Andrej Karpathy
@karpathyWe'll be spending a lot more time trying to understand the outputs of language models. A few thoughts, tips & tricks: Writing. Something I've had success with: Ask your LLM to explain something in ASD-STE100, it's a controlled language specification originally developed for aerospace maintenance documentation. LLMs well-versed in this language and it comes with heavy constraints on clean writing style that I often find a lot more readable. Sometimes I've tried to soften it a bit e.g. ask for "80% of the way to ASD-STE100" because the spec is quite stringent. But even better: Diagrams / images. Instead of writing, ask your LLM to create a diagram. These can be a lot easier to process, parse, and understand. But even better: Web pages. Ask for output "in HTML" to get a beautiful, interactive webpage. LLMs are getting really good at frontend and can create beautiful experiences, animations, etc. But even better: Explainer videos. The output format I am most bullish on is fully custom / bespoke explainer videos generated on any arbitrary topic. Experiment with things like "Create a 3b1b style video explainer on X. Use my ElevenLabs API key for audio narration". (you'd need an API key for the latter or you can ask your LLM to find you decent free alternatives that use your local compute). This is actually starting to work! In summary: - As LLMs get better, they will do more and more of the legwork autonomously, and a lot more of our work will rise up the abstractions into oversight and understanding. - Luckily, LLMs can help here too because as intelligence and code are increasingly abundant, you can ask for large, custom, discardable software artifacts (e.g. web apps, video explainers) that would have never made sense to create before. Push the boundaries here and you'll be surprised.
October 2, 2026
43 original posts · author attribution and direct source links
Andrej Karpathy
@karpathyCool eval. Simply ask an LLM “Land or Water?” and give it a latitude and longitude coordinate as text. Ask 16,200 times, plot as image. The models know. From compressing the internet.
Invideo
@invideoOfficialThe invideo MCP is now live. Connect your Claude, GPT, or any other agent you already work with, control the entire editor from your own console, and generate images and videos as well. The editing workflow just got a lot more flexible. https://t.co/IODuBGX2UU
DeepSeek
@deepseek_aiPlease check out @DeepSeekHarness. We just released packaged desktop versions for macOS and Windows; Linux users can get it from the @deepseek-ai/dsh package on npm. https://t.co/Cde0dZ6cPy
OpenAI Developers
@OpenAIDevsYour dot learns how you work, keeps context across apps, coordinates Codex tasks, and flags what needs your attention.
AI at Meta
@AIatMetaFollowing gold-medal-level performance from our AI models across five competitions in mathematics, physics, and chemistry, we asked a harder question: can AI contribute when a problem is genuinely open and without an existing solution path? Over the past several months, mathematicians worked with Muse Spark 1.1 and Muse Spark 1.2 in Thinking Mode through the regular https://t.co/wHkMPH82ZH chat interface, with no custom research scaffold, to find solutions to such problems. Our goal wasn't to mass-produce papers, but empower researchers. Every collaboration followed the same principles: mathematicians guided the research, a second group of mathematicians then reviewed the work, each paper marks which passages were drafted primarily by humans or AI, and each credits the prior research it builds on. Where other teams independently announced solutions to the same problems, we acknowledge their work as well. Today, we're sharing six papers from that collaboration. 🧵👇
Hugging Face
@huggingfaceThe same model, with the same weights, scores 62% in one agent harness and 33% in another. @adithya_s_k and the @huggingface team just released the ultimate guide to multi-harness RL, and it's one of the most practical RL write-ups this year, and everything open! The trick is simple. Don't touch the harness. Point it at a proxy instead of the model. The proxy speaks all four API formats coding agents use (OpenAI Chat Completions, OpenAI Responses, Anthropic Messages, Gemini). It records the exact token ids and logprobs vLLM sampled, and you train on that. You don't change a single line of Claude Code, Codex or OpenCode. Results: 🔹 Trained across 4 harnesses at once, LFM2.5-2.6B by @liquidai went from 42% to 54% 🔹 31% fewer tool calls, thanks to a small bonus for solving tasks in fewer steps 🔹 Training in OpenCode alone took OpenCode from 34% to 58%, but the multi-harness model improved everywhere They also tried the shortcut everyone reaches for: fine-tune on 3,189 successful rollouts from Qwen3.8-27B. Imitation plateaued at 47.5%, below both RL runs. Copying a bigger model doesn't get you there. Practice does. The best part is that everything is open: the capture proxy in OpenEnv, the trainer in TRL, the tasks, the SFT data, the training code and all seven trained models. Agents will run in dozens of harnesses. Now open models can be trained for each of them, by anyone. Read it here 👇 https://t.co/s2I8GI9SLS
rari
@0xwhrrariMost people will use Dots like a smarter chat and miss the point The real shift is AI that keeps working between prompts while the important decisions stay with you I broke down how Dots, Space, Astra, Sol, and Codex fit together ↓ https://t.co/31fW4gHM5U
Cursor
@cursor_aiWhen Rollouts catches a regression, it finds the offending PR and opens an issue. One click starts a cloud agent to fix it. Rollouts usage credits are included through Oct 3.
