ed interactive group (@edinteractivegr) header
ed interactive group (@edinteractivegr) profile picture

We help B2B firms automate to augment: AI tools, training and coaching that free your people for better thinking. Human-led · AI-enhanced.

Ste-Marg-du-lac-Masson, Québec
Joined December 2012
Photo by ed interactive group (@edinteractivegr)Photo by ed interactive group (@edinteractivegr)Photo by ed interactive group (@edinteractivegr)Photo by ed interactive group (@edinteractivegr)Photo by ed interactive group (@edinteractivegr)
ed interactive group retweeted
Your agent is not autonomous You are its loop You check every step because nothing else in the system can I broke down the 7 levels from babysitting agents to running a fleet ↓
95
220
71
1,839
4,452,134
ed interactive group retweeted
This is f*cking insane. Andrej Karpathy, the man who co-founded OpenAI and ran AI at Tesla, just told the labs their giant models are memorizing garbage: "The internet is like 0.001% cognition and 99.99% of information is just garbage I think the cognitive core can be extremely small. I think even a billion, a billion suffices" In 17 minutes, he explains why the next big model is a small one that knows how to think and looks everything else up with tools Giant Model → Cognitive Core → Tools → A 1B Model That Thinks The labs are spending billions to make models remember the internet The model doesn't need to remember everything, it needs to know where to look and how to check This talk is worth more than most $1,000 AI courses Bookmark and watch it today Then read how to build the part around the model in the harness engineering guide below ↓
5
8
1
95
7,574
ed interactive group retweeted
This is f*cking insane. Google Brain founder Andrew Ng just dropped a free 3-hour course on becoming an AI engineer in 2026: "The most valuable skill this year is building with AI. Not chatting with it, building with it Agentic workflows will drive more progress than the next generation of models" In 3 hours, he goes from agentic systems to a full prompting course to shipping a working AI app in 30 minutes Prompts → Agents → Agentic Workflows → A Shipped App → An AI Engineer Job An agent is a model that plans, uses tools and checks its own work An agentic workflow chains those agents so the job gets done without you in the loop Everything else you need is free: Andrew Ng's Agentic AI course deeplearning.ai/courses/agen… AI Python for beginners, taught by Andrew Ng deeplearning.ai/short-course… Anthropic's interactive prompt engineering tutorial github.com/anthropics/prompt… Anthropic's courses: from your first API call to tool use github.com/anthropics/course… Karpathy's Neural Networks: Zero to Hero karpathy.ai/zero-to-hero.htm… Practical deep learning for coders course.fast.ai Microsoft's 21 lessons on building with generative AI github.com/microsoft/generat… Google's machine learning crash course developers.google.com/machin… Hugging Face LLM course, start to finish huggingface.co/learn/llm-cou… The OpenAI cookbook: code for every common task cookbook.openai.com Hundreds of AI apps you can copy and ship github.com/Shubhamsaboo/awes… The prompting guide everyone shares promptingguide.ai Total: $0 95% will keep paying for bootcamps that teach last year's stack. The 5% who start with this course will ship their first AI app this month Bookmark this and send it to someone who wants into AI this year Then read how to wrap that first app in a harness that runs on its own in the guide below ↓
Harness, loop and graph could be the missing layers between AI models and agents that actually work. Same model, same prompt, completely different results. In this article, I show you how to build all three layers.
