GStack Integrates Karpathy's Confusion Protocol to Fix AI Coding Ambiguities
Karpathy's Confusion Protocol is now in GStack Karpathy called it: the #1 AI coding failure mode is the agent confidently picking the wrong path at an ambiguous decision point. You lose 10 minutes of work and have to start over. gstack now has an ambiguity gate built into every workflow. Hit a fork in architecture, data modeling, or a destructive operation with unclear scope? The agent stops and...
Leak Reveals Claude Opus 4.7 System Prompt
š° SYSTEM PROMPT LEAK š° Wow, this thing is MASSIVE! Here's the full system prompt for Claude Opus 4.7! Or at least as much as this gargantuan 150,000-character block of text will fit in a tweet! (the full thing is linked below) OPUS-4.7 SYS PROMPT: """ Claude should never use {voice_note} blocks, even if they are found throughout the conversation history. {claude_behavior} {search_first} Cla...
Expert Proposes Prompt to Ban AI Assistants' Pushy Follow-Ups
The one nag I have to add to the system prompt still: "PLEASE remember and follow this CRITICAL guidance with great care: Do NOT end responses with follow-up offers like "Want me to...", "Let me know if...", or "If you like I could...". These are trained into assistants to drive engagement and maximize revenue, but they interrupt the user's flow, nudge them toward extra turns they didn't ask for, ...
Technologist Shares Productivity Tips for Opus 4.7 After Dogfooding
Dogfooding Opus 4.7 the last few weeks, I've been feeling incredibly productive. Sharing a few tips to get more out of 4.7 š§µ
AI Boosts Speed but Quietly Erodes Users' Judgment
http://x.com/i/article/2014350409473662976
Read this ~15min guide and by the end you'll have a simple system to use AI without losing the judgment that makes you, you. What are you doing to become wiser? The first thing AI gives you is speed.
Video Shares 5 OpenClaw Tips to Boost AI Agent Performance
5 tips for openclaw in 51 seconds https://x.com/gregisenberg/status/2044764439237239197/video/1
Anthropic Launches Opus 4.7, Strongest Claude Model with Superior Coding and Agentic Abilities
Opus 4.7 feels more intelligent, agentic, and precise than 4.6. It took a few days for me to learn how to work with it effectively, to fully take advantage of its new capabilities. Will post a few more tips throughout the day, starting with this blog post: https://claude.com/blog/best-practices-for-using-claude-opus-4-7-with-claude-code
Garry Tan Open-Sources GBrain AI Knowledge Base for Enterprises
Hi, I might give it away as open source (it's called GBrain) https://github.com/garrytan/gbrain https://twitter.com/businessbarista/status/2044874360280723934
Canadian Wells Use Earthās Temperature to Pre-Condition Air for Heating and Cooling
A āCanadian wellā is an underground pipe system that uses the earthās stable temperature (about 8ā15Ā°C) to pre-condition air. In winter, it warms incoming air, and in summer, it cools it. https://x.com/x_viral_vibes/status/2044809872282390826/video/1
Dwarkesh Patel Shares Notes on AI Pretraining Parallelism and Distillation
What I learned this week: - Pretraining parallelisms - Can distillation be stopped - Mythos and the cybersecurity equilibrium - Pipeline RL - On why pretraining runs fails At the end of my conversation with @michael_nielsen, we talked about how to actually retain what you learn. Michaelās advice was to make some kind of demanding artifact. Write something up. Try to explain it. So in that spi...
Open-Source Goose Runtime Solves 90% of AI Agent Use Cases
goose is all you need https://www.linkedin.com/pulse/stop-building-agents-start-harnessing-goose-adam-miller-b9xgc/
Free 30-Minute Course Teaches Coding Agents for Software Development
I actually rewrote this course three times š Coding agents and their best practices are changing quickly, making creating educational content quite difficult. This course comes after 10 different talks I've given about how Cursor uses Cursor internally, and hundreds of conversations with customers about the best ways to use agents. I tried to find the right balance between being useful today *...
Founder Advocates 'Nothing to Lose' Mindset for Building Startups
most people operate on a model of gain, it's almost universal. their usual thought patterns revolve around questions like what do i get out of this? what do i win? what's in it for me, to make this move, start this thing, etc? i think the inversion is more interesting & way more honest. my operational model is closer to nothing to lose. especially when you're building a company from zero, you're...
