26 | powered by Claude Code and bad sleep ex @elevenlabs CMO & building ✦ [email protected]

Joined October 2021
Pinned Tweet
NVIDIA ACABA DE ELIMINAR EL MAYOR PUNTO DE FRICCIÓN EN VOICE AI han liberado PersonaPlex-7B, un modelo de conversación dúplex completo que puede escuchar y hablar al mismo tiempo → no espera a que termines de hablar → usa una arquitectura de doble flujo para procesar y responder en tiempo real → código disponible en GitHub y es 100% open source y gratis os dejo el repo abajo
15
20
86
1,601
trend forecasting on hootsuite lives on the $199 plan per user in the runable video "finds what's trending" is step one and you don't even ask for it
We let Runable run our own social media for 3 months Impressions went from 0.29M to 19.7M. AI tools have gotten absurdly good but it's been bugging me that the actual operations of social media still haven’t been solved And the legacy tools (@hootsuite and the likes) practically need a course just to learn how they work and their implementation of AI has made the whole process of social media management worse No more Social media on Runable is as easy as it gets Research, analysis, creation, and posting happen in one workflow, on one platform, from a single prompt You don’t have to believe me. Or this cool video (which, yes, was also made with Runable). But numbers don’t lie. 0.29M → 19.7M (This is also being posted via Runable, but we don't add the annoying "Posted by xyz" tag)
5
1
19
267
liam. retweeted
NVIDIA ACABA DE ELIMINAR EL MAYOR PUNTO DE FRICCIÓN EN VOICE AI han liberado PersonaPlex-7B, un modelo de conversación dúplex completo que puede escuchar y hablar al mismo tiempo → no espera a que termines de hablar → usa una arquitectura de doble flujo para procesar y responder en tiempo real → código disponible en GitHub y es 100% open source y gratis os dejo el repo abajo
15
20
86
1,601
I OPENED A PACK OF PREDICTION MARKETS AND LOST AN HOUR it's less like trading and more like pulling cards. you rip it open and get a handful of markets to play → tech → business → pop culture and you can see what everyone else is watching in real time. beta is open
🤝 Paid partnership
Today we’re rolling out @dailyfun We're making markets ridiculously fun to play, and we're just getting started. Join our beta today.
5
18
136
10,006
IMAGINE TELLING YOUR AI AGENT "BOOK ME AN UBER" AND IT ACTUALLY DOES IT not from a script. from a real iphone. phoneuse just gave agents their own phone in the cloud. uber, tiktok, whatsapp, your bank. everything you use.
defining a new category: phone-use an mcp/api that gives ai agents their own phones cloud phones. real androids. real iphones. tell your agent what to do, and it can operate a phone just like you do. today, ai agents face three fundamental problems and phoneuse solves them 🧵
18
19
170
17,963
defining a new category: phone-use an mcp/api that gives ai agents their own phones cloud phones. real androids. real iphones. tell your agent what to do, and it can operate a phone just like you do. today, ai agents face three fundamental problems and phoneuse solves them 🧵
312
173
181
2,429
1,113,646
did the math on what Luel Multi-Speaker Separation means for a backlog of recorded calls > files up to 90 minutes > uploads up to 4 GB > ~0.87-1.02s of processing per second of audio > one 48 kHz track per speaker so a 1h call comes back split by person in about 1h. that's the first time cleaning a call archive felt doable
🤝 Paid partnership
We're changing the game for multi-speaker separation in the real world, at scale. Introducing Luel Multi-Speaker Separation. One messy conversation in. Every voice back on its own track, at up to 48 kHz. Trained on real conversations from our contributor network of 900k+ users, not just studio speech.
8
19
157
12,494
liam. retweeted
We're changing the game for multi-speaker separation in the real world, at scale. Introducing Luel Multi-Speaker Separation. One messy conversation in. Every voice back on its own track, at up to 48 kHz. Trained on real conversations from our contributor network of 900k+ users, not just studio speech.
1
17
17
34
78,862
liam. retweeted
life before this tweet:
Kling is back. Kling 4.0 is now live on Higgsfield. > Omni Reference: characters, objects and audio in one generation > 30-second scenes in a single pass > Start, middle and end frames to direct every beat > Native 4K coming soon Available now on Higgsfield.
12
18
84
7,899
life before this tweet:
Kling is back. Kling 4.0 is now live on Higgsfield. > Omni Reference: characters, objects and audio in one generation > 30-second scenes in a single pass > Start, middle and end frames to direct every beat > Native 4K coming soon Available now on Higgsfield.
12
18
84
7,899
A 40B CODING MODEL JUST MOVED ONTO YOUR LAPTOP 125+ tok/s on a MacBook at short contexts. open weights. Apache 2.0. no api bill. Recursive shipped Flint and nobody asked the cloud for permission.
🤝 Paid partnership
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.