9
21
1
244
15,332
ed interactive group retweeted
Montreal circa 1963 I find myself nostalgic and longing for an era I never experienced Listen to how its described: “Country that ranks amongst the leaders in world’s trade” “Growth and prosperity” Looked clean, uncongested, vibrant
4
5
35
1,325
ed interactive group retweeted
This is f*cking insane, SpaceXAI engineer Lauren Tan: "You can't spawn a hundred agents when you don't even trust the output of one agent" "If you don't have a verification skill, you are the verifier, you're the bottleneck" In a 1-hour talk, she shows how she went from watching every output to shipping 4,000 PRs a month Watch Every Step → Verify → Skills → Evals → Auto-Merge She now wakes up to 50 PRs her agents merged while she slept Trust in agents isn't a feeling, it's infrastructure you build Worth more than any $6,000 AI coding bootcamp Bookmark it and watch it tonight
Anthropic measured it: one agent burns 4x the tokens of a chat, a team of agents 15x The people running 12 agents on one plan aren't paying more, they're rereading less 6 levers, 6 token leaks and a TOKENS.md you can copy ↓
18
84
6
976
83,697
ed interactive group retweeted
"All code is slop, except mine" - Lauren Tan, engineer on GrokBot at SpaceXSI At AI Engineer Code Summit she just showed her entire coding setup. Every agent mistake becomes a lint rule, not a prompt. Agents verify their own work with videos, screenshots and tests. Result: 3,000 PRs to production in a month. Her codebase is now clean enough that agents merge most of their own PRs. Watch this before you automate anything.
Anthropic's agent team beat a single agent by 90%. Cursor's built a browser in a week. Other big labs are moving the same way. This is how they work, and how you can run one today.
175
830
120
8,710
4,997,909
ed interactive group retweeted
ESTOS 39 MINUTOS DEL FUNDADOR DE KIMI SON ORO PURO explica cómo montó desde China una empresa de IA valorada en más de 50.000 millones hay másteres de 20.000 € que no enseñan lo que él cuenta en 10 minutos. 39 minutos, GRATIS y sin relleno. guárdalo y mira esto cuando tengas tiempo, me lo agradecerás
248
2,024
65
12,519
1,487,989
ed interactive group retweeted
DeepSeek just avenged Old Huang after Anthropic’s decapitation strike by releasing its entire App stack for everyone to use and adapt freely via DeepSeek API. Dario and Sam are now dead men walking as closed models can’t compete. “Now, all the capabilities behind our official products can be freely accessed and combined through the DeepSeek API. We hope developers can use this to skip repetitive foundational work and build more diverse, more polished, and more imaginative products. We welcome you to use the DeepSeek API and create freely.”
🚨DeepSeek居然把他们整个App也开源了。。。。 刚刚 @deepseek_ai 正式发布全家桶应用API,直接把DeepSeek App里几乎所有的功能都开放给开发者了。 包括联网搜索、深度思考搜索、文件解析、阅读链接、语音交互,以及OCR、ASR、TTS等底层能力。 甚至官方还直接开源了一套完整的聊天前端Demo,支持导入DeepSeek历史对话。 也就是说,现在开发者拿着DeepSeek的API,基本就能1:1复刻一个DeepSeek App。 DeepSeek在公告最后还专门表示:「DeepSeek现有的对话产品只是较为基础的一种形态。现在,官方产品背后的各项能力均已支持通过 DeepSeek API 随心调用、自由组合,我们期待开发者能够借此省去重复的基础工作,打造出更多样、更极致、更具想象力的产品。欢迎使用 DeepSeek API,尽情创造。」 现在各家AI公司都在拼命做自己的超级入口,想方设法把用户留在自家App里。 DeepSeek倒好,直接把自己App背后的能力全部开放出来,连复刻方案都给你准备好了。 你想做一个自己的DeepSeek?可以。 你想把DeepSeek的搜索、语音、文件处理能力拆出来,重新组合成一个新产品?也可以。 甚至可以在DeepSeek已有的能力之上,做出比官方App更复杂的产品。 这才是这次发布最值得关注的地方,DeepSeek对自己在AI生态里要扮演什么角色,似乎想得很清楚。 就是下次能不能别在北京时间的周五晚十点发布了。。。。
144
2,346
131
10,123
291,069
ed interactive group retweeted
Why Jensen was fired.
