broadfield digital founder | architecting ai for brands
Joined June 2025
- Tweets4.3K
- Following281
- Followers50.7K
- Likes7.4K
Pinned Tweet
Absolutamente recomiendo usar Cache Control en Claude Code
Probablemente uno de los Mods que te ahorrará la mayor cantidad de tokens y ayudará a mantener tus sesiones activas por mucho más tiempo
Añade esta barra para la caché de 5 minutos de Claude y envía notificaciones para ayudar a mantener la caché activa
Comando de instalación:
npx claude-code-templates@latest --mod observability/prompt-cache-control
Stefan Friedman retweeted
este ha sido el progreso de la IA en los últimos 3 años 👀
Stefan Friedman retweeted
Introducing Bivia.
A multimodal AI decision model built to complete entire workflows, not just answer individual questions.
We've spent 3+ years building and deploying AI in production across semiconductors, energy, manufacturing, and other complex industries.
Today, we're bringing that experience to everyone.
underrated line in the sunra api launch:
"the model you call is exactly the model that runs"
anyone who's shipped on an aggregator knows the "why does the output look worse today" feeling
they also itemize every request and every charge in the console
pricing is each model at its maker's own api price, pay as you go
that combo is what i'd want before putting video gen in prod
Stefan Friedman retweeted
los que montaron una startup revendiendo modelos de IA con un margen encima viendo esto
Stefan Friedman retweeted
AN API THAT ANSWERS YES OR NO USING LIVE WEB SEARCH, AND SHOWS YOU THE SOURCES
I have a small tool I built with Claude: I give it a company name and it tells me who’s already promoting them on X.
A lot of what it needs to check is yes/no stuff like “did this company just raise?”
Right now the model answers from whatever it remembers, and sometimes it’s just wrong.
Parallel’s Decisions API checks the web first.
In their demo it rechecked 426 companies in under 10 seconds for 85 cents.
Plugging it into mine this week to see if the answers actually hold up
#ParallelPartner
🤝 Paid partnership
Introducing the Parallel Decisions API: the first web-grounded decisions engine.
Decision models like Jev can answer in an instant, but they lack the recency of current news. Now you can get both in a single call.
Decisions API is live today at $0.001 per request.
Try it with 5,000 free requests: platform.parallel.ai/play/de…
Here are some examples we've been experimenting with:
- Fact-check content (@Grok is this true?)
- Refresh stale banking records
- Classify companies and leads with fresh signals
- Backtest trading strategies
- Track competitor e-commerce inventory
i usually ignore anything voice for AI, most of it never sounds human
this changed my mind:
>install this for a tech support company
>use it with chatgpt or claude
>24/7 automated customer support
there’s no way you can’t make $10k/mo with this
🤝 Paid partnership
introducing prism, the world’s first speech-to-speech model you can’t tell from a human in conversation
prism has:
> 300ms response time, 4x faster than OpenAI GPT-Live
> outperforms every other s2s model on intelligence (with NO reasoning)
> the first s2s model customizable through reinforcement learning
we’re a team of 35 with $35M in funding, and we’re unlocking s2s model adoption faster than openai, google, and x.
the first 100 people to sign up here get 1 month free: rime.ai/prism?utm_campaign=p…
Stefan Friedman retweeted
the launch did 21M views and the best part was happening off camera the whole time
please tell me someone fixed that wire for vanessa
if this is what made the BTS cut i need to see what didn’t
how long did you guys spend filming this
Stefan Friedman retweeted
Earlier this month we introduced Griffin, the first AI model to pass the video Turing test.
The launch post got a lot of traction (~21M views), but filming it with Griffin running live on set was its own adventure...
Here are some of my favorite BTS clips:
Stefan Friedman retweeted
Every AI shopping agent assumes you already know what you want.
Most of the time you don't.
→ Agents: "tell me what to buy"
→ Shopping apps: a grid of thumbnails
→ scene: a feed of lifelike AI images where you can buy everything you see
Generative commerce > agentic commerce
🤝 Paid partnership
We raised a $7M pre-seed to make shopping fun again.
@daletlabs generates media just for you: like walking into a mall built around your taste. And you can buy everything in-app.
Built by researchers from Google, Runway and MIT. Backed by @lightspeedvp and amazing angels.
