全部 AI 动态

man I love telling ai to take a large piece of work and break it up into small stacked PRs that you can land individually it lets you build something big quick but then you can do a final pass to make it production ready by focusing on individual pieces in order breaking it up was so much work before

Exciting news: Qwen3.8-27B by @Alibaba_Qwen just landed in Code Arena: WebDev at #9 overall with 1595 pts. It is the only model in its size class in the top 10, and also reshapes the Pareto Frontier! It is only 6 ranks behind the much larger Qwen3.8-Max. For scale: Gemma 4-31B which was released back in April sits at #80. Congrats to @Alibaba_Qwen on the release! See thread for more highlights on Qwen3.8-27B.

Grok Bot is insane. I asked it to find every video where Dr. Roger Seheult discusses sunlight, red light, infrared light, and mitochondria, including MedCram videos and interviews with Andrew Huberman, Eric Berg, Steven Bartlett, and others. A few minutes later it had: • Found and transcribed 38 videos into one Word document • Picked the 13 best clips explaining the science • Cut and combined them into one 19:35 video I can watch straight through • Created a markdown file summarizing each clip, the science it covers, and the exact timestamp in the original video No scripts. No code. I just told it what I wanted in plain English, and it turned days of work into minutes. As a non-coder, this is the first AI agent that actually feels like a superpower.

Grok Voice Think Fast 2.0 is now #1 on the Artificial Analysis Speech-to-Speech Index. This index measures whether voice agents can reason over the speech it hears, resolve real customer issues, and correctly complete tasks using agent tools.

Qwen 3.8 27B Q4 quant destroyed Q8 at voxel island creation ⛏️ We gave our Atomic Dynamic Q4, Q5, Q6 and Q8 quants the same voxel island tasks: - Cliffside Hamlet - Autumn Road - Winter Cabin - Jurassic Volcano Isle - Pirate Cove Sunset - Overgrown Ruins - Steampunk Sky Island Q8 wasn’t always the best, with Q4 often times producing a comparable or better result, which shows that lower quants are also a viable and more accessible option. AD-Q5_K_M is our top-pick, it fits on a 32GB MacBook Air with 32K content and picks the same next token as the BF16 original 97.3% of the time

ex-Apple engineer gave Grok 4.6 two jobs inside Cursor, went to sleep, and opened the results live the next morning: no babysitting, no checking every generation, just a model left running on real work for hours. • 00:42 - reveal the website Grok redesigned overnight • 23:21 - go from a voice prompt to a full software stack • 45:43 - inspect the generated code + architecture • 01:21:26 - Grok 4.6 vs Opus 5 • 01:46:31 - live PR + Cloud Agent workflow Most coding demos test an AI for 5 minutes. This tests the thing that matters for agents: can the model keep working when you stop watching it? SpaceXAI built Grok 4.6 specifically around longer-running agent tasks, coding and more ambitious visual work. The endgame isn’t prompting faster. It’s giving an agent a job at night and reviewing finished work in the morning. Worth watching before you choose which model runs your overnight agent loops.

http://x.com/i/article/2091179277962113024

Kimi K3 on my shader test: "create a visually interesting shader that can run in twigl-dot-app make it like an infinite city of neo-gothic towers partially drowned in a stormy ocean with large waves." "Make it better" Very good model, not Sol Max or Fable, but great open weights

Alright, it happened. After a slow start with Grok Bot I just had my mind blown. Twice I did a screen recording with audio of me cleaning up my desktop after a week that resulted in hundreds of random files that all needed particular sorting It's a monotonous task I've always wanted to outsource but didn't really trust other platforms to give that level of access But the mind blowing part was how my Chief of Staff LEARNED not just what I do but HOW and WHY I do things From a screen recording with my voice giving instructions...and because the video file was large my CoS automatically compressed the file under the limit to watch it It then gave me feedback on what to change for how I record part two so it can learn even better And there's more. I planted a quick question in my video recording to see how closely it was watching and listening. You can see my CoS asking me the question in the second screenshot, it's the last chat bubble Soon enough I'll have this painful task automated away. And undoubtedly many, many others in time as well. Thank you @bot This really will be the entry point for many and the beginning of a true paradigm shift. I also got my wife set up with Grok Bot today. This platform is easily accessible by anyone and everyone

a skill people at Anthropic have been using a lot recently: ELI5 /eli5 <what you want explained> "explain like I'm someone who knows nothing about this topic, using a HTML artifact with big pictures and few words"

I launched Grok Bot for 24 hours, gave it 3 jobs, and stopped treating AI like a chatbot. I had it scan live X + web data for emerging AI products, research what people were actually saying about them, filter the obvious engagement bait, compare sources and turn the strongest signals into a structured brief. At the same time, I could have separate research paths looking at competitors, code or market opportunities instead of forcing everything through one conversation. That’s the difference I underestimated. I don’t want to sit inside a chat writing another prompt every 5 minutes. I want to define the objective, give the agent access to the right tools and data, let it execute, then step back in when judgment or approval is actually needed. The old workflow was prompt → answer → copy → repeat. The workflow I’m moving toward is assign → execute → verify. Once AI stops waiting for your next message, it stops feeling like software you use and starts feeling like work you delegated.

one of my favorite interactions i’ve been playing with lately on @ChatGPT: long press the +, drag to a recent photo, and let go to attach it instantly you can even select one while the photos are still flying out and and it'll morph right into the composer lmk what y'all think!