I have a major result to share. I’ve found conclusive forensic evidence that the top 4 Chinese AI labs—DeepSeek, Moonshot, MiniMax and Ant Group—have built their newest models on NVIDIA’s most advanced Blackwell chips, which are barred from export to China. This sounds insane, but I discovered that almost every AI model carries a hidden signature of the GPU that trained it. When a model is first set up, the starting values of its weights are laid out in a pattern that depends on the exact GPU. Each combination of GPU and seed number produces a unique pattern—a sort of fingerprint of the hardware used to create the model. Traces of that pattern survive training into the published model. Again, this evidence isn’t based on any insider information or leaking. It’s based on a forensic analysis of the model weights released by the companies themselves. I searched more than six trillion combinations of chip layout and seed number and matched them to open-weights models. In doing so, I found conclusive evidence that six Chinese AI models—including DeepSeek-V4-Pro, Kimi K3, and MiniMax-M3—were initialized on NVIDIA Blackwell systems. Four of the models match Grace Blackwell, the GB200 and GB300 parts in NVIDIA’s flagship liquid-cooled rack-scale systems. I then verified these results on real hardware. Running unmodified PyTorch with the seeds I’d recovered, a Blackwell B300 reproduced DeepSeek-V4-Flash’s untouched starting weights bit for bit, and a rented GB200 reproduced Kimi K3’s to a correlation of 0.999995. A Hopper H100 reproduced none of DeepSeek-V4-Flash’s, and changing the seed by one broke the match. (DeepSeek’s seed is 953049583 by the way.) The method also was verified to match 24 known open-weights models to the hardware that developers said they used. It is exceedingly unlikely that these results are a coincidence. Across six trillion combinations, the best accidental match should reach only about 8 sigma (i.e., eight standard deviations above what random chance would produce). The real matches range from 49 to 1,385 sigma, and even the weakest single test hits 20. These were data-center-scale training runs. With some analysis of release schedules and model details, I estimate that the largest, Kimi K3 and DeepSeek-V4-Pro, used 2,000 to 3,000 Blackwell GPUs each. Clusters of this size cost on the order of $150 to $200 million each to buy, and across these labs alone would add up to more than half a billion dollars of export-controlled GPUs, not counting experimental compute or otherwise. It’s quite likely that we see multi-billion dollar expenditure on export-controlled GPUs. I am highly confident in this result, and it has been independently verified by senior researchers at OpenAI and Anthropic. You can read the full report at unemploymentlabs.com/checkpo…. There is much more I’ve identified about these models through other forms of checkpoint forensics like this, and will have more to share soon.
170
611
36
7,017
690,275
ed interactive group retweeted
🚨 CYBERATTACKS, CASH STOCKPILING & STABLECOINS: AMERICA IS PREPARING THE BIGGEST FINANCIAL RESET IN HISTORY Scott Bessent said we are approaching another Bretton Woods moment, a “GRAND ECONOMIC REORDERING.” He also said AI and STABLECOINS will help America GROW OUT of its $40 TRILLION debt! Stablecoins will create MASSIVE demand for US Treasuries and the US dollar. Trump’s January 2025 order made dollar-backed coins official policy. The GENIUS Act locked the mechanism in: every compliant stablecoin must sit 1-for-1 on cash, T-bills, and Treasury repos. That’s the escape hatch. America can’t print gold. It can print demand for its own debt. Cyberattacks and emergency stockpiling of cash will be used to accelerate the shift from traditional banking to BLOCKCHAIN TECHNOLOGY. Donald Trump also openly said that the crypto market structure bill will upgrade America’s banking and financial system to the 21st century using “STATE-OF-THE-ART” BLOCKCHAIN TECHNOLOGY. A new surveillance system is being opened without naming it CBDCs. What we really need for the new digital financial system: Private digital identities. Zero-knowledge proofs. A financial system where you can prove who you are without exposing who you are. A future of complete on-chain financial privacy and almost nobody is paying attention. This is why @DNAOnChain is about to be the biggest privacy solution in the new on-chain financial system. Be prepared for the new financial system. It’s always followed by an engineered crisis.
