs/v3/ 06 Anime.js ▸ animate text, SVG and interface elements animejs.com/documentation/ 07 Rive ▸ build characters that react to user input rive.app/ 08 Animations .dev ▸ study timing, easing and interaction details (paid course) animations.dev/ 09 Carbon Motion ▸ plan movement around hierarchy and purpose carbondesignsystem.com/build… 10 Easing Wizard ▸ preview curves and tune the feel of a transition easingwizard.com/ 11 Google Fonts ▸ choose typefaces for titles and readable labels fonts.google.com/ 12 Three.js ▸ build custom 3D scenes, lighting and cameras threejs.org/ 13 React Three Fiber ▸ compose reusable 3D scenes in React r3f.docs.pmnd.rs/getting-sta… 14 Spline ▸ design interactive 3D scenes visually spline.design/ 15 Poly Haven ▸ get CC0 models, materials and HDRIs polyhaven.com/ 16 Blender ▸ model assets, animate objects and render scenes blender.org/ 17 SVGOMG ▸ clean vector assets before animating them jakearchibald.github.io/svgo… 18 Pixabay Music ▸ find licensed music and sound effects pixabay.com/music/ 19 Mediabunny ▸ read, write and convert media in the browser mediabunny.dev/ 20 FFmpeg ▸ encode, resize and prepare the final export ffmpeg.org/documentation.htm… the loop: write the brief → choose the type → map the keyframes → build the scene → add sound → review the render → ship start with one product workflow. give every movement a job. save this, then read the full guide for Opus 5.5 Motion Engineering ↓
Opus 5.5 can run a whole motion design studio Sell a 20-second launch clip built from real product screens Copy the motion engineering prompts and delivery process Build your offer before the next product launch ↓
27
55
1
333
14,987
Anton Oberhuber retweeted
Flights bound for Dammam‚ including one that was originally scheduled to Riyadh, are currently holding away from the airfield. flightradar24.com/airport/dm…
3
65
9
421
198,269
Anton Oberhuber retweeted
Anthropic pays $650,000 a year for this exact stack. Ten courses. All free. This is the interview, written out. math → the network by hand → language → the model from scratch → code-first deep learning → fine-tune → agents → tools → evals → the system you ship. 01 Math, visual. 3Blue1Brown. Essence of Linear Algebra ▸ youtube.com/playlist?list=PL… 02 The network, by hand. Andrej Karpathy. Zero to Hero ▸ karpathy.ai/zero-to-hero.htm… 03 Language. Stanford CS224n. Chris Manning ▸ web.stanford.edu/class/cs224… 04 The model, from scratch. Stanford CS336. ▸ cs336.stanford.edu/ 05 Code first. fast. ai. Practical Deep Learning for Coders ▸ course.fast.ai/ 06 Fine-tune. Hugging Face LLM Course ▸ huggingface.co/learn/llm-cou… 07 Agents. Hugging Face Agents Course ▸ huggingface.co/learn/agents-… 08 Tools. Microsoft. AI Agents for Beginners ▸ github.com/microsoft/ai-agen… 09 Evals. Anthropic courses. Prompting, tool use, evaluations ▸ github.com/anthropics/course… 10 Ship it. Full Stack Deep Learning. LLM Bootcamp ▸ fullstackdeeplearning.com/ll… save this ⭣
12
54
1
227
7,512
Anton Oberhuber retweeted
Grok sees everything on X in real time, Opus 5.5 thinks before it writes almost nobody runs them together this is f*cking insane... i wired them into one 10-step agent harness: Grok scouts, Opus verifies and builds, i just press "yes." it turned Claude into a full research + content studio send it to Claude, and thank me later then read the full roadmap with examples below
Grok Bot just got a serious brain upgrade With Claude Opus 5.5 entering the stack, one Bot could run an entire AI team that researches, writes code, uses tools, verifies results, and delivers finished work Here's how to build it in 10 steps:
18
32
1
274
28,825
Anton Oberhuber retweeted
🇨🇳 China wastes 1/4 of its wind and solar power 24% of potential wind and solar output went unused in 2026 so far, up from 9% in 2024 and 4% in 2022, per CREA. That is 401 TWh, close to what 🇪🇸 Spain and 🇵🇭 the Philippines generated together last year. ⚡ The problem is geography. The biggest hubs sit in Inner Mongolia, Xinjiang and Gansu, far from the coastal cities that use the power, and transmission hasn’t kept up. 🏭 Coal adds to it. Coal plants still sign annual contracts covering 60–70% of the previous year’s output, so they keep running while wind and solar get curtailed. ⚠️ The open question is whether grid lines and storage can be built as fast as new panels and turbines. Source: CREA via Bloomberg (@SStapczynski )
1
16
1
46
3,607
Anton Oberhuber retweeted
10 GITHUB REPOS THAT TEACH AI TO SEE, HEAR AND UNDERSTAND THE WORLD. 1. Qwen2-VL — powerful vision-language AI. Give it images, documents or videos and let it reason about what it sees. github.com/QwenLM/Qwen2-VL 2. LLaVA — connect vision with language models. Turn an LLM into an AI that can actually look at images and talk about them. github.com/haotian-liu/LLaVA 3. Florence-2 — compact vision foundation model. Use one model for captioning, detection, segmentation and other vision tasks. github.com/microsoft/Florenc… 4. Grounding DINO — find objects from text prompts. Tell the model what you're looking for and it can locate those objects inside images. github.com/IDEA-Research/Gro… 5. SAM 2 — segment almost anything. Give it an image or video and isolate objects with surprisingly precise masks. github.com/facebookresearch/… 6. Whisper — turn speech into text. A powerful open-source speech recognition model that can understand multiple languages. github.com/openai/whisper 7. ImageBind — connect different types of data. Bring images, text, audio, depth, thermal and motion into one shared embedding space. github.com/facebookresearch/… 8. Open-Sora — open-source AI video generation. Generate videos from text and explore how modern video models work under the hood. github.com/hpcaitech/Open-So… 9. AudioCraft — generate and understand audio with AI. Experiment with music generation, audio synthesis and other generative audio models. github.com/facebookresearch/… 10. PaddleOCR — extract intelligence from images. Turn screenshots, scanned documents and real-world images into structured text. github.com/PaddlePaddle/Padd… All are open-source. Bookmark this. AI is no longer just reading text — it's learning to understand the world.
