Jason @pythonagents

Andrej Karpathy spent 2h showing how he actually uses AI day to day he's a co-founder of OpenAI and led AI at Tesla, so when he shows how he works, it’s worth watching and the whole session is just him telling the machine what he wants in simple terms, like he's briefing a coworker watch what's actually happening the entire time: > he describes the task in normal words > it goes off and does the work > he glances at the result and nudges it with one more sentence that's the whole skill, and you've had it since you learned to talk the only gap between that and a worker that runs on its own is handing that sentence a schedule and the tools to act check his work, then build the version that keeps working when you stop

Chrome @0xchromium

5 days ago

x.com/i/article/2061…

9 83 651 1.7M 3K

It's WILD how much value is in this video, but only 1,300 views??

Hermes Desktop Just Changed Everything Hermes just became way easier to use. No terminal, no complicated setup, no command-line stress. The new Hermes desktop app gives you a visual way to run a powerful AI agent. Here’s what it does: → Installs Hermes automatically. → Lets you choose OpenRouter, Anthropic, OpenAI, or local models. → Saves every chat session. → Creates separate profiles for different workflows. → Stores persistent memory. → Gives you access to 68 built-in tools. → Supports MCP connections. The biggest unlock? Hermes can learn skills from tasks and improve the more you use it.

This is the best founder you've never heard of. Adam Foroughi's company prints $5 billion in cash a year with just 400 people. When his stock price dropped by 92%, he borrowed money to buy back $6 billion in stock. That bet made the company more than $60 billion. The entire C-suite is just 4 people. To this day, Adam approves every hire. This episode has the highest insight-to-minute ratio of any conversation I've had so far. Here’s our full conversation: 0:00 The $6B Buyback That Made $60B 2:15 Borrowing Money To Buy Back Stock At A Discount 5:02 Why VCs Passed On AppLovin In 2012 9:00 From App Discovery To Ad Platform 14:45 Beating Google's AdMob With Performance Marketing 19:30 No Board For Six Years 30:12 The China Deal That Almost Blew Up 37:45 The Convertible Note Pivot And KKR 46:30 Buying Gaming Studios To Get Data 51:45 Losing Trust With Game Developers 58:20 The 2022 Crash And How He Kept His Team 1:02:00 Building An Hyper Competent & Efficient Company 1:07:25 Why Every New Hire Needs His Approval 1:19:06 The Axon 2 Inflection Point 1:21:15 One Great Engineer Now Beats A Hundred Includes paid partnerships.

youtu.be/kB9a5nXCwkA?si…

this chinese developer making $320k/year as a solo contractor his secret: 5 AI agents running in parallel, each one a specialist architect, coder, reviewer, tester, ops they don’t share context, don’t step on each other, just ship he takes on projects meant for teams of 5-8 engineers delivers in half the time keeps the entire budget found this video on bilibili at 3am and watched it four times guy sitting at his desk, two monitors filled with code, and he’s barely touching the keyboard here’s what’s happening on his screen: > agent 1 (architect): designs system structure, breaks down features into tasks, decides what gets built first > agent 2 (coder): writes the actual implementation based on architect’s specs > agent 3 (reviewer): checks every piece of code for bugs, edge cases, security issues > agent 4 (tester): generates test cases, runs them, reports failures back > agent 5 (ops): handles deployment, monitoring, infrastructure five separate claude code instances running simultaneously each one has its own system prompt, its own context, its own specialty they communicate through a shared task queue, not through each other that’s the key insight - no shared context means no conflicts agent 2 doesn’t know what agent 3 is doing agent 4 doesn’t care what agent 1 decided they just pick up tasks, complete them, move on he showed his contract history: > 3D rendering pipeline for a gaming studio: $25k > automated trading dashboard: $33k > enterprise CRM rebuild: $44k all completed solo, all delivered early, all clients thought they were hiring a team the code on his screen is python with blender integration - complex stuff that would normally require 3-4 specialists he’s shipping it in days while the client expects weeks while he’s explaining the system to camera, commits are happening in the background, tests running, deployments going out all while he’s literally not touching the keyboard his API costs run about $2k/month his revenue averages $26k/month that’s a 13x return on his AI investment this is the new solo developer playbook don’t compete with teams become the team

regent0x @regent0x_

a month ago

x.com/i/article/2049…

41 82 725 946K 3K

🚀 DeepSeek-V4 Preview is officially live & open-sourced! Welcome to the era of cost-effective 1M context length. 🔹 DeepSeek-V4-Pro: 1.6T total / 49B active params. Performance rivaling the world's top closed-source models. 🔹 DeepSeek-V4-Flash: 284B total / 13B active params. Your fast, efficient, and economical choice. Try it now at chat.deepseek.com via Expert Mode / Instant Mode. API is updated & available today! 📄 Tech Report: huggingface.co/deepseek-ai/De… 🤗 Open Weights: huggingface.co/collections/de… 1/n