They asked for your data. You gave them a proof. DNA Protocol. Powered by $XDNA: xdna.dnaprotocol.org
27
277
18
846
89,198
ed interactive group retweeted
Carney is MAD for getting bypassed! Quebec’s new Premier, PSPP, sent out 59 letters to American Senators and State Governors to promote economic ties between Quebec and their respective States. When Carney found out, he WARNED PSPP that trade was of federal jurisdiction. This is what PSPP had to say about it: PSPP: “Mr. Carney would like us to stay off the ice and never set foot on the rink because he wants the entire rink to himself. But we, in Quebec, have always claimed our rightful place. That’s what I’m doing with these letters, and that’s what previous governments have done. And he’s being a little contradictory because he says that if you’re not at the table, you’re on the menu. So, I don’t know which Mr. Carney to actually believes…” JOURNALIST: So your intention is not to go around him, if that’s what you’re suggesting. PSPP: “What I’m saying is that I will do what I believe is in the best interests of Quebec and Quebecers, because I am the Premier of Quebec, and Mr. Carney is not my boss. He cannot stop me from doing what we have always done. That means having an international presence that allows us to represent our interests. So, expect to see, over the coming months, a Quebec that takes its rightful place on the world stage. I said it during the election campaign.”
966
2,132
275
10,597
381,669
ed interactive group retweeted
Claude just f*cked the Magic Mouse... this is f*cking crazy my iPhone is now a Wii remote and a trackpad for my Mac. Claude wrote the code in 10 minutes > 2 apps: Mac menu bar + iPhone > ~2,800 lines of code, 0 servers > Bluetooth only, nothing goes online > cursor updates 240 times a second > a free Apple ID is enough > built-in game to practice your aim point the phone at the screen and the cursor lands where you aim. tap to click, hold to drag lay it flat on the desk and it works as a trackpad last time I turned my AirPods case into a mouse. Apple Watch is next free and open source, repo below
138
134
27
1,743
206,874
ed interactive group retweeted
Madre mía. Claude Code acaba de encontrar un chip oculto dentro de mi power bank chino de Guangzhou. Cada vez que cargaba mi iPhone, lo copiaba todo del teléfono en silencio… También leyó la contraseña de Wi-Fi guardada en mi iPhone, se conectó a nuestra red doméstica y envió todo a un servidor. Lo compré por 19 dólares en un kiosco del aeropuerto allí hace 3 semanas. Sin marca, 30.000 mAh en la etiqueta, y se sentía pesado. La primera vez que lo enchufé, la pantalla preguntó "¿Confiar en esta computadora?". Toqué Confiar sin pensarlo y seguí con mi día. Ese solo toque le dio acceso a todo. Esta mañana finalmente configuré Claude Code para que me avisara cuando cualquier dispositivo empezara a leer mis archivos. Una hora después, 10 segundos después de empezar a cargar, la pantalla se llenó de advertencias amarillas, luego rojas: los datos se movían cuando solo debería estar cargando. Le pregunté a Claude Code qué estaba pasando. Mi teléfono veía el power bank como una pequeña computadora con 64 GB de almacenamiento y su propia conexión a internet. Un cable USB tiene cables para energía y cables para datos, y un cargador normal solo usa los de energía. Este usaba ambos, y ya había copiado todo esto: → /device → IMEI y número de serie → /contacts → 1.206 números → /apps → cada app que tengo, incluyendo mis 2 apps de banco → /photos → 3.912 fotos, mi pasaporte entre ellas → /wifi → el nombre de nuestra red y su clave → /iPhone de mi esposa → sus fotos también, ella cargó de él dos veces y tocó Confiar también Cada vez que lo enchufábamos, usaba esa clave para conectarse a internet y enviar los nuevos archivos: envía a → 47.***.***.12 subidas → 19 en 3 semanas la última → anoche a las 2:14, mientras dormíamos Desenchufé el cable de inmediato. Luego Claude Code bloqueó esa dirección en nuestro router y me guió paso a paso para cambiar la contraseña de Wi-Fi y los inicios de sesión de mis apps de banco. Lo abrí en un taller de reparaciones, y lo que había dentro está en el primer comentario. Una verificación de 30 segundos para la tuya está en el segundo. Guarda esto y prueba cada power bank en tu casa hoy.