4
23
65
4,318
Anton Oberhuber retweeted
sonnet 5.5 is f**king insane send this to Claude Sonnet 5.5 and watch it design your whole one-person company: the folder, the roles, the weekly schedule. Steal this before everyone starts using it:
1
20
57
3,591
Anton Oberhuber retweeted
GPT 6 Astra needs a f…cking operating system around it 8 GitHub repos to build the stack around one model 01 OpenAI Agents SDK ▸ github.com/openai/openai-age… → agents + handoffs + guardrails + tracing MEMORY 02 Mem0 ▸ github.com/mem0ai/mem0 → persistent memory across sessions 03 Graphiti ▸ github.com/getzep/graphiti → temporal knowledge for changing facts TOOLS 04 Composio ▸ github.com/ComposioHQ/compos… → 1000+ toolkits + authentication + actions EXECUTION 05 E2B ▸ github.com/e2b-dev/E2B → isolated environments for agent code EVALS 06 DeepEval ▸ github.com/confident-ai/deep… → test complete agent trajectories OBSERVABILITY 07 Langfuse ▸ github.com/langfuse/langfuse → traces + evals + production debugging CONTEXT 08 LlamaIndex ▸ github.com/run-llama/llama_i… → retrieval + context pipelines the architecture: input → context → agent → handoff → memory → knowledge → tools → sandbox → eval → trace → retry the model decides how smart the system can become the architecture decides whether it can actually work
20
14
89
4,629
Anton Oberhuber retweeted
Andrej Karpathy (ex-OpenAI, ex-Tesla) just released a prompt that turns Opus 5.5 into a 24/7 quant research desk that actually researches markets instead of just making shit up It’s f*cking unreal... send it to Opus 5.5 and let it build the entire research pipeline. then read the full guide for 24/7 Quant Trading Desk with Opus 5.5
Grok Bot + Opus 5.5 is godmode for building your own 24/7 AI quant desk. 1M token context. always-on reasoning. wire it into market data and backtesting to research new strategies. set it up correctly and your quant desk keeps working while you sleep
15
50
261
26,286
Anton Oberhuber retweeted
Anthropic's engineer, Andrej Karpathy (ex-Tesla) just shared ASD-STE100 prompting system almost nobody uses this is f*cking insane... i applied it to my Opus 5.5 & Fable 5.5 motion prompt and it turned Claude into a premier motion studio. send it to Claude, and thank me later. Then read the full roadmap with examples below.
10
15
139
21,742
Anton Oberhuber retweeted
This is f***king gold A FACELESS YOUTUBE CHANNEL THAT’S ONLY 12 DAYS OLD IS ALREADY ON TRACK TO MAKE $4K IN ITS FIRST MONTH. The first video pulled 272K views. The second added another 136K. One viral channel could just be luck. Then the creator’s student launched a completely separate faceless channel using the same production system. Just 13 days later, he’s got; 10K+ subscribers One AI generated video made $500+ Neither channel started with an existing audience, apersonal brand, or a huge content library AI handled much of the production, while the creators focused on repeating a proven format and recognizable packaging instead of reinventing every upload. One channel going viral is interesting. Two different channels getting similar results with the same system? That’s when it starts looking less like luck and more like a repeatable process.
23
137
2
623
36,384
Anton Oberhuber retweeted
lauren tan just f*cking lost it and built a prompt for a whole team of grok bots this is absolutely insane. one prompt, and your grok bot turns into a team try feeding it to your own grok bot and thank me later 👇
22
86
5
707
62,477