mostly here to learn and share 🌬️
Joined March 2023
- Tweets7.1K
- Following776
- Followers42.5K
- Likes9.8K
ethan retweeted
**Back by Popular Demand—Extended for 7 More Days!**
The **Flova × Seedance 2.5 Savings Event** has been extended through **Oct 27**.
Save up to **80%** on Seedance 2.5 and lock in the discounted rate for **365 days**:
🎬 **Seedance 2.5**
- 480p: From **$0.022/sec**
- 720p: From **$0.044/sec**
More limited-time offers:
- Image 2.5: From **$0.011/image**
- Wan 3.0 Prime: From **$0.022/sec**
One more week to create more for less—don’t miss the final extension! ✨
ethan retweeted
the part of this launch that matters most might not even be Flint itself.
Recursive’s Eureka Machine helped improve the architecture, training, and inference stack used to build it.
a system that helps research better models, which can then help accelerate the next round of research.
that’s a genuinely interesting feedback loop. @RichardSocher @recursive_si
Introducing Flint, @Recursive_SI’s first model: a fast local coding model built with our automated research system, the Eureka Machine.
• Best-in-class fast local coding
• Up to 2x faster than comparable models: 125+ tokens/sec on a MacBook Pro, 150+ on RTX Spark
• Built from scratch in under 4 months
Private, unmetered intelligence on your own machine. Open weights available now on @huggingface, run it locally on your favorite framework or try it for free on @opencode or @pidotdev.
ethan retweeted
running capable coding models locally without handing your code to a cloud API is the future i’ve been waiting for.
Flint is a 40B open-weight coding model designed to run on consumer hardware.
your machine. your code. your model.
big step for local AI. @recursive_si
Introducing Flint, @Recursive_SI’s first model: a fast local coding model built with our automated research system, the Eureka Machine.
• Best-in-class fast local coding
• Up to 2x faster than comparable models: 125+ tokens/sec on a MacBook Pro, 150+ on RTX Spark
• Built from scratch in under 4 months
Private, unmetered intelligence on your own machine. Open weights available now on @huggingface, run it locally on your favorite framework or try it for free on @opencode or @pidotdev.
ethan retweeted
eight cofounders with backgrounds spanning OpenAI, Meta, DeepMind, robotics, and reinforcement learning.
then they build a 40B coding model and the automated research system that helped create it.
Recursive is taking the idea of self-improving AI seriously, and Flint is its first concrete artifact. @RichardSocher
Introducing Flint, @Recursive_SI’s first model: a fast local coding model built with our automated research system, the Eureka Machine.
• Best-in-class fast local coding
• Up to 2x faster than comparable models: 125+ tokens/sec on a MacBook Pro, 150+ on RTX Spark
• Built from scratch in under 4 months
Private, unmetered intelligence on your own machine. Open weights available now on @huggingface, run it locally on your favorite framework or try it for free on @opencode or @pidotdev.
ethan retweeted
the long-term promise of local AI isn’t just saving money on API calls.
it’s having models that can work directly with your code and data, on hardware you control, without every interaction depending on a cloud service.
Flint feels like a meaningful step in that direction.
@recursive_si
Introducing Flint, @Recursive_SI’s first model: a fast local coding model built with our automated research system, the Eureka Machine.
• Best-in-class fast local coding
• Up to 2x faster than comparable models: 125+ tokens/sec on a MacBook Pro, 150+ on RTX Spark
• Built from scratch in under 4 months
Private, unmetered intelligence on your own machine. Open weights available now on @huggingface, run it locally on your favorite framework or try it for free on @opencode or @pidotdev.
ethan retweeted
open-weight AI is a lot more useful when you can actually run it yourself.
Flint is released under Apache 2.0, so developers can download it, run it locally, fine-tune it, and build on top of it.
less dependence on someone else’s API. more control over your own stack.
@recursive_si
Introducing Flint, @Recursive_SI’s first model: a fast local coding model built with our automated research system, the Eureka Machine.
• Best-in-class fast local coding
• Up to 2x faster than comparable models: 125+ tokens/sec on a MacBook Pro, 150+ on RTX Spark
• Built from scratch in under 4 months
Private, unmetered intelligence on your own machine. Open weights available now on @huggingface, run it locally on your favorite framework or try it for free on @opencode or @pidotdev.
ethan retweeted
PREDICTION MARKETS JUST GOT A NEW GAME LOOP
daily.fun is turning market discovery into something that feels more like opening a mystery pack than staring at charts.