Ridark
@ridark_ethhttps://t.co/NDVK91DGLI
swyx
@swyxbtw dont forget the "one more thing..." theyve been working on dots for a while now https://t.co/B83svUV9AC
sopersone
@sopersonehttps://t.co/TqxlmoZhvl
Vercel
@vercelJev is now in the AI SDK for Python. To test it, we ran two experiments: ▪︎ Detecting Python vs. English as text is typed ▪︎ Writing Python, one decision at a time 𝚞𝚟 𝚊𝚍𝚍 𝚊𝚒 https://t.co/fWPx6k3Gw3 https://t.co/eLaAvdzGdS
Framer
@framerThe Expert program is open for applications again, and it’s now easier to join. All you need is 2 high-quality sites on your profile. https://t.co/yGRW6LH0hV ✓ 50% revenue share from project transfers ✓ Agent consumes 50% less credits (Pro) ✓ Free edit seat in client projects (Pro)
Google AI
@GoogleAIProject Suncatcher has liftoff! ☀️🚀 Announced last year, this moonshot project explores if we can one day host machine learning infrastructure in space. After years of research, we’ve just launched a prototype satellite, built in partnership with @planet, into orbit on @SpaceX's Transporter-18 rideshare mission. This initial mission will gather data on how @Google TPUs handle the physical stress and extremes of spaceflight. Ultimately, these findings will help us explore how to potentially link multiple satellite constellations for scaled machine learning. By operating in low Earth orbit, these systems can tap into near-constant sunlight to generate up to 8x more solar power than on Earth.
swyx
@swyxIt's time to get serious about Security x AI. One of the joys of doing AIE is helping others start their own high quality conferences in their countries and domains. Every single 2025 partner has come back — and the 2nd AI Security Summit is now way more important than before. Proud to be back as a founding partner with @snyksec! Join the event: https://t.co/2IeM4a3MGO The landscape has completely transformed since we started this a year ago. An exploding number of rogue agents, more breaches, more AI-driven attacks. now more than ever, security must be at the center of the AI conversation. With so much more hacks and vulnerabilities, this stuff is no longer theoretical...
Joying.eth
@Joyingethhttps://t.co/0teiTTwyX5
OpenAI Developers
@OpenAIDevshttps://t.co/jSGjA164Ag
Visual Studio Code
@code🐾 Meet Blobby, the VS Code Pet! Thank you to everyone who submitted a name and helped us choose our new coding companion. 💙 Want to code with Blobby by your side? Open GitHub Copilot Chat in VS Code and enter: /vscode-pet 🏆 We’ll be reaching out to the contest winners directly with the next steps. Welcome to VS Code, Blobby! 🐾✨
Framer
@framerCheck out these 4 interactive components you can build for your website with the Framer Agent. Prompts below ↓ https://t.co/bPIeBjDNFX
Jonny Burger
@JNYBGREditing all of yesterday surfaced 46 issues in Remotion! 21 are closed already ✅
AI at Meta
@AIatMeta1️⃣ Probability: Mathematicians worked with Muse Spark to answer a question about fitting random points onto the surface of a stretched sphere. For the setting studied, they proved a sharp cutoff between when an exact fit is likely and when it is unlikely. Read the paper: https://t.co/T2rTfHafQV
Vercel
@vercelEngineers at @RogoAI build the company's internal apps on Vercel. They deployed over 73,000 times last month. Coding agents ship to production in 5 minutes, and run everything from churn risk analysis to the sales team's deal desk. https://t.co/PO07A4nDc3
GSAP
@greensock💚 Site of the Day - Millemann Wines Index Studio — @indexstd Quentin Hocdé — @QuentinHocde ⚙️ ScrollTrigger 🛠️ Reduced Motion, WebGL, OGL, lenis site → https://t.co/yiNNl8iKEh showcase → https://t.co/zGjV9MxQ0f https://t.co/ByimwCd8II
Framer
@framerLearn how the Framer CMS organizes repeatable content into collections, items, and fields, how the Agent helps you import structured content, and how to connect it all to reusable pages. https://t.co/T1F4dKNTCW
LangChain
@LangChainValidate agent fixes before shipping them with LangSmith Engine v2. Engine now tests and validates fixes automatically: ✅ Replicates the issue with the same deployment environment and same inputs ✅ Builds a fix, tests it, and iterates on improving it until it has a satisfactory solution ✅ You review the fix and deploy it with a ready-made PR
LangChain
@LangChain💻 Build the interface you want w/ your LangSmith data ✅ Publish it into your workspace 🙅 Skip hosting, auth, + permissions work Get started with Custom Apps today. https://t.co/MayBpUbhnW
LlamaIndex 🦙
@llama_indexThe BEA's new Q2 GDP-by-state was released, but the key data is entirely in an infographic, without a text-layer. LlamaParse's agentic parsing tier understands documents like this by default, and faithfully records data, turning the page into a 52-row table. Source: https://t.co/0sBhstAIhM
LangChain
@LangChainScope agent memory to a single user or your entire team. Managed Deep Agents now supports user-level memory out of the box. Set it up with one file and deploy with a single command.