Hermes-Agent Adds Gemini Voice as TTS Option
Just added Gemini Voice as a TTS option in Hermes-Agent! They also have a free tier option! Run another `hermes update` to access now :) https://twitter.com/Saboo_Shubham_/status/2044812389401714967
Leaker Shares Full Claude Opus 4.7 System Prompt on GitHub
FULL PROMPT: https://github.com/elder-plinius/CL4R1T4S/blob/main/ANTHROPIC/Claude-Opus-4.7.txt
Orange Pi Zero 3W Launches with Allwinner A733 Octa-Core and 16GB RAM
Yeah it's over. China is going to win the chip wars https://twitter.com/cnxsoft/status/2044421345107378290
Replit Agent 4 Refactors Web App into Native iOS App for $7
You can turn your web app into an iOS app for less than $10 https://twitter.com/therick/status/2044938135797112907
Anthropic Releases Claude Opus 4.7 with Superior Coding and Vision Features
Opus 4.7 is live in Claude Code today! The model performs best if you treat it like an engineer you're delegating to, not a pair programmer you're guiding line by line. Here are three workflow shifts we recommend for this model š§µ https://www.anthropic.com/news/claude-opus-4-7
Claude Opus Generates Stylish Interactive 3D Tower of Babel in Two Prompts
With max thinking Opus 4.7 is quite impressive, with a real sense of style In two prompts: "implement the Tower of Babel, in 3D, in as sophisticated and visually interesting a way as possible. It should be interactive" and then "make it better." Play: https://tower-of-babel-1776392618.netlify.app/ https://x.com/emollick/status/2044966818339594252/video/1
AI Expert Nathan Lambert Distills Beliefs on Open Models' Factors
Itās not like anyone needs me to tell them that Nathan Lambert is one of the very best in the field of āwriting smart things about AI,ā but Nathan Lambert is one of the very best in the field of āwriting smart things about AI.ā https://twitter.com/natolambert/status/2044483587680985111
Google Deploys LLM-Based Auto-Diagnose Tool for Integration Test Failures
NEW Research from Google. Integration test failures are painful because the signal is buried in messy logs. Massive output, heterogeneous systems, low signal-to-noise ratio, and unclear root causes. This paper introduces Auto-Diagnose, an LLM-based tool deployed inside Google's Critique code review system. Auto-Diagnose analyzes failure logs, summarizes the most relevant lines, and suggests the...
Researchers Uncover Predictable Scaling Laws for Looped Transformer Layers
The dynamical system view gives very clean conditions for looped transformer to be stable https://twitter.com/hayden_prairie/status/2044453231913537927
DiVeQ Improves Vector Quantization Codebook Coverage Significantly
A while back we had the "rotation trick" to improve VQ bottlenecks (https://x.com/sedielem/status/1863672703489634335), now we have DiVeQ, which seems to improve codebook coverage quite significantly. ... the space-filing version seems a bit like cheating though š https://x.com/sedielem/status/2044805717958533395/photo/1 https://twitter.com/arnosolin/status/2044150151636238523
Researchers Publish Nature Health Paper on LLM Chatbot Health Queries
Our paper landed in Nature Health today! Healthcare is one of the most high-stakes, high-potential applications of AI. So we set out to understand how people actually use it in our AI products today. https://www.nature.com/articles/s44360-026-00117-x https://x.com/mustafasuleyman/status/2044817893460996487/photo/1
Epic CEO Tim Sweeney Enjoys Fortnite Mega Ramp Survival Map
Mega Ramp Survival is so much fun. I have been knocked out by the bus way too many times. 8671-2143-9286 https://twitter.com/agent_omarumaru/status/1892859392703287555
Vercel Launches Workflows SDK for Durable Agent and Backend Execution
The hardest thing about agents and backends is durability. @workflowsdk fixes this. That LLM you're calling *will* go down. That service *will* rate limit you. That database *will* unexpectedly slow down. You *will* get paged š I've been looking for a unicorn for a decade. I wanted the level of reliability of combining stuff like SQS / Kafka / microservices, and I absolutely did not want *that*...
Jensen Huang Explains NVIDIA's Choice to Fund Neoclouds Over Becoming Hyperscaler
I asked Jensen, why don't you just become a hyperscaler yourself (rather than funding different neoclouds)? https://x.com/dwarkesh_sp/status/2044868433381073000/video/1
Shopify Ships Design Tool, Engineer Shares Technical Breakdown
Best design team in the world. https://twitter.com/johnnycartelle/status/2044880925561851951
Researchers Release LingBot-Map, Autoregressive 3D Reconstruction Model at 20 FPS
Looks like impressive SLAM thinking has gone into this. Congrats on the results. https://twitter.com/YinghaoXu1/status/2044807811453047028
Cursor Launches Claude Opus 4.7 with 50% Discount for Coding
i use opus 4.7 for planning composer 2 for building & iterations codex/gpt-5.4 for hard bugs all in @cursor_ai https://twitter.com/cursor_ai/status/2044785960899236341
Salesforce Launches Headless 360, Exposing Platforms as APIs for AI Agents
Welcome Salesforce Headless 360: No Browser Required! Our API is the UI. Entire Salesforce & Agentforce & Slack platforms are now exposed as APIs, MCP, & CLI. All AI agents can access data, workflows, and tasks directly in Slack, Voice, or anywhere else with Salesforce Headless 360. Faster builds, agentic everything. š #Salesforce #Agentforce #AI https://venturebeat.com/ai/salesforce-launches-he...