→ crack open packs to discover markets
→ jump between tech, business and pop culture
→ see what people are paying attention to
this could get dangerously fun to try lol
Today we’re rolling out @dailyfun
We're making markets ridiculously fun to play, and we're just getting started.
Join our beta today.
ethan retweeted
Cómo me sentí después de encontrarme con esta cuenta
ACABAN DE SOLUCIONAR EL MAYOR PROBLEMA DEL VIBECODING
La IA ya escribe tu código.
Pero nadie comprueba de verdad que funcione con usuarios reales, datos de meses y servicios que se caen.
¿La idea?
En vez de lanzar a producción y rezar para que tus usuarios no encuentren el bug, Breken monta una copia realista de producción alrededor de tu app y la rompe a propósito ANTES de que llegue a nadie.
La IA primero:
→ crea usuarios, datos con historial y procesos de fondo
→ reinicia servidores y tumba servicios a mitad de tarea
→ prueba los órdenes raros en los que pasan las cosas
→ y cuando algo se rompe, lo arregla y demuestra el arreglo
Eso significa menos cobros duplicados, menos datos filtrados y menos noches de guardia apagando fuegos que ningún test vio venir.
Y no se queda ahí: vuelve a comprobar cada bug en cada release para que no reaparezca.
Funciona con APIs, apps y agentes de IA.
Tienen prueba gratis 7 días entrando con GitHub, sin tarjeta👇
ethan retweeted
este ha sido el progreso de la IA en los últimos 3 años 👀
ethan retweeted
🚨 AI AGENTS COULD ALREADY USE COMPUTERS. NOW THEY HAVE THEIR OWN PHONE
It’s called PhoneUse, and it launches today.
You give your agent a cloud phone, a real Android or iPhone, tell it what to do, and it uses the phone just like you would.
First came computer-use. Then browser-use. Now there’s phone-use.
What it can do:
→ Order an Uber and text you updates along the way
→ Research trends on TikTok
→ Get around old apps with no API, like insurance apps
→ Complete two-factor verification
→ Read your notifications and act on its own: if your flight is delayed, it messages you
And here’s the big part: most mobile apps have no API and no MCP. Until now, agents couldn’t get into them. With a phone, they can.
It connects through MCP or an API, so you can use it from the agent you already have.
The world still runs on mobile. Now your agent does too.
Save this before you forget 🔖
Link in the comments 👇
ethan retweeted
me: i'll learn video editing this year
Kling 4.0 on Higgsfield:
30-second scenes in one pass
me: so anyway, i'm a director now
ethan retweeted
In a year, "I shot this" and "I generated this" will look exactly the same.
Kling 4.0 is a big step there
ethan retweeted
wait... so AI agents don't need an API anymore?
@phoneuse just put an agent on a real phone and let it scroll TikTok by itself.
five phones, running at the same time, logging every trend that keeps showing up.
this is what agents look like when apps stop being a wall
ethan retweeted
my wallet seeing Kling 4.0 live on Higgsfield:
"here we go again"
ethan retweeted
FIVE PHONES SCROLLING TIKTOK AT ONCE. NOBODY TOUCHING A SCREEN.
that's the Phoneuse demo, and the scale is what stands out.
→ each phone opens TikTok on its own
→ checks what's working right now
→ logs the trends that keep repeating
→ all five running in parallel
what used to be one person's daily research is now a job that runs by itself.
and it's not just TikTok:
→ no API for Instagram
→ no API for Uber
→ no API for WhatsApp
Phoneuse gives the agent a real phone, so if it runs on a phone, an agent can use it.
for agencies, that means the hours spent hunting trends go into execution and landing more clients with the same team
ethan retweeted
my take: most people are going to look at kling 4.0 and only see "30 second videos"
that's not what got my attention
what got me is that i can keep my character, my product and the audio together in one go
that's the stuff that used to take me hours to fake
if the 4k lands as promised, this is going to be hard to beat
ethan retweeted
real talk, i don't get excited about ai video launches anymore
but being able to set the start, middle and end of a scene is a bigger deal than people think
it's the difference between hoping the ai gets it right and actually telling it what you want
kling 4.0 on higgsfield is the first one that gets that
i've tried almost every ai video model this year and they all had the same problem
the first few seconds look great, then the face starts drifting and you spend more time fixing it than making anything
kling 4.0 is the first time it looks like someone actually focused on that instead of just making the clips longer
that's the update i was waiting for
ethan retweeted
imaginate estudiar audiovisuales 4 años de tu vida y que de la nada saquen esto 😭😭
ethan retweeted
the guy who sold me a $497 course on stitching ai clips together watching this: