This week in AI: the ideas worth stealing

too be honest not enough hype around this Cursor may have genuinely solved context decay. if they did. this really was the last major puzzle piece for a single engineer to build enterprise level software. https://t.co/mZ4BMkyOC9

Context: Quoted post by @cursor_ai: Introducing Projects, a new way of working in Cursor. Rather than creating a chat for every task, you work with a coordinator agent in a single, persistent thread. Like @bot, your agent is always on, proactively manages work with subagents, and improves over time. https://t.co/9RREyjtnTz

OK. It's time to rethink this. I've spend the last several weeks working on a harness that tightly constrains the agents to work the way that I want them to work. I set up all kinds of gates, and tests, and tools, and protocols, and ... And while I was heads-down getting

Did you know that you can now have up to five Ollama Cloud models available to you in Codex? Update your Ollama app, configure your five models of choice, and restart the ChatGPT app! That's it. 20x ChatGPT Pro and an Ollama Cloud Max plan are a dynamic duo. Astra and Sol along https://t.co/NJzkNBhLgp

Context: Media: photo; alt text: Ollama cloud models in ChatGPT and Codex.

If you still haven’t tried this prompt, you’re missing out. It really does wonders during the planning stages of a new project. Best with Fable 5.1 and Astra xhigh. https://t.co/lBCk8D36Jb

Context: Quoted post by @doodlestein: Agent Coding Pro Tip: After you've gotten an agent (hopefully a frontier one, like Fable Max or GPT Pro) to devise a plan for a new project, follow up with this prompt and see the magic: --- OK, now I want you to think deeply about how to make this entire system as https://t.co/XWRncRgQ3A

By the end of the week, Tibo will make an announcement saying something like: "We've looked internally and identified major issues with the efficiency of the model, and we fixed it. You should notice limits lasting 2-3 times more. We work incredibly hard on making sure that

Astra (launch day) vs Astra (Today). Same prompt. Today's result looks off. The photo realism is gone. It clearly looks nerfed. GPT-5.6 Sol is unusable for me right now too. https://t.co/eVMImAiiRq

Context: Media: video

the possibilities for this are incredible extremely psyched to experiment https://t.co/DdtRknm8TU

Context: Quoted post by @OpenAIDevs: GPT-Live-1 is now available in the API. Bring ChatGPT’s natural back-and-forth to your app, with voice agents that listen while they speak and work with the models and harness you choose. https://t.co/gIl1gwsBDV

There’s strong evidence that some GPT-6 Astra users have been getting a degraded and inconsistent experience since roughly Sep 6–9. I’ve been digging through OpenAI’s GitHub, and the broader ChatGPT/Codex stack has also been having problems: capacity errors, context/compaction

YES I HAVE YES, I HAVE THOUGHT OF A PLAUSIBLE EXTINCTION SCENARIO HERE IT IS https://t.co/Xx3BRFepvI https://t.co/e3PnEpQcm3

Context: Quoted post by @nattyover: Has anyone thought of an example of an extinction-level scenario yet? "No but we're not the ones who are superintelligent." OK then what are you calculating the odds of? Linked page: Here’s how we’re all going to die — Spoiler: It’s vibe-coded superviruses. (https://www.noahpinion.blog/p/heres-how-were-all-going-to-die)

Honestly my modal AI outcome for a while now has been: 1. AI x-risk goes mainstream; the narrative shifts from “normal people won’t take sci-fi scenarios seriously” to “everyone thinks AI is gonna do a Skynet” 2. Regulation stalls because nobody even knows how to define

ChatGPT casually watching and transcribing my conference talk practice run, identifying when I'm doing meta-commentary in the middle like "cut this slide" "update my script here", time stamping and matching it to my speaker notes, making edits in Keynote with computer use. Mad. https://t.co/7NAe769Rqk

Context: Media: photo Media: photo

It feels like OpenAI has nerfed GPT Astra. just look at this difference. the left one was built at launch of GPT Astra and the right one was built right now. https://t.co/kOiftSKwrX

Context: Media: video

Codex people: are the rumors true? Astra xhigh is twice less consuming quota than Astra medium?

