prototype_27 retweeted
La guide touristique qui fait semblant que tout va bien alors que Paris est en feu 😂😂😂😂😂😂😂
prototype_27 retweeted
Un couple de touristes japonais à Paris le 17 octobre qui se sont retrouvés sous les gaz lacrymogènes en sortant de leur visite du Musée du Louvre #17octobre
prototype_27 retweeted
🚨🇫🇷 ALERTE INFO !
La situation reste TRÈS TENDUE à Paris 🤯
Des véhicules de police sont PRIS POUR CIBLE par la foule dans le secteur de la place de la République.
(@LucAuffret)
prototype_27 retweeted
An Nvidia researcher couldn't stand how lifeless VTuber voices sounded.
So he trained a voice model on a single 4090 GPU in his bedroom. A year later, his company says it had $21 million in annual recurring revenue.
The company is Fish Audio, legally Hanabi AI, founded in 2025 by Shijia Liao and Rissa Cao. Liao, the chief scientist, was a video research engineer at Nvidia. Cao, the CEO, spent years in voice AI at Amazon Alexa and Meta.
Liao open-sourced his model as Fish Speech, which passed 31,000 stars on GitHub. It spread through game developers, anime fans, and people dubbing characters for fun.
Cao's diagnosis of text-to-speech is that getting the words right and getting the delivery right are different problems. "A voice built to read never learns to perform," she wrote.
So Fish Audio controls the delivery. You type a script with tags like [whispering], [laughing], [sighing], or [pause], then clone a voice from a short clip or pick one from a library of more than 2 million voices. The company says it has 15,000+ natural-language controls, and that listeners preferred its S2.1 Pro model 66% of the time in blind tests against leading competitors.
HeyGen uses its voices for avatars, and voice-agent companies like LiveKit use them for calls. Fish says enterprises and developers now make up two-thirds of its revenue.
On July 28, its first anniversary, Fish Audio announced a $52 million seed led by Coreline Ventures and Capital Today, with 359 Capital, Play Time, HF0, and others joining. The team had grown from 3 to 22 people. Cao said it raised to build more advanced models and serve enterprises.
It also made its API free to developers through August. Fish says a rebuilt inference stack, running 8,000+ tokens per second on one H200, cut costs enough to give the model away.
On October 13, it launched Drama 3, an open-weight model, meaning anyone can download it. You can control how a single word sounds and shift a voice mid-sentence. It went live the same day on Runway, Pika, OpenRouter, Vercel, ComfyUI, and Nous Research's Hermes Agent. Fish now says it has 15M+ users, up from 8M+ at the seed.
prototype_27 retweeted
Bro caught a King cobra with bare hands and started extracting venom. One slip, and the video ends differently.
prototype_27 retweeted
🧠 OpenAI enterre doucement l'interface de chat classique.
Avec Intelligent UI, GPT-6 ne se contente plus de répondre avec du texte, il génère des interfaces interactives, des boutons et des graphiques à la volée.
En devenant visuel, ChatGPT se transforme en un véritable système d'exploitation dynamique.
On ne cherche plus seulement une info, on manipule un outil créé sur mesure.
Est-ce le début de la fin pour les applications traditionnelles à usage unique ?
prototype_27 retweeted
Grok Bot Dash
Track important metrics, get updates on tasks, and create new widgets in chat.
They should actually ship this.
prototype_27 retweeted
Elon explains why he is still working even tho he's worth $1 Trillion dollars. Great @elonmusk ❤️
#ElonMusk #elon_musk #elon #tesla #Starship #SpaceX #Starlink #AI #robots #trillionaireviews #musk #satellites #technology #technews #gineus





