Grok Voice Think Fast 2.0 just ranked #1 on Artificial Analysis’ new Speech Agent Arena for highest Task Success Rate This benchmark actually measures what matters Real people talk to hidden voice agents across practical scenarios. Task Success Rate tracks whether the AI understands the request, calls the correct tools, and successfully completes the job It’s not just about sounding natural. It’s about whether the voice agent can actually get things done Grok finished #1

I am going to start to share the leaderboard at VulcanBench more regularly. VulcanBench is the only benchmark I know that benchmarks across effort levels, on an eval suite of 100% real engineering tasks. Getting ready to sunset eval suite 3 and eval suite 4 is almost ready to go. Here's the current leaderboard for Eval Suite 3, Grok 4.5 High is the highest scoring model still with Fable 5 Low, yes Low, right behind it. What I've been able to uncover with VulcanBench is that way too many people are running models at Max effort, thinking that buys them more accuracy, when really it just costs more and uses more tokens so takes more time. If you're still in the mode of, tell my agent to do something then go get coffee, you're probably still living in the past. You can move faster, with higher accuracy, the key is not thinking you need Max effort all the time.

Grok Bot might be the first tool that lets one non-technical person run an entire business with a team of AI agents. My friend Billy runs his whole newsletter business on Grok Bot agents, and I think we're about to see 100,000+ businesses like his. BEST PRACTICES: 1. The agents run on a shared cloud computer, so running your newsletter, your X, and your receipts all in one place creates context bloat and burns tokens fast. One mission per setup. 2. Start with a Chief of Staff. Give it access to your existing docs (Notion, Slack, Gmail), have it audit the business, then tell you the top three agents to build first to drive revenue. 3. Perfect a task with the Chief of Staff before spinning up a new agent. Have it do the outbound sales once, review it, and only then say "now build a bot that does exactly that." You earn each new hire by proving the task works first. 4. Constraints are the feature. You get a limited number of agents, one thread per bot, like DMs with a teammate. It forces you to stay mission-oriented instead of spinning up a bot for every random idea. 5. You make the decisions, not the agent. Billy's team spent three weeks unable to pick where content should live. At some point you say "we're doing Notion, no more tinkering" and move on. 6. Run week one with no new agents. Build the team, learn to fly the plane, just execute. Week three is when you find the real gaps and expand, someone to man the inbox, someone for the Shopify shop. 7. Then add routines so it works while you sleep. Ask your Chief of Staff what recurring jobs would move the business forward overnight, and it builds the automations that run without you. Thanks to @billyjhowell for sharing the sauce on @startupideaspod (follow for more). Grokbot is really cool. Watch below: http://youtu.be/qQluNEfSVHk?si=3L2cAB71NQSuqxWU

BREAKING: Grok 4.6 just took the #1 spot on CursorBench 3.2 — while delivering a massive efficiency advantage. ⚡💻 • Grok 4.6 Extra High — 70.8% | $2.81/task • Fable 5 Max — 70.5% | $17.32/task • Opus 5 Max — 70.0% | $8.23/task • GPT-5.6 Sol Max — 67.2% | $5.69/task Grok achieved the highest score while costing roughly 6× less than Fable 5 Max and nearly 3× less than Opus 5 Max per task. For AI agents, raw intelligence is only part of the equation. The ability to maintain high performance across long coding tasks without burning massive amounts of compute could be a major advantage. Grok’s agentic coding efficiency is becoming seriously impressive. 🚀 Source: CursorBench 3.2

I asked GPT-5.6 Sol to create the most Claude-y possible parody image and what it came up with is pretty great and dead-on.

One of the best things I gave my Grok Bot... Was its own email address. Now it can: → Send emails → Receive replies → Verify accounts → Sign up for websites → Use email inside its own workflows Takes 2 minutes... Now it can do a LOT more without me.

Grok 4.6 just took the #1 spot on CursorBench 3.2.....and the efficiency is insane Here's the cost comparison: • Grok 4.6 Extra High — 70.8% | $2.81/task • Fable 5 Max — 70.5% | $17.32/task • Opus 5 Max — 70.0% | $8.23/task • GPT-5.6 Sol Max — 67.2% | $5.69/task Grok achieved the highest score while costing roughly 6X less than Fable 5 Max and nearly 3X less than Opus 5 Max per task That’s what makes Grok so powerful for agents Top-tier intelligence is great.....but top-tier intelligence that can keep working across long coding tasks without burning ridiculous amounts of compute is even better Grok’s agentic coding efficiency is insane

Our general-purpose coding agent just scored 100% on the ARC-AGI-3 interactive reasoning benchmark. NVIDIA AVO completed all 183 levels across all 25 public environments, figuring out what to do with no instructions, explicit rules, or stated goals.