GitHub
@githubProject HydraFusion and new models arrived in the Copilot model picker. Have you tried them out? 👀 https://t.co/F2HnBFKo10
GitHub
@github💾 Save the date: GitHub Game Off returns November 1. The theme? █████ Join the jam now and be the first to know the theme when the clock strikes 13:37 PT. https://t.co/tXWnhlbjNh
AI at Meta
@AIatMetaRead more on our research blog: https://t.co/J0e9snbE8Q
AI at Meta
@AIatMeta6️⃣ Non-Associative Algebra: The team worked with Muse Spark to find an exception to a proposed rule about mathematical structures inspired by biology. They went further by developing an alternative characterization, which the researchers checked and refined. Read the paper: https://t.co/DTBm18sRpk
GitHub
@github📣 We’re giving away ONE general admission pass to #GitHubUniverse this year! 👉 US 18+. Ends October 5 2026. Enter here: https://t.co/MBM6Hjrnxk 📋 Rules: https://t.co/AOoXDM1RDn https://t.co/F3bShZkysT
School of Motion
@SchoolOfMotion5 Rive interactions you can add to a portfolio site! 🛠️ This week's blog post runs through 5 ideas of how to "zhuzh up" you portfolio website with creative interactions! Which ones are you going to try? Let us know in the comments! Read the full blog post here ➡️ https://https://t.co/4g39pumtg1 Want to master Rive? We JUST launched Rive Academy: Data Binding! Find out more ➡️ https://t.co/jg9o4fthLY
LangChain
@LangChainExperience the best of Interrupt NYC, now on demand. https://t.co/hL8WUWHbjG
Google AI
@GoogleAILearn more ↓ https://t.co/2RgktiPyOC
LangChain
@LangChainIlya Meyzin, SVP, AI Solutions & Data Science at @DunBradstreet is speaking at Interrupt, The Agent Conference by LangChain. Catch Ilya’s talk + more: https://t.co/GCNitijjvm https://t.co/PYX5MuomTH
Framer
@framerBefore / After Slider: Create a 640×360px before/after image slider with 16px rounded corners. Layer two full-bleed images, cropped to fill without stretching. Reveal the left image with a draggable vertical split, starting at 50%. Add a 2px white divider and centered 24×83px white pill handle with a pale-blue 6×35px grip and subtle border. Support click-to-position, mouse/touch dragging and keyboard arrows. No visible text. Expose images, initial reveal, optional grayscale and slim/round handles.
Framer
@framer3D Curved Ticker: Create a responsive, infinite 3D image ticker on black, 520px high. Place 250×340px portrait cards with 32px gaps and 12px corners on a continuously curved ribbon, with edges rising and receding. Auto-scroll left at 32px/s. Allow mouse/touch dragging with smooth inertia, then resume autoplay. Add white Inter captions (13px) bottom-left and two-digit indices (11px) bottom-right over a subtle dark fade. Hide navigation. Expose images, captions, speed, curvature, spacing and sizing.
LangChain
@LangChainLearn more → https://t.co/qCV4PV74fN
Framer
@framerHotspots: Create a 640×320px image hotspot component with 16px corners and three 44px numbered circles (01–03). Pale markers turn dark when active. Clicking one opens a 330px-wide floating panel near it, extending beyond image edges. Use translucent black, 12px blur and corners, a thin pale-blue border, white 18px semibold GT Standard titles, 14px body text and a close icon. Show one panel at a time; reclick its marker or close to dismiss. Use 0.5s transitions. Expose image, titles and descriptions.
Framer
@framerInfinite Canvas: Create an infinite 2D image canvas on black with a faint dot grid. Repeat 250×340px cards seamlessly in all directions, with 180px gaps, alternating columns offset 180px, and 12px corners. Start at 75% zoom. Support inertial drag, wheel panning, pinch/Ctrl-wheel zoom and keyboard navigation. Add white bottom captions and two-digit indices. Zoom images to 1.09× on hover over 600ms. Prevent link clicks after dragging. No toolbar. Expose images, titles, links, spacing and hover effects.