Qwen3.6-27B can now run locally! 💜 Run on 18GB RAM via Unsloth Dynamic GGUFs. Qwen3.6-27B surpasses Qwen3.5-397B-A17B on all major coding benchmarks. GGUFs: huggingface.co/unsloth/Qwen3.… Guide: unsloth.ai/docs/models/qw…

Qwen @Alibaba_Qwen

2 months ago

🚀 Meet Qwen3.6-27B, our latest dense, open-source model, packing flagship-level coding power! Yes, 27B, and Qwen3.6-27B punches way above its weight. 👇 What's new: 🧠 Outstanding agentic coding — surpasses Qwen3.5-397B-A17B across all major coding benchmarks 💡 Strong

538 2K 13K 3.7M 5K

🚀 Meet Qwen3.6-27B, our latest dense, open-source model, packing flagship-level coding power! Yes, 27B, and Qwen3.6-27B punches way above its weight. 👇 What's new: 🧠 Outstanding agentic coding — surpasses Qwen3.5-397B-A17B across all major coding benchmarks 💡 Strong reasoning across text & multimodal tasks 🔄 Supports thinking & non-thinking modes ✅ Apache 2.0 — fully open, fully yours Smaller model. Bigger results. Community's favorite. ❤️ We can't wait to see what you build with Qwen3.6-27B! 👀 🔗👇 Blog: qwen.ai/blog?id=qwen3.… Qwen Studio: chat.qwen.ai/?models=qwen3.… Github: github.com/QwenLM/Qwen3.6 Hugging Face: huggingface.co/Qwen/Qwen3.6-2… huggingface.co/Qwen/Qwen3.6-2… ModelScope: modelscope.cn/models/Qwen/Qw… modelscope.cn/models/Qwen/Qw…

this part of the KIMI K2.6 launch blog is insane: > it deployed Qwen3.5-0.8B model locally on a Mac. > coded and optimized its inference in Zig > (never knew you could do that) > improved throughput from ~15 to ~193 tokens/sec > made it 20% faster than LM Studio > did 4,000+ tool calls, >12 hours of execution, 14 iterations

Kimi.ai @Kimi_Moonshot

2 months ago

Meet Kimi K2.6: Advancing Open-Source Coding 🔹Open-source SOTA on HLE w/ tools (54.0), SWE-Bench Pro (58.6), SWE-bench Multilingual (76.7), BrowseComp (83.2), Toolathlon (50.0), Charxiv w/ python(86.7), Math Vision w/ python (93.2) What's new: 🔹Long-horizon coding - 4,000+

942 2K 18K 7.5M 8K

if you've been paying attention, you know there's been a serious vibe shift in AI this month 3 major AI labs (anthropic, openai, amazon) all just went hard into biology. they're literally telling you this is where the next wave of ai value capture is here's what happened:

do you understand what Kimi just did.. because nobody is talking about it yet.. in January, DeepSeek open-sourced a model that crashed Nvidia $600 billion in a single day.. forced Sam Altman to make ChatGPT free the same week. now a second Chinese lab just did it again.. Kimi K2.6. an AI coding agent that runs 4,000 tool calls in a single session.. 12 hours of continuous execution. 300 parallel sub-agents building entire codebases from one prompt.. open-source.. weights on HuggingFace right now. SOTA on SWE-Bench Pro. SOTA on BrowseComp. SOTA on HLE. beating every closed model behind billion-dollar paywalls... OpenAI charges $200/month for Pro. Anthropic just raised at $60 billion. Google is spending $75 billion on infrastructure this year. and two Chinese labs.. funded on a fraction of that.. keep giving away for free what American companies tell investors requires tens of billions to build. the AI arms race isn't US vs China. it's closed vs open. and closed keeps losing. this is the second time in four months. there will be a third.

Kimi.ai @Kimi_Moonshot

2 months ago

Meet Kimi K2.6: Advancing Open-Source Coding 🔹Open-source SOTA on HLE w/ tools (54.0), SWE-Bench Pro (58.6), SWE-bench Multilingual (76.7), BrowseComp (83.2), Toolathlon (50.0), Charxiv w/ python(86.7), Math Vision w/ python (93.2) What's new: 🔹Long-horizon coding - 4,000+

942 2K 18K 7.5M 8K

In 14 minutes, this Anthropic engineer who wrote "Building Effective Agents" will teach you more about building them right than most developers figure out on their own in months. Bookmark this for the weekend. Then read the builder's guide below.