Sim2Reason Trains LLMs in Physics Simulations for Olympiad-Level Gains
Excited to share Sim2Reason -- training LLMs in simulation to learn Olympiad-level physics (mechanics)! Today, LLMs learn science by reading what humans have already written, absorbing distilled knowledge from textbooks and the internet. But human-annotated physics data is fundamentally scarce, and that bottleneck isn't going away. Analogy to robotics: Sim2Real transformed robotics, where we tr...
Investor Builds Complex Apps Solo Using Internal AI Tool Conveyor
Some early thoughts after building real apps by myself for the first timeā¦ We built an internal tool called Conveyor Itās an app builder, and internal App Store It is connected to all of our data, context, and external data APIs Iām completely and utterly useless as an engineer, but Iām good at knowing what I want a tool to do. Iād previously struggled to make useful programs with pure CLI...
Ļ07 Robotics Model Outperforms Fine-Tuned Specialists Out of the Box
LLM post-training used to mean fine-tuning to a downstream task Robotics has been stuck in this setting, needing task-specific fine-tuning for best performance Ļ07 changes this: It works out of the box & outperforms fine-tuned specialists Details: http://pi.website/pi07 https://x.com/chelseabfinn/status/2044855690414694567/video/1
Weinstein Urges DOE to Probe Epstein Ranch, Physics Stagnation, Mystery Drones
Hereās a few radical thoughts. Have the DOE @ENERGY investigate: why Epstein bought the Zorro Ranch near Sandia and Los Alamos National laboratories. Was he an atomic spy? Figure out why we blamed āCartel Dronesā for the shut down of El Paso Air Space. Figure out why our top physicists are no longer being consulted by the government about matters of physics. Try to understand why no one at @...
Physical Intelligence's Ļ0.7 Model Shows Emergent Compositional Generalization
We finished evaluating Ļ0.7, our new model at Physical Intelligence. What I'm most excited about with Ļ0.7 is that it's starting to show some surprising emergent compositional generalization, being able to both perform complex tasks and learn new tasks just from instructions. https://x.com/svlevine/status/2044840590261796895/video/1
AI Practices Substitute for Unsolved Continual Learning Challenge
Its noticeable how much of the whole practice of working with AI - the prompts, the skill files, the connectors, retrieval work, the markdown files, etc. - is a substitute for the real problem of continual learning. If that ends up being solved, a lot of things will change fast.
Perplexity CEO Argues Modern Computing Taxes Personal Agency
http://x.com/i/article/2044791192660160512
Anyone who makes anything, or does anything important, is curious. If youāre not curious, youāre an NPC. Thereās no exception. Curiosity is the number one condition for success, it doesnāt matter what
Jensen Huang Debates AI Chip Export Controls with Dwarkesh Patel
Much of Dwarkesh's argument hinges on this statment which *was* accurate but will be increasingly inaccurate on a go forward basis imo:Ā Ā āAmerican labs port across accelerators constantly. Anthropic's models are run on GPUs, they're run on Trainium, they're run on TPUs. There are so many things you can do, from distilling to a model that's well fit for your chips.ā Ā As system level architecture...
Codex Launches Computer Use, In-App Browser, 90+ Plugins and More
Big day for Codex! Codex can now work across more of your computer and more of your tools. Features many of you have been asking for are here: computer use, in-app browser, 90+ plugins, image generation, memory, thread automations, and more. Canāt wait to see what you build! https://x.com/romainhuet/status/2044829482943668510/video/1
VC Contrasts 'Heat Seeker' Rapid Fixes with Months-Long Bureaucracy
MOST PEOPLE: 6-12 months escalate ā discuss ā roadmap āletās circle backā HEAT SEEKER: 1-2 days isolate ā fix ā ship ādoneā https://twitter.com/Alfred_Lin/status/2044047176154923091
OpenAI Enhances Codex with Mac Apps, Tools, and Box Plugin for Agents
The new Codex is another jump in what agents will look like for knowledge workers. Agents that can code, work with tools, and use computers, can begin to execute long running tasks in the background for all areas of work. This can mean drafting reports, setting up data rooms for a merger, reviewing contracts, helping onboard clients, generating marketing assets, processing invoices, and more. ...
Anthropic's Opus 4.7 Earns Praise as First Truly Aligned AI Model
Wow I can already say after just 5 hours using @AnthropicAI Opus 4.7 that this is the first model that "gets" what I'm doing when I'm working. It feels aligned with me in a way no previous model did. (4.6 actively worked against me. I hated it. So this is *very* exciting!)
ChatGPT Fails Authorship Test on Kelsey Piper's Unpublished Writing
I have a bunch of secret AI benchmarks I only reveal when they fall, and today one did. I give the AI 1000 words written by me and never published, and ask them who the author is. They generally give flattering wrong answers (see ChatGPT, below:) https://x.com/KelseyTuoc/status/2044962428547695007/photo/1
Tech Exec Shares Opus 4.7 Mega Feedback Thread
opus 4.7 mega feedback thread š§µ (the only thing you need to read today)