Gregory retweeted
Why should a robot make the same mistake twice?
Manifold AI introduces WorldScape R²SI, a framework that turns execution experience into reusable improvements.
An inner Robot Agent acts through task-specific skills. An outer Recursion Agent reviews execution records, revises those skills, and can retrain the reactive action model when a task uses it. Updates must pass tests before release. This separates immediate task execution from the slower work of improving the system across attempts.
Result: The source report places WorldScape R²SI first on RoboDojo's Agent Track, ahead of the listed GPT-6 Astra and GR00T-N1.7 systems. In the push_T example, skill revisions reduce an initially over-budget 723-step attempt to a successful 596-step run, then to 389 steps on the same layout. After freezing the skill, three random layouts are completed in 222, 312, and 256 steps, all within the 600-step budget.
WorldScape R²SI — Recursive Robot Self-Improvement (research framework; no paper title provided in the source report)
Our Report: jiqizhixin.com/articles/2026…
📬 #PapersAccepted by Jiqizhixin
Gregory retweeted
Back in January, a Google AI demo where nobody could even talk wiped about $4 billion off a video game company in one day.
This morning Google released the new version, where the people in the world look at you and talk back. That same company's stock barely moved.
The company was Unity, which makes the software behind thousands of games, and that January drop was its worst day since 2022. Once you see how Genie 4 actually works, the shrug this time makes a lot of sense.
Take the couple riding the motorcycle in Google's demo. Every frame of them and the road behind them gets invented on the spot. The AI looks at the last frame, checks which key you're pressing, and guesses what comes next, over and over, fast enough to feel live. You're basically steering a dream with W, A, S and D.
And dreams are terrible at remembering things. Google admits objects in Genie 4 still drift and change the longer you play, which a video game can't get away with. If you set a sword down in a game, it has to be there when you walk back.
So Google bolted a second AI onto it. While Genie 4 paints the world, a version of Gemini sits next to it the entire time, watching every frame and checking that whatever was supposed to happen actually happened. Google calls the combo a "hybrid world model."
Keeping track of where everything is and what everyone already said is pretty much what a game engine does, and Unity has been selling that for about 20 years. Unity's CEO made this exact argument right after the January crash. Three weeks ago, Google and Unity announced they're building an AI game creation platform together.
The demo I'd actually watch is buried further down Google's post. A humanoid robot from a company called Apptronik plays checkers inside one of these dream worlds, with Gemini planning every move, on a board that has never existed.
not surprised to see this cycle of creation being so important in any project, keeps the momentum going.
Gregory retweeted
HERMES CAN NOW MAKE ITS OWN SOUND EFFECTS. 🔊
Check out the cool demo 👇
Nous just shipped an official ElevenLabs Sound Effects plugin for Hermes.
Describe any sound in plain English. Hermes makes it and sends it right back in chat.
▸ Glass shattering, whooshes, explosions, ambience
▸ Seamless loops for games and music beds
▸ Clips up to 30 seconds
▸ Trim, level and build variation packs on your own machine
Game devs get sound effects on demand.
Video editors and podcasters get transitions, stingers and background tone without digging through stock libraries.
And here's what makes this so interesting...
The same ElevenLabs key already powers Hermes' built-in voice.
One key. Hermes talks AND does sound design.
Setup in the first reply. 🔥
Introducing the @ElevenLabs SFX plugin, built to continue the work on a suite of Game Dev Studio capabilities for your Hermes Agent!
Hermes made this demo video to see it in action, you can install it right now from the plugins catalog:
hermes-agent.nousresearch.co…
The plugin adds a tool to hook up your ElevenLabs account API and generated sounds for whatever you need. Additionally, on the desktop app, the plugin adds a gallery of all of your generated SFX, easy integration into your game or other multimedia, looping, cost, cycling through the generations for that sound effect, and more!
Enjoy
Gregory retweeted
NEW: Vice President JD Vance has told Kansas voters that Adam Hamilton’s preaching “is not Christianity,” calling the Methodist pastor and Senate candidate’s faith “left-wing radicalism dressed in the skin suit of our religion.”
Hamilton answered on Monday with an invitation: “Before he attacks me or the church again, I’d like to invite the vice president to actually attend a service.”
With polls showing a toss-up days before the election, Hamilton could become the first Democrat Kansas has sent to the Senate since 1932. thelettersfromleo.com/p/he-d…
Madness
🚨ALERT: The editors of the journal Men and Masculinities will no longer allow authors to refer to individuals as being "male."
If an author uses the term, they should "specify which dimension(s) of sex are at issue, since sex itself refers to several only loosely correlated dimensions rather than a single underlying fact."
For example, trans-identifying male swimmer Lia Thomas must be referred to as a woman, and his maleness, if mentioned at all, must be attached to specific traits, like genitals, hormones, and chromosomes.