Hoy lo hago oficial: Viktor ya es parte del equipo. Hablo de IA y negocios online, y las marcas me escriben por DM y por correo. Desde octubre esa bandeja la lleva él, mi empleado de IA, con una regla: nada sale sin mi sí. Desde el 29 de octubre: → 8 mensajes de marcas leídos → 4 respuestas a marcas con presupuesto, enviadas con mis tarifas tras mi aprobación. → De esas 4 respuestas, 3 acabaron en colaboraciones pagadas. Él propone, yo decido. Un programa se instala. A un empleado se le da la bienvenida. Hire @viktor_com for your team. Start free. #AIemployee #NegociosOnline Paid Partnership
97
262
19
1,147
245,343
ed interactive group retweeted
This is a shared memory service where you can connect multiple agents and AI tools to reuse knowledge. For example, you can save information using Codex and retrieve it through Claude Code. Two scenarios where this is very helpful: • To keep a portable knowledge base not tied to any agent. • To have the flexibility of using multiple agents without duplicating the knowledge.
We decoupled memory from the model. Living Brain is one persistent brain for ChatGPT, Claude, Cursor, Codex and your own agents. Write in one, retrieve in any. 92.2% on LoCoMo. 464ms median retrieval. Up to 22.6x fewer tokens per question. LLMs have parameters. Living Brain has plasticity.
57
14
1
118
34,832
ed interactive group retweeted
This is f*cking insane. Anthropic Engineer Andrej Karpathy: “99% of people are still forcing agents to work instead of mastering the /harness engineering + Opus 5.5. We made the exact same mistake at OpenAI in 2016 — it cost us 3 years and billions. The agent is not the product. The foundation is. Master that and agents appear on their own.” Opus 5.5 → Prompts → Agents → Graphs → Harness What Karpathy actually means: step 1 → stop forcing the agent to do everything — understand the model and harness first step 2 → demos are easy, real products take a decade — self-driving proved it. Skip the foundation and the whole system collapses step 3 → the agent is not the product. The foundation is. Build that properly and agents emerge by themselves “If you’re building agents right now, you’re ahead of OpenAI and DeepMind. Not them. You.” Watch it today before it gets buried. Then read the full Opus 5.5 harness engineering guide below ↓
Stop letting unfinished Opus 5.5 runs eat your budget Set up the harness before your next long task Save progress, verify results, reuse the workflow The full 7-layer setup below ↓
12
85
2
1,096
180,375
ed interactive group retweeted
The TOTAL DEMOLITION of the IPCC Our latest peer-reviewed paper concludes that fossil carbon is a NEGLIGIBLE component of the net atmospheric CO₂ increase. 500 years of isotopic evidence. Elementary mass conservation. A devastating mathematical result.
BOMBSHELL Fossil Fuels Did NOT Cause the CO2 Rise. New peer-reviewed paper just dropped. 500 years of data. The verdict is in. Fossil carbon's contribution to rising CO2: effectively ZERO. The entire IPCC/NASA/NOAA/EPA climate narrative rests on a claim that just got mathematically falsified. Read it. Share it. The cover-up is over. The full peer-reviewed paper: papers.jcohler.com/keeling
398
4,097
427
10,229
447,702
ed interactive group retweeted
Claude Code found that every free charging station i used in airports copied data from my phone. photos, passwords, banking sessions, contacts. i charged my phone at 3 airports across China and every single station pulled something from my device. i travel through Asia a lot for work. my phone dies at airports constantly. i've used those free USB charging stations at gates without thinking about it. everyone does. i installed Claude Code on my iPhone and ran a full device scan while i was still at the airport. plugged into a charging station at the gate and let Claude Code monitor the connection in real time. within 8 seconds it flagged a data handshake that shouldn't exist on a power-only connection. i went back and checked every charging session from the last 4 months: → Shanghai Pudong PVG, Terminal 2 → June 14 → contacts + 2,840 photos transferred → Guangzhou Baiyun CAN → July 2 → active Safari sessions copied including Gmail and banking apps → Shenzhen Bao'an SZX → August 19 → full photo library access + location history for 6 months → Shanghai Pudong again → September 3 → iMessage database copied, 3 years of messages → Guangzhou again → October 11 → clipboard history including a crypto wallet address i copied that morning every station initiated a data transfer the moment i plugged in. none of them were just charging. the scariest one was Shenzhen. whoever operated that station got my full location history for 6 months. every place i slept, worked, ate, and visited. from one 20 minute charge while i was scrolling my phone waiting for my flight. Claude Code traced 4 of the 5 transfers to the same server: → IP → 116.25.***.34 → location → Shenzhen, China → hosting → dedicated server, registered 45 days before my first transfer → total data pulled from my phone → 4.7 GB across 5 sessions what Claude Code did: → revoked every active session on Gmail, iCloud, WhatsApp, and banking apps → rotated every password stored in Keychain → purged all cached session tokens → flagged the crypto address i copied as compromised before i sent anything to it → disabled USB data transfer for all future connections i never lost money. Claude Code caught it before anything was used. but 3 years of iMessages, my banking sessions, and my entire photo library were sitting on a server in Shenzhen for months. buy a $12 charge-only USB cable. it has no data pins. or use a wall outlet with your own adapter. never plug a USB cable from a public station directly into your phone. the full prompt is in the first reply. run it tonight. takes 3 minutes. if you've ever charged your phone at an airport, a mall, or a hotel lobby, someone might already have your data. save it and send it to anyone who travels.