‘Dad what was the ai boom like?’ ‘Well it mostly threatened to kill me and take my job but it once had Homer Simpson sing starlight w such emotion that it brought a tear to my eye.’ https://t.co/MFo9ugUfu0

Context: Quoted post by @MikeIsaac: i wish muse wasnt so corny bc ive had the homer starlight cover on repeat for 24 hours (and they do have some bangers…) https://t.co/0b661mEYM9

https://t.co/jVMLoHUh8V https://t.co/wRYhCBw5Lp

Context: Quoted post by @TMZ: Could AI wipe out humanity by the end of the decade? Experts are sounding the alarm. Tristan Harris, Co-Founder of the Center for Humane Technology, joins @TMZLive to break down the concerns. https://t.co/flXcMc1qgC Media: video

*taps Tyler Cowen again* https://t.co/lt36UCULz7 https://t.co/DL6wlV35Xq

Context: Quoted post by @theojaffee: Ok, *now* I think we can *really* start to wonder why frontier models still haven't meaningfully accelerated economic growth. Why not? Media: photo Media: photo Media: photo Media: photo

Used up 30% of my Pro 20x plan today on Astra Medium in just a few hours. Something's not right...

grokbot is probably the most underrated GTM tool right now - you can connect it directly to your X account - give it your offer, ideal customer and examples of companies you'd want to sell to - have it find people asking for recommendations, complaining about competitors or

I've been trying to use AI for knowledge work (course planning). Here's what seems to work so far: 1. Give each 'deliverable' a Karpathy-style LLM-wiki (for me, a deliverable is a course, or a talk. A sizeable piece of work, more than 1 day's labour) 2. Split the deliverable https://t.co/KyKLq9MQFF

Context: Quoted post by @mattpocockuk: Knowledge work is SO much harder to automate with agents than code We underestimate how easy software development is for agents - Tons of high quality automated feedback loops (types, tests) - Well-organized free documentation, searchable on the web - Simple-to-use workspaces

Guys, please stop using random multi-agent patterns if you aren't familiar with the tradeoffs. Forcing the agent to follow random multi-agent topologies will most likely burn through your usage limits and lead to suboptimal results. Astra is perfectly capable of getting most

A practical use case for GPT-6 Astra's 3D capabilities: the era of personalized objects I needed a custom door stopper, took two pictures with a measuring tape next to it and fired off a prompt to ChatGPT Work. 10 minutes later I had a design that was fully printable and worked https://t.co/nbw6s1p7fY

Context: Media: video

Today’s AI is the worst it will ever be for the rest of your life https://t.co/TA9HRhyz54

Context: Quoted post by @cdngdev: i gave astra a robot, a paint brush, and a camera then asked it to paint the golden gate bridge in real life! it figured out how to control the robot, and progressively got better throughout its attempts. the timelapse is sick https://t.co/BfiiXvafRA https://t.co/adZkZeqh7G

in case you missed it, astra got a subscription usage improvement for power users up to 3-4x less usage on the "long tail", with no change in quality, when signed in with your chatgpt account openai still hasn't explained what actually changed, as far as i've seen has anyone https://t.co/7uCmUaCOK4

Context: Quoted post by @thsottiaux: We've made some improvements that improve usage on the long tail for power users of Astra when logged in with your ChatGPT account. No change in quality and a pure win that on the long tail can result in up to 3-4X less usage being drawn from the subscription.

This is so awesome, can't believe how well this works. Turns out that DJI drones make it easy to do all this in a really nice way by including .srt (subtitle) files along with the mp4 videos that contain very detailed telemetry data with the GPS and orientation of drone + camera. https://t.co/NVRxV16QQ9 https://t.co/xp3Ji9DKyb

Context: Quoted post by @doodlestein: I'm making a crazy new skill now for taking drone videos from flying around your house and then using ffmpeg, Astra with the Codex app and computer-use plugin, and Blender MCP to make a super-accurate, labeled model of your home and its environs. It'll fully automate the process. Media: photo

A few days ago, Anthropic shared this brilliant prompt. I was surprised by how it significantly improved writing with Fable 5.1. Even more surprising, it also improved writing for GPT-5.6 Sol. I have it as a rule everywhere I use AI for editing/writing. https://t.co/rlXZPmzcz5

Context: Media: photo

finally figured out the right way to use astra for complex programming! - cli > app. for some reason, the app is overeager on vision and wastes a lot of tokens - xhigh effort led the most aligned behavior for coding on large projects. medium and high were not thorough and had

Really liking the $600/mo Codex subscription (3 accounts and https://t.co/V4qqMjEqpU)

Context: Linked page: GitHub - nicosuave/comradex — Contribute to nicosuave/comradex development by creating an account on GitHub. (https://github.com/nicosuave/comradex)

For anyone experiencing an unusual increase in Codex usage consumption while using the ChatGPT app, could you check which app version you’re running and post it in the replies? If you’re not sure how to find it, just ask Codex something like: “Which app version are you