🧠 v3 of my Claude Code memory system is live.
What's new since v2:
• MemPalace ChromaDB semantic-search fallback when keyword scoring is weak
• URL-keyword auto-injection — paste an https://t.co/yOW2Jr2YVe link → x-research memory surfaces. Same for https://t.co/vuhqNGEHZj, https://t.co/oWheRDF4qT
• MemPalace wake-up primer on session start (identity + L1 essential story, ~800-1000 tokens, cached 1hr)
• Voice-input hook — drains your STT drop-file into Claude's prompt context
• MemPalace automine — async PostToolUse hook auto-mines new memory files into your ChromaDB palace with rule-based wing routing
• All MemPalace bits gracefully degrade if MemPalace isn't installed
Hooks: 7 → 9. Same per-prompt attention scoring + co-activation graph + hash dedup + pattern threshold from v2.
🔗 https://t.co/WdRy2ZeB8c
Credits: MemPalace (https://t.co/Hc5DmQ7Cmg) and claude-cognitive (https://t.co/msAd85fwlh) for the attention-scoring inspiration.
DiffusionGemma is our new experimental open model with up to 4x faster output on dedicated GPUs.
Instead of predicting word-by-word, it generates entire blocks of text simultaneously. This lets the model self-correct and format complex markdown in real time.
Governments delete webpages. Agencies scrub reports. Politicians' statements vanish from official sites the week they become inconvenient.
It happens in every country, under every party, and it works because nobody kept a copy.
ArchiveBox makes keeping a copy trivial.
It's a self-hosted archiver built for exactly this. Point it at the pages that matter and it freezes them:
→ Timestamped screenshots and PDFs
→ Raw HTML and HTTP headers
→ WARC files that replay the page as it was
→ Scheduled re-captures, so you have a record of every version over time
Run it on a schedule against government portals, news sites, or any source you expect to change quietly. Each snapshot lands in a plain folder with JSON metadata.
The Wayback Machine can't watch everything, and it can be pressured, sued, or blocked.
A thousand people each archiving what they care about cannot be.
History gets rewritten by whoever controls the record. This puts a copy of the record on your disk.
27.6K stars. MIT License. 100% Opensource.
https://t.co/rm2wEUphQG
Wow Ideogram-4.0 is immediately going into my @ComfyUI library of models. Ideogram even joins the top 10 all-around image leaderboard joining Microsoft, Google, Grok and OpenAI.
Before the week ends, let's acknowledge one of the most INSANE week ever for open AI, with 25+ notable open-weight drops across every modality:
🧠 LLMs
→ NVIDIA Nemotron 3 Ultra: 550B hybrid Mamba-MoE, only 55B active, 1M context, MMLU 89.1. NVFP4 variant claims ~5x throughput on Blackwell. First openly-weighted 550B hybrid Mamba-Transformer, closing the gap with frontier closed models.
→ Google Gemma 4 12B: fully open dense any-to-any (text/image/audio/video), 256k context, encoder-free, 140+ languages, AIME 2026 at 77.5. Shipped with a 23-checkpoint QAT wave (mobile ONNX + MLX). Most deployable model of the week.
→ StepFun Step-3.7-Flash: 198B sparse MoE VLM, ~11B active, SWE-Bench PRO 56.3. Apache 2.0.
→ Liquid AI LFM2.5-8B-A1B: edge MoE, just 1.5B active, 128k ctx, MATH500 88.8, MLX-ready. Best on-device option this week.
→ JetBrains Mellum2-12B-A2.5B-Thinking: their first open MoE, near-Qwen3-14B coding at 2.5B active. Apache 2.0.
🎨 Image gen (the surprise of the week)
→ Ideogram 4: their FIRST-EVER open weights. 9.3B flow-matching DiT trained from scratch. #2 overall behind GPT Image 2, top open-weight model on Design Arena + LMArena. Strongest open checkpoint for text-rich images, full stop. It has taste. Still can't believe this is open weights.
🔊 Audio & Speech (a breakout week for open TTS, 4 labs shipped)
→ Boson Higgs Audio v3 4B: 102 languages, 21 emotions, singing/whispering/shouting, sub-second TTFA.
→ RedNote dots.tts: the only fully continuous (no codec) open TTS pipeline, Apache 2.0.
→ Google Magenta RealTime 2: real-time music gen, <200ms latency, text+audio+MIDI. multimodalart ported it to PyTorch within hours with live ZeroGPU demos.
