this is not a joke, this is literally what happened this week, bonsai 2 on a 12gb 3060 built a whole game through hermes agent overnight, the full 5 hour build is below 🧵
Why has France, for decades, failed to create a single large tech or finance company?
No large hedge funds. No large tech companies. No meaningful AI companies.
No homegrown Google. No Facebook. Not even an Instagram or WhatsApp.
Its national LLM champion, Mistral, failed fast.
What’s going on?
Genuine question.
ENTRETIEN. Allemagne, Ukraine, États-Unis, Chine : les dernières prophéties d'Emmanuel Todd sont à découvrir dans Marianne, en kiosque dès jeudi !
👉 https://t.co/VvB227jB05 @frederictaddei
if 12gb of vram is the entry point of the local ai quest, then 2x dgx spark with 256gb of unified ram or vram is the point where you don't miss the frontier anymore.
now i am running qwen 3.8 flash next fp8 on my 2x dgx spark with vision and full context, it builds insane stuff you have no idea of and it's better at frontend design than any of the humans i've met.
i know because i run this every day on my desk, run experiments and put it through things you could not imagine yet, and most of them i cannot post, so what i post is games and builds and stacks. so here is hermes agent running on telegram, working on the build i am working on to demonstrate.
Clairement le meilleur move de ma vie entrepreneuriale.
Quel plaisir de voir Hermès fouiller dans la banque, les mails ou sur Amazon / Aliexpress / Ebay pour retrouver et réconcilier tous les justificatifs.
❌ Destinos de viaje que NO valen para NADA:
• Bali 🇮🇩 (tráfico y postureo)
• Dubái 🇦🇪 (centro comercial en el desierto)
• Maldivas 🇲🇻 (aburrimiento en un resort)
• Mikonos 🇬🇷 (trampa para turistas)
• Tulum 🇲🇽 (postureo e inseguridad)
✅ Destinos que SÍ merecen la pena ↓
Run open models like Gemma 4 completely offline in the Antigravity SDK.
Built on Google AI Edge’s LiteRT, you can now run Gemma 4 directly on your local GPU. Zero API costs, total data privacy, and no internet required.
This is amazing, but that 70GB is compressed.
Decompressed, the resolved TeX contains lots of info...
📚 2.856 MILLION papers
📝 ~247 BILLION characters
🧠 estimated 78.6–80.9 BILLION tokens
So obviously you aren't stuffing arXiv into a giant context window. 😂
But Local AI can ...
🔎 search/filter locally
📚 retrieve a handful of relevant papers
🧠 feed only those chunks to your local LLM
🚫 no cloud/API required
The metadata for ALL 3.15M papers is only 1.6GB. 👈
So you could keep a tiny local index for discovery + the 70GB corpus for retrieval + a local model for answering.
Basically, it's your own offline scientific search engine sitting on an SSD. 👀
Currently reading Sailor who Fell from Grace with the Sea and the front cover tore off after accidentally getting wet from the sea as I’m reading Part two when Ryuji is being torn from his freedom.