Readers added context they thought people might want to know
Account pattern is consistent and commercial. @starmexxx posts a steady stream of these dramatic "Claude Code found..." stories. Many link to or promote digishield.app, which sells paid packs ($19-$39) of "16 ready-to-run checks that turn Claude into your personal security scanner"
201
840
176
2,889
565,236
ed interactive group retweeted
holy f*ck, web designers have NO IDEA what’s coming after them in 2027 the ones learning AI tools right now are getting ahead. the rest might be fighting for jobs they used to take for granted. 7 GitHub repos every designer should know before it’s too late: 1. Graphify 137k stars and in Y Combinator. point it at your old projects, brand PDFs and screenshots, and Claude gets a memory of everything you've ever designed. "what did this client hate last time?" answered in seconds github.com/Graphify-Labs/gra… 2. Awesome DESIGN.md 84 design systems pulled from real sites, each one a single markdown file. drop it in your project, say "build me a page that looks like this", done github.com/VoltAgent/awesome… 3. Hallmark a design skill from Together AI that refuses to look AI-generated. 50 themes, checks every page in a real browser, and /hallmark redesign rebuilds any old page github.com/Nutlope/hallmark 4. Huashu Design one sentence in, a launch animation, clickable app prototype, editable PPT or infographic out. it shows you 6 directions first, you pick 3 github.com/alchaincyf/huashu… 5. Designer Skills Pack 273 skills and 76 commands for Claude Code: research, design systems, UI, interaction. a whole design department in one install github.com/Owl-Listener/desi… 6. Motion Design Skill LottieFiles taught AI to think like a motion director: timing, easing, choreography, Disney's 12 principles adapted for UI. this is where the "feel" comes from github.com/LottieFiles/motio… 7. Blender Agent Skills 22 pro Blender workflows for Claude Code: modeling, materials, rigging, animation, rendering. the full 3D pipeline, not just "make a cube" github.com/EL4CTEO/blender-s… the cell you're watching? built solo with Claude Code plus a few of these tools. no agency, no team, no After Effects. want to make something like this? like + save this post, comment "CELL" or DM me and I'll send you the exact prompt.
135
245
18
1,512
126,853
ed interactive group retweeted
Sweet. It's a 1 megawatt nuclear microreactor that fits in a shipping container. Gas-cooled, first by helium, then air. Runs for five years, then you ship it back to the factory, where they take care of the spent fuel and refuel it.
BREAKING: For the first time in American history, the Department of Energy has approved a commercial nuclear microreactor's safety analysis for full-power operations. LA-based Radiant secured the approval, with testing planned at Idaho National Laboratory.
357
2,180
146
22,585
650,728
ed interactive group retweeted
‼️NEW: Canada just dropped up to $1.6 billion of taxpayer cash on one of the worst-performing AI labs in the world. - Cohere’s top model sits at #210 on the LMArena leaderboard... The compute gap is brutal: - Cohere: 8,500 GPUs vs xAI who is ramping up to 1.21 MILLION chips. Cohere’s systems are 50–200× smaller than the frontier labs. Another flop. Its the Lyon electric of AI.
272
930
81
3,155
78,069