Avid @Av1dlive

2 months ago

x.com/i/article/2044…

56 320 2K 4.2M 10K

🚨 This is absolute GOLD. The @AnthropicAI engineer who literally wrote "Building Effective Agents" just dropped a 14-minute masterclass. saves you months of headaches trying to figure this out alone. bookmark for the weekend + read @Av1dlive's great guide below 👇

Avid @Av1dlive

2 months ago

x.com/i/article/2044…

56 320 2K 4.2M 10K

I was in class last week. One of my students raised their hand mid-lecture and asked something that stopped me for a second. "Why is every AI tool built on Python? C++ is faster. Rust is faster. Even Java is faster. So why Python?" Honestly it’s a fair question. And the answer reveals something really interesting about how the AI industry actually works. Let me explain this properly. 🧵

A 19-year-old took apart an IBM PC and realized something that would destroy two of the world's most powerful computer companies: "The emperor had no clothes." Michael Dell opened up his IBM 5150 and found nothing made by IBM inside. Just assembled parts from other manufacturers. The same parts available to anyone. This was the insight that mattered: If IBM doesn't actually make the components, why are they charging 50-70% more than it costs to buy the same parts separately and assemble them? So Dell did the obvious thing. He bought IBM PCs at retail price. Upgraded them with better components. Sold them directly to doctors and lawyers. His first 9 months: $6 million in revenue. First full year: $33 million. IBM and Compaq called him a "mail order company" and a "garage shop operation." They weren't wrong about the garage part. But they completely missed the point. They were obsessed with their brand, their distribution channels, their retail partnerships. They thought those things were the moat. What they actually were: the tax on inefficiency. Dell had found the gap between what something cost to make and what people were charged to buy it. And he exploited it ruthlessly.

@unknown_expert2 @dsiroker Second this

What is the best local model to run on ANY computer or device? My AI answers: docs.google.com/document/d/1D0…

Gemma 4 just dropped. Here's what it actually means for builders: What it is: Google's open-source AI model. Runs on your own machine. No API. No usage bill. No data leaving your server. 4 sizes: → 31B - full reasoning, runs on a workstation → 26B MoE - fast, activates only 3.8B params per inference → E4B / E2B - runs on a phone, offline, real-time audio + vision Why the license matters: Apache 2.0. Ship it in a product, fine-tune it, modify it. No restrictions. No royalties. No legal grey area. Where to use it: → Replace OpenAI API calls in your SaaS. Every call you move saves margin. → Routine agent tasks (summarize, classify, simple replies) don't need Claude. → Sensitive data workflows where you can't send data to a cloud API. → Offline features on mobile. The number worth remembering: 256K context. Feed it a full codebase. It won't lose track. Available now on HuggingFace, Ollama, Kaggle, Google AI Studio.

Google DeepMind @GoogleDeepMind

2 months ago

Meet Gemma 4: our new family of open models you can run on your own hardware. Built for advanced reasoning and agentic workflows, we’re releasing them under an Apache 2.0 license. Here’s what’s new 🧵

373 1K 9K 4.0M 6K

let me explain what Google just did: → they’ve just released their most capable open models yet → Gemma 4… built from the same research behind Gemini 3… four sizes… all running on your own hardware → the 31B dense model and 26B mixture of experts model deliver what Google is calling “frontier-level intelligence” on a personal computer... no cloud required… your data stays on your machine → the 26B MoE only activates 3.8B parameters at a time… meaning it runs fast without needing massive compute → the 2B and 4B models are built for phones and edge devices… text, image, and audio support they can see and hear in real time… 140+ languages natively → 256K context window on the larger models… enough to analyze full codebases or handle long multi-turn agent workflows → native tool use built in… these models can plan steps and call tools on your behalf without extra wiring → Arena Elo scores: Gemma 4 31B hit 1464 and the 26B hit 1453… competing with models 20-30x their size as of today… GLM 5 at 754B scored 1469… Kimi k2.5 at 1100B scored 1464… Gemma is doing this at a fraction of the parameters → Apache 2.0 license… fully open weights, commercially permissive… and the first time Google has done this with Gemma → 400 million downloads and over 100,000 community variants since the first Gemma launched → available now on Google AI Studio, HuggingFace, Kaggle, and Ollama the open source AI race just took a massive leap forward imo running frontier-level reasoning on your laptop without sending a single byte to the cloud completely changes the game for privacy, speed, and cost and the fact that a 26B model with 3.8B active parameters is competing with models 20-30x its size tells you where this is heading running models locally? you gotta get this set up today

Google @Google

2 months ago

We just released Gemma 4 — our most intelligent open models to date. Built from the same world-class research as Gemini 3, Gemma 4 brings breakthrough intelligence directly to your own hardware for advanced reasoning and agentic workflows. Released under a commercially

729 3K 20K 7.7M 10K