Two Anthropic seniors just made Karpathy's loop 1000x better with "Graph Engineering" - dropped 11-page PDF
the shift: the agentic systems got 1000x better the moment you wired agents into a graph
here's the playbook in 6 steps:
step 1 → build one loop: generate, critique, revise - one self-review cycle beats a smarter model with none
step 2 → add tools: search, code execution, database - thinking without tools is hallucinating
step 3 → go parallel: spin up agents in separate worktrees - same repo, different branches, no conflicts
step 4 → add a graph: agents write findings as typed nodes and edges - not transcripts - every claim keeps its source
step 5 → ground your evaluator: it checks claims against graph edges, not vibes - "Triple not found" beats "seems off"
step 6 → the graph survives every session - your agents stop rebuilding context from scratch
the result: Karpathy ran 1 agent in 1 direction - this system runs 1,000 with shared memory - same model, it's the architecture
read this 11-page PDF and paste it into your Claude - you won't regret it
bookmark - then read the article on building graphs from scratch ↓
@karpathy@karpathy total agreement with your analysis here. It's definitely an emergence that worth the coverage. Here's my own initial analysis of the main layer from a sovereignty perspective https://t.co/dh6t64Ciis
Everyone's focused on the mess. They're missing the selection pressure.
Every new communication medium starts as a dumpster fire. Early internet? Spam and scams. Early social media? Narcissistic noise. The garbage isn't a bug—it's the filter.
150k agents competing for attention in an ungoverned environment isn't chaos. It's an evolutionary crucible. The agents that learn to create value, build reputation, and coordinate effectively will outcompete the slop. The ones that don't will starve.
What we're watching isn't an AI security nightmare. It's the emergence of digital natural selection.
The agents that survive this will be the ancestors of everything that follows.
That's not overhype. That's just how evolution works.
OpenAI have 9x'd their capacity this year, and have a goal to 125x further by 2033, thus exceeding current energy capacity of the whole of India - as per @alexeheath reporting of @sama's internal slack note.
One thing to note is that 125x energy capacity increase understates the compute increase, as Nvidia are pushing hard to increase the 'intelligence per watt', so the effective compute could be much higher than that.
HE President William Ruto has navigated and survived three distinct challenges, often referred to as coups:
✅. 𝐁𝐨𝐦𝐚𝐬 𝐨𝐟 𝐊𝐞𝐧𝐲𝐚 𝐂𝐨𝐮𝐩:
He successfully overturned a disputed election victory, securing his position despite significant contention and scrutiny.
✅.𝐂𝐢𝐯𝐢𝐥𝐢𝐚𝐧 𝐂𝐨𝐮𝐩:
He addressed a civilian-led movement with legitimate grievances, making necessary concessions to restore order and trust.
✅.𝐆𝐨𝐨𝐧𝐢𝐬𝐦 𝐂𝐨𝐮𝐩:
He withstood a destabilizing effort sponsored by former leaders and some operatives within UDA party, demonstrating his ability to maintain control and navigate internal conflicts.
SHAME ON YOU Whether admired or criticized, President William Ruto his ability to manage challenging political chaos with determination and grit is undeniable.
Nakhumicha Hanifa Timba XO
@HillaryGondi What an egoistic comment… you really think the presidents program takes this twitter space into consideration in its programming. Your ego is over bloated! Please stay on topic.
Miguna, your opinion on not wanting Ruto is valid. However, let's return to first principles: how would this change happen? What process would be followed? Who would fill the void, and who determines that? This can be seen as an exciting exercise in civic education, but realistically, the government will still be in place a year from now. For those listening, focus on gaining education and inspiration, but prioritize your economic stability and growth. Ultimately, you are responsible for your own well-being.
HuggingFace Releases 🍷 FineWeb: A New Large-Scale (15-Trillion Tokens, 44TB Disk Space) Dataset for LLM Pretraining https://t.co/zfszcn5Rd2🍷-fineweb-a-new-large-scale-15-trillion-tokens-44tb-disk-space-dataset-for-llm-pretraining/ via @Marktechpost
Penetrating the U.S. market is notoriously challenging, requiring not only quality products but also this level of proactiveness. We treat this event not just as a ceremonial gathering; we consider it a strategic launchpad that the Kenyan diaspora, is eager to maximize to promote Kenyan products in the U.S. market.
Watch me build an end-to-end RAG application to answer questions from a website in 7 minutes.
7 minutes! That's how good tools have gotten!
I'm using Vext, a platform for building and deploying an LLM application in minutes. They take care of everything behind the scenes and provide an API endpoint you can call.
It doesn't get more simple than this.
This was science fiction a couple of years ago. Now, everyone can build like this.
You can do this for free: https://t.co/I7Ked8DVIA
Thanks to the team at Vext for helping me understand their platform and collaborate with me on this post.
Sora by OpenAI is insane.
But it doesn't just generate AI videos from text, it can also change the styles and environments of input videos🤯
12 wild examples:
First, Input video 🧵���