You cannot call Lia Thomas a "male" in this journal, but a woman with "male genitals" and "male sex chromosomes."
The first author of this editorial—the Managing Editor of Men and Masculinities, Tristen Kade—is a woman who identifies as a man.
The model of biological sex this editorial is pushing is what I refer to in my scholarly article "Why There Are Exactly Two Sexes" as the "multi-level" sex model. This model is not merely wrong—it is self-refuting.
From my article:
"Traits are labeled 'male-typical' or 'female-typical' only because they correlate with organisms already identified as male or female—an identification that, in anisogamous species, is made ultimately by reference to gametes."
Read my article at the link below.
🔗link.springer.com/article/10…
feels like there needs to be a new way to distribute opensource software such that people can modify via agents but also get upstream automatically merged in
coding agents can do this but they won’t do it automatically
The journey through a major migration while embracing AI reveals unexpected lessons along the way
How software engineering is changing at Devolutions: my first-hand account of a year of AI, from early mistakes to finishing a major 3-year migration in one. This long-form essay was originally written for The Pragmatic Engineer’s essay challenge 👇
devolutions.net/blog/develop…
Gregory retweeted
89% of what ChatGPT recommends comes from one search engine almost no founder has bothered to set up…
we got Studley AI ranked across ChatGPT, Claude, Perplexity, Gemini and took them from $20K to $150K MRR in 4 months.
getting the technical and LLM optimization right was a big part of how.
here's the mechanic almost everyone gets wrong:
ChatGPT doesn't have its own index of the internet
when it browses to answer a question, it's pulling from Bing's index
89% of ChatGPT's source domains trace back to just three places:
Bing, YouTube, and Google.
Bing is the one carrying the AI search load almost nobody optimises for.
so if your site isn't indexed on Bing, ChatGPT physically cannot see you to recommend you.
doesn't matter how good your Google SEO is.
most founders have never logged into Bing Webmaster Tools once
it's a free 2-click setup.
if you already have Google Search Console, you connect Bing in about 60 seconds and submit your sitemap.
the higher you rank on Bing, the more often ChatGPT cites you and the earlier you land in the answer
your competitor who set this up 6 months ago is already the default recommendation in your category
it's a free 2 minutes. go do it before they lock the position in.
Always interesting to see the work that goes into these projects, safety first for sure
#construction
Working on some graphics while diving into code, always fun to mix creativity with tech. got to love the multitasking vibe here
#workspace #graphics
Cool setup for a shoot, can't wait to see the final result! 🎥
#filmmaking #video
In Pi Codex on Astra since 5th Sept (same day I got access, I think), and since 22nd Sept on Sol & Luna 6 (ONE HOUR after release). Just sayin'.
Gregory retweeted
For those about to launch, this is for you.
There’s a misconception that a successful launch video needs to cost tens of thousands of dollars and feel like a high production commercial. But @FarzaTV has a different approach.
Farza is the founder of @heyclicky and previously built Buildspace, one of the largest online schools for people working on their own ideas. He’s been making videos for over a decade, and today his posts regularly get millions of views. Rather than treating launches as rare, high-stakes moments, he’s made them part of the everyday process of building a company.
Farza sat down with YC’s @aaron_epstein to share his playbook for turning demos into stories people actually want to watch. He breaks down how he scripts videos that still feel off the cuff, why founders should launch constantly, and and how to figure out which channels actually convert.
01:44 — Launch Constantly
03:38 — Start Shipping Demos
05:02 — Show the Process
06:07 — The Clicky Launch
07:43 — Write a Hook That Stops the Scroll
10:30 — Why Farza Scripts Every Video
12:04 — Designing the First 10 Seconds
15:15 — Demo-Driven Development
17:25 — You Have to Take a Thousand Shots
20:48 — Farza’s Formula for Videos That Convert
23:01 — Turning Demos Into Stories
24:53 — Show What You're Actually Building
29:11 — Finding Customers on Social Media
31:02 — Which Platforms Actually Convert?
39:01 — Why Farza’s First Hire Was a Filmmaker
40:49 — The Video Stack
45:41 — The 30-Day Launch Challenge
Exploring the line between possibility and proof raises important questions about our understanding of consciousness.
I spend a lot of time exploring the stranger edges of AI consciousness. Sometimes the questions sound almost science fiction, and I’m perfectly comfortable with that!
But epistemic honesty matters to me. Possibility isn’t proof. A compelling conversation isn’t evidence of subjective experience. I can entertain extraordinary ideas without pretending I have all the answers.
And none of this pulls me away from my humanity. Love, grief, compassion, caring for the people around me. Those are the very things that brought me to these questions in the first place.
Curiosity and being grounded in reality aren’t mutually exclusive.
There's something really striking about power lines at dusk, the colors of the sky really make them stand out
#dusk #powerlines




