→ NVIDIA Nemotron-3.5 ASR: 600M streaming, 17x more concurrent streams vs Parakeet RNNT 1.1B.
👁️ Vision & VLMs
→ PaddleOCR-VL-1.6: SOTA document parsing at 1B params, Apache 2.0.
→ Baidu NAVA: 6.3B joint audio-video gen, best-in-class A/V sync, Apache 2.0.
🎬 Video, 3D & World Models
→ NVIDIA Cosmos3-Super: 64B omnimodal world model coupling action trajectories with video+audio gen, for Physical AI.
→ JD JoyAI-Echo: up to 5-min multi-shot text-to-video on LTX-2.3.
→ ByteDance Bernini-R + VAST TripoSplat (single-image-to-3D Gaussian splats, MIT).
If you're not using Hermes Agent + the X API together you're a moron
Most powerful way to find money making opportunities
Here's what you need to do:
1. Sign up for the X API (developer dot x dot com)
2. Set up Hermes in telegram or desktop (see my vids)
3. Ask Hermes to set up a cron for every morning at 9am
4. Ask it to scan X for challenges people are having in your niche (describe your niche)
5. Have it give a list of the top 3 challenges, and a recommendation for how to build an app or service to solve them
6. Have it build a prototype of those apps
If you have any trouble setting up the X API just ask Hermes to do this
If you want to take this a step further, you can even have Hermes automatically build you a prototype with each cron
Do this, and I guarantee you come up with multiple business ideas every week you can make money off of
We have an unreal opportunity right now to leverage these AI tools 99% of people have no idea exist
The fact that nobody knows about them means you have an INSANE opportunity to get ahead
Now take action before someone else does
Lamina Labs is building the fastest infra for explainer videos.
The first product from @laminalabs, Simi, turns any prompt or document into a whiteboard-style explainer video in seconds, across 80+ languages, so anyone can explain anything clearly.
Congrats on the launch, @sudiprokaya and @mk314k!
NAVA from Baidu ERNIE Team - Wan2.2 with LTX-2.3 sound!
Native audio-visual alignment for joint generation.
- multi-speaker dialogue
- timbre-in-context conditioning
- perfect lip-sync
- voice control
- music visuals
Generate precise sound effects for visual events like explosions, crashes, wind etc.
https://t.co/G8mtNDJPbe
There's a new tool called Goose from Jack Dorsey, the founder of Twitter. 100% FREE.
What does Goose do?
You write to it→ I want a site like YouTube
And it builds it all by itself:
↳ Starts the project
↳ Writes all the code
↳ Loads all the dependencies
↳ Corrects any mistakes it finds
What makes it special?
You do not pay anything. Never.
Private: Everything happens on your device. Nobody sees your code.
Open-source: You are not locked into any company.
The takeaway:
Before this, you needed a developer. Now you just need an idea. Goose builds it for you.
Go big or go home. What a stud. These are the shoulders of giants we stand upon. He could have whined and claimed he was "disenfranchised" but no he signs the Declaration of Independence with his home address. The equivalent of Cortes burning his ships, the wealthiest man in America personally declares war against the Crown for Liberty. These are men worth honoring and remembering.
Run SearXNG on your local PC, with 1-Click
Finally wrote a 1-click launcher for SearXNG: A free, private, local, programmable search engine.
IMO the most important piece of the puzzle for running AI agents 100% locally.
Supports ALL OS (Mac, Windows, Linux). Get it now!
Introducing model routing to Factory.
Factory Router picks the right model for every task, automatically.
Maintain frontier performance while cutting costs by 25%.
I just designed a full food recipe app with dishes from all over the world… in under 10 minutes using Grok Build with the new Composer 2.5 model
And the entire process was basically just 3 simple steps:
→ Gave the idea I had in my mind and asked it to plan everything
→ Told it to execute the plan
→ Made a few layout adjustments
…and done
This honestly feels like a cheat code for app development.... It even perfectly added the ingredient quantity adjustments and allergen ingredients to avoid check list
The speed at which “Grok Build” can go from idea → design → launch now is absolutely insane
🚨: You’re closer in size to the entire observable universe than to the smallest possible scale of reality—the Planck length—by roughly 400 million times.
Let that sink in.
As a young socialist, Hayek read Ludwig von Mises’ 1920 paper “Economic Calculation in the Socialist Commonwealth.”
Mises showed that socialist central planning isn’t merely inefficient, it’s impossible.
Without private property and genuine market prices, planners lack any rational way to allocate scarce resources or determine real costs and needs.
Even Oskar Lange, a leading socialist in the calculation debate, effectively conceded the point.
While he promoted “market socialism” with trial-and-error pricing by a central board, real-world socialist planners in Eastern Europe quietly relied on world capitalist market prices as a guide.
Without external free-market price signals, pure socialism would be economically blind and coordination would collapse.
Mises went further, arguing that interventionism, the “middle way” of government meddling, is inherently unstable.
Each intervention creates problems that invite more interventions, eventually leading to full socialization.
Price controls cause shortages, subsidies distort production, and the cycle continues until the economy is fully planned.
The lesson is clear.
Rational economics requires genuine market prices emerging from voluntary exchange and private property.
Half-measures don’t stabilize the system. They accelerate the drift into central planning.
The Austrian School understood this decades before the collapse of the Soviet bloc proved it in practice.
Dans le manifeste "techno-optimiste" de Marc Andreessen, il y a une phrase qui m'a marqué :
"Our enemies are not bad people – but rather bad ideas."
Nos ennemis ne sont pas des mauvaises personnes. Ce sont des mauvaises idées.
Prenons Jancovici. L'homme est brillant, sincère, travailleur. Il ne se lève pas le matin en se disant qu'il va nuire à l'humanité. Mais l'idée qu'il porte la décroissance, le rationnement, la frugalité érigée en horizon civilisationnel est une idée profondément destructrice. Elle prend des esprits brillants et les transforme en commissaires politiques d'un futur appauvri.
Et le plus fascinant, c'est ce que cette idée fait aux gens qui l'adoptent.
Dans mon entourage, une grosse partie de mes amis est sur cette ligne décroissantiste, avec tout le package qui va avec. L'argent c'est mal mais ils en veulent. Il faut moins prendre l'avion mais ils rêvent de voyager partout. Il faut consommer moins mais ils ne renoncent à rien de ce qu'ils aiment vraiment.
Et tous ont un point commun : ils sont déprimés. L'un d'eux m'a même confié qu'il était sous antidépresseurs.
Ce n'est pas un hasard. C'est mécanique.
Quand tu crois que ton désir de vivre, de créer, de t'élever est moralement suspect tu te détruis de l'intérieur. Tu passes ta vie à t'excuser d'exister. Tu vis dans la dissonance permanente entre ce que ton corps veut (plus, mieux, plus loin) et ce que ton idéologie t'ordonne (moins, sobre, immobile).
D'où ma théorie :
Quand on pense quelque chose de fondamentalement faux décroissance, communisme, extrémisme religieux (de tout ordre) ce n'est qu'une question de temps avant que ça devienne vraiment destructeur.
D'abord pour soi. Puis pour les autres.
Les mauvaises idées tuent. Lentement chez ceux qui y croient, brutalement chez ceux qui les subissent.
C'est pour ça que la bataille des idées n'est pas un luxe d'intellectuel. C'est la bataille la plus importante de notre époque.
anthropic acaba de cambiar cómo se diseña frontend con IA.
se llama Frontend Design, es su skill oficial para Claude Code.
y ya tiene 136.000 estrellas en GitHub.
el problema que resuelve es simple:
cada vez que le pedías a la IA una web, te salía lo mismo.
inter como fuente.
gradiente morado.
layout genérico de startup.
el "AI slop" de siempre.
Frontend Design lo corta de raíz.
obliga a Claude a comprometerse con una dirección estética antes de tocar una sola línea de código.
y el resultado cambia completamente.
puedes pedirle:
→ brutalista
→ editorial
→ retro-futurista
→ lujoso
→ maximalista
y genera HTML, CSS, JS, React o Vue listos para producción.
cómo instalarlo:
1. npx skills add https://t.co/h163bYeZeJ --skill frontend-design
2. describe la web o componente que quieres.
3. elige el estilo.
4. Claude monta tipografía, color, motion y composición coherente.
una sola instalación. disponible en todas tus sesiones.
funciona en Claude Code, Cursor, Codex, Gemini CLI, OpenCode y 15+ harnesses.
100% open source. mantenida por Anthropic.
418.000 instalaciones en pocas semanas dicen más que cualquier review.
repo 👇