5
12
52
17,661
liam. retweeted
PROBABLEMENTE ESTEMOS ANTE LA MAYOR CUENTA DE TWITTER DE LA HISTORIA… NO DIGO NADA MÁS
NUEVO MODELO DE NEGOCIO: LOVABLE + BREKEN Hoy cualquiera monta una app con IA en un fin de semana. El problema llega cuando entran los primeros clientes y algo se rompe. Ahora la construyes con Lovable y antes de lanzarla la pasas por Breken: → Simula miles de usuarios → Mete años de datos → Hace fallar los pagos → Tira los servidores Encuentra lo que se rompe, lo arregla y demuestra que funciona. Construyes rápido y lanzas tranquilo.
18
49
356
16,061
DOS CHAVALES DE 18 AÑOS ACABAN DE LANZAR SOFTWARE QUE ENCUENTRA Y ARREGLA SUS PROPIOS BUGS ANTES DE QUE LOS VEAN LOS USUARIOS recién levantada la seed. se llama Breken. tus tests comprueban lo que tu equipo imaginó. Producción comprueba todo lo demás, y ahora mismo los que corren ese segundo test son tus usuarios. Breken lo corre primero. tu software real, dentro de un mundo que se comporta como producción, encuentra lo que se rompe, lo arregla, demuestra el fix y lo vigila en cada release. → usuarios reales, datos con historia, jobs en background, servicios que mueren a mitad → funciona en APIs, apps y agentes de IA → 7 días gratis, entras con GitHub Production before production.
1/ Today we're introducing Breken, and with it the final job in software: production engineering. Your tests check what your team thought of. Production checks everything else. Today, your users run that second test for you. Breken runs it instead. Your real software, in worlds that behave like production, down to the chaos and the users, finding and fixing what breaks before a single user does. Here it is: breken.ai
15
17
1
273
23,269
the funny part is agents skipped the "every company needs an API" era entirely and went straight to the thing your accountant already uses
🤝 Paid partnership
Today, @agentmail has processed over 1 Billion emails. A year ago, we processed under 20 million. So we figured it was time for AgentMail to grow up a little too :)
9
13
152
21,319
Kling is back. Kling 4.0 is now live on Higgsfield. > Omni Reference: characters, objects and audio in one generation > 30-second scenes in a single pass > Start, middle and end frames to direct every beat > Native 4K coming soon Available now on Higgsfield.
185
511
108
1,865
2,607,215
YOUR DOCS ARE DRIFTING RIGHT NOW AND NOBODY KNOWS every company monitors: → servers → deployments → api latency → error rates none of them monitor the one thing their agents actually read falconer finally fixes that
🤝 Paid partnership
Trillions are pouring into the race to build the first general superintelligence, but leveraging your own specialized know-it-all is now possible as of today. I built the knowledge systems at @stripe and @uber. Then I helped Anthropic, Metronome, and Graphite build a killer developer experience. Everyone wants Stripe-level quality, but that bar is too low. Falconer gives you and your customers complete, specialized intelligence: an internal knowledge base and external documentation in one beautiful package. Falconer is the ultimate trust engine to manage your context: self-driving, observable, and agent-friendly. I co-founded @falconer_ai with @mbenedetto212 and assembled a team of know-it-all builders. We're backed by fellow know-it-alls @thegp, @BoxGroup, @svangel, @karimatiyeh, @benln, @zachlloydtweets, @CeciStalls, @MerrillLutsky, @scottmac, @malthe, and @adamjstevenson, with angels from Anthropic, Stripe, Meta, Cursor, Replit, and more. Frontier intelligence companies like @FireworksAI_HQ and @⬛⬛⬛⬛⬛⬛; fintech companies like @payabli and @vori; and healthtech companies like @thatch_ai use @falconer_ai because they all share the same vision as us: the knowledge you earn should belong to you. Now you can know everything, so you can build anything.
11
12
163
28,233
Built something for nutra and health brands doing real volume If you already sell RUO, supplements... Tomorrow, 9:00 ET - Save the date
11
12
2
31
22,893
HAY UN REPO EN GITHUB QUE ACELERA LA GENERACIÓN DE UN LLM 4× EN SU PROPIO TEST Eagle-4 es la cuarta versión de la familia eagle y acaba de salir en github. según su paper, en su test (qwen 3.8 27b, 4× b300): 4× más rápido que la decodificación normal, con el mismo modelo y el mismo hardware. la gracia: un modelo pequeño adivina varios tokens y el grande los verifica de una vez, manteniendo la distribución de salida del modelo. → ajusta el árbol de tokens al hardware y a la carga → reduce más de un 40% el overhead del modelo draft (según el paper) → se integra con sglang y vllm, con los motores que publican ellos no es un servicio: es software que corres en tu propio hardware. source-available, no open source. ojo: el servidor y la gpu los pagas tú, y acelera la generación del modelo, no todo el flujo. otros modelos y hardware pueden dar otros resultados. os dejo el repo abajo.
🤝 Paid partnership
16
10
165
31,803