NUEVO MOD: El modo Claude Code ahora puede gestionar tus archivos .env sin ver nunca un solo valor
/env abre un panel para editar tus claves, y cuando Claude necesita una, pregunta con un recuadro informativo que te indica exactamente dónde encontrarla
Cualquier valor guardado de 8+ caracteres se elimina antes de que Claude lo vea
Claude incluso valida la longitud de la cadena para asegurarse de que hayas pegado el token correcto
/plugin marketplace add davekiss/env
/plugin install env@davekiss
Stefan Friedman retweeted
SOMEONE JUST FIXED VIBECODING'S BIGGEST PROBLEM
AI writes your code now.
But sometimes it leaves serious security holes that can take down your app, or your whole business.
Shannon finds them first. It's an open-source AI pentester with 48K+ GitHub stars.
Point it at your app and it:
→ hunts for vulnerabilities
→ fires real exploits to prove them
→ hands you a PDF report + SARIF
No exploit, no report.
Got a Grok subscription? You can run it with no extra API bill
Repo below👇
POV: your AI agent runs 24/7 on your laptop and costs you $0 in tokens
that's basically what LFM3 from Liquid AI unlocks
before: rent a giant model in the cloud, pay per call
after: a 2B model on-device, fine-tuned on your own data, faster than everything in its class
they even ran 300K agents in parallel on the Enron emails 💀
🤝 Paid partnership
Introducing 𝙻𝙵𝙼𝟹: our new generation of Liquid Foundation Models. With a new tri-hybrid architecture and new multimodal encoders, the LFM3 family is the ideal model suite for specialized intelligence.
> 6 model instances, across text, audio, vision, and omni modalities, generative and decision models.
> 𝙻𝙵𝙼𝟹-𝟷.𝟽𝙱: text, 256k context window.
> 𝙻𝙵𝙼𝟹-𝚅𝙻-𝟸𝙱: vision-language model with video-input support for spatial understanding
> 𝙻𝙵𝙼𝟹-𝙰𝚞𝚍𝚒𝚘-𝟸𝙱: tool calls, voice chat, reasoning in 18 languages.
> 𝙻𝙵𝙼𝟹-𝙾𝚖𝚗𝚒-𝟸𝙱: an omni model combines vision, audio, and language.
> 𝚍𝟷-𝟸𝙱 and 𝚍𝟷-𝙾𝚖𝚗𝚒-𝟸𝙱: decision models with typed outputs and calibrated probabilities in a single pass. Omni adds audio input.
> fine-tune on 1 GPU.
> a new tri-hybrid architecture
> architecture optimized for 14 processor classes across GPUs, CPUs, and NPUs.
> 𝙻𝙵𝙼𝟹-𝚅𝙻-𝟸𝙱 is a beast. It performs competitively with Gemma 4 12B (6x larger), and Gemini 3.5 Flash Lite, and GPT-5.6 Luna, with 1.9B parameters.
> 𝙻𝙵𝙼𝟹-𝟷.𝟽𝙱 is #1 on the Pareto frontier of device intelligence 𝚙𝚒𝚙𝚎𝚝𝚝𝚎.𝚕𝚒𝚚𝚞𝚒𝚍.𝚊𝚒
Every modality for specialized intelligence. Pick the one that fits your use case, customize it, and run it on-device.
1/ 🧵
Stefan Friedman retweeted
Introducing 𝙻𝙵𝙼𝟹: our new generation of Liquid Foundation Models. With a new tri-hybrid architecture and new multimodal encoders, the LFM3 family is the ideal model suite for specialized intelligence.
> 6 model instances, across text, audio, vision, and omni modalities, generative and decision models.
> 𝙻𝙵𝙼𝟹-𝟷.𝟽𝙱: text, 256k context window.
> 𝙻𝙵𝙼𝟹-𝚅𝙻-𝟸𝙱: vision-language model with video-input support for spatial understanding
> 𝙻𝙵𝙼𝟹-𝙰𝚞𝚍𝚒𝚘-𝟸𝙱: tool calls, voice chat, reasoning in 18 languages.
> 𝙻𝙵𝙼𝟹-𝙾𝚖𝚗𝚒-𝟸𝙱: an omni model combines vision, audio, and language.
> 𝚍𝟷-𝟸𝙱 and 𝚍𝟷-𝙾𝚖𝚗𝚒-𝟸𝙱: decision models with typed outputs and calibrated probabilities in a single pass. Omni adds audio input.
> fine-tune on 1 GPU.
> a new tri-hybrid architecture
> architecture optimized for 14 processor classes across GPUs, CPUs, and NPUs.
> 𝙻𝙵𝙼𝟹-𝚅𝙻-𝟸𝙱 is a beast. It performs competitively with Gemma 4 12B (6x larger), and Gemini 3.5 Flash Lite, and GPT-5.6 Luna, with 1.9B parameters.
> 𝙻𝙵𝙼𝟹-𝟷.𝟽𝙱 is #1 on the Pareto frontier of device intelligence 𝚙𝚒𝚙𝚎𝚝𝚝𝚎.𝚕𝚒𝚚𝚞𝚒𝚍.𝚊𝚒
Every modality for specialized intelligence. Pick the one that fits your use case, customize it, and run it on-device.
1/ 🧵
Stefan Friedman retweeted
Releases like this don't come around often.
Liquid AI just dropped LFM3: a family of ~2B open-weight models covering text, vision, audio, omni and decision tasks, all built on a single shared backbone.
LFM3-1.7B was trained to work inside agent harnesses like Hermes Agent and Pi, and it handles 256K context.
In Liquid's own evaluation on WebArena-Verified, it hits 38.4, compared to 28.9 for Qwen3.5-9B. And you can fine-tune these models on one GPU.
Big potential here for local agents that run 24/7.
🤝 Paid partnership
Introducing 𝙻𝙵𝙼𝟹: our new generation of Liquid Foundation Models. With a new tri-hybrid architecture and new multimodal encoders, the LFM3 family is the ideal model suite for specialized intelligence.
> 6 model instances, across text, audio, vision, and omni modalities, generative and decision models.
> 𝙻𝙵𝙼𝟹-𝟷.𝟽𝙱: text, 256k context window.
> 𝙻𝙵𝙼𝟹-𝚅𝙻-𝟸𝙱: vision-language model with video-input support for spatial understanding
> 𝙻𝙵𝙼𝟹-𝙰𝚞𝚍𝚒𝚘-𝟸𝙱: tool calls, voice chat, reasoning in 18 languages.
> 𝙻𝙵𝙼𝟹-𝙾𝚖𝚗𝚒-𝟸𝙱: an omni model combines vision, audio, and language.
> 𝚍𝟷-𝟸𝙱 and 𝚍𝟷-𝙾𝚖𝚗𝚒-𝟸𝙱: decision models with typed outputs and calibrated probabilities in a single pass. Omni adds audio input.
> fine-tune on 1 GPU.
> a new tri-hybrid architecture
> architecture optimized for 14 processor classes across GPUs, CPUs, and NPUs.
> 𝙻𝙵𝙼𝟹-𝚅𝙻-𝟸𝙱 is a beast. It performs competitively with Gemma 4 12B (6x larger), and Gemini 3.5 Flash Lite, and GPT-5.6 Luna, with 1.9B parameters.
> 𝙻𝙵𝙼𝟹-𝟷.𝟽𝙱 is #1 on the Pareto frontier of device intelligence 𝚙𝚒𝚙𝚎𝚝𝚝𝚎.𝚕𝚒𝚚𝚞𝚒𝚍.𝚊𝚒
Every modality for specialized intelligence. Pick the one that fits your use case, customize it, and run it on-device.
1/ 🧵
Stefan Friedman retweeted
Ideogram Vector 1.0 from @ideogram_ai is now on Sunra. Type a prompt or drop in a sketch, and get back an editable SVG.
→ Sharp at any size, from a wristband to a festival screen
→ Swap the colours
→ Edit any shape, point by point
→ Any style: line, engraving, airbrush, pop
Make your first vector at sunra.ai
AI just found a bug in a math proof that experts missed for decades
then it fixed it
the experts who checked the fix are in their 80s
🤝 Paid partnership
Stefan Friedman retweeted
The largest math proof in human history is now computer-verified.
Today, Axiom announces a complete Lean auto-formalization of the Classification of Finite Simple Groups, the "periodic table of symmetry."
About 15,000 pages, built by over 100 mathematicians across generations.
Stefan Friedman retweeted
holy shit. Axiom just formally verified the largest proof in math history
the Enormous Theorem (CFSG) by the numbers:
1. 15,000 pages
2. 500 papers
3. 100 authors
4. 30 years of work
5. 0 humans who have read all of it
AxiomProver checked every single step in Lean in 24 days.
~40M lines of machine-checked proof, 3x the size of Fermat's Last Theorem 🤯
🤝 Paid partnership