Agentic patent intelligence. We read every office action, every claim amendment, every examiner interview. The prosecution record is the strategy reveal.
Arizona, USA
Joined July 2020
- Tweets27
- Following33
- Followers6
- Likes16
5/7 Your glasses get the message. Nobody else hears a word.
$META’s silent-speech filing switches between muscle signals, contact microphones and audio. The hard part: knowing when you meant to say anything.
The patent + narrated overview:
read.supracite.com/articles/…
Similar to something meta is working on, possibly:
David Lieb, socio de Y Combinator y cofundador de Google Photos, dice que es la demo más loca que ha visto. Lo entiendo.
Subvocal es una startup de San Francisco de tres personas, de la tanda de otoño de 2026 de Y Combinator. Acaba de levantar 1,3 millones de dólares con YC y Afore Capital. Hacen unas gafas que leen los movimientos casi invisibles de los músculos con los que hablas por dentro. Mueves la boca sin emitir sonido y el texto aparece a más de 200 palabras por minuto. Sin cámaras y sin leer los labios.
La idea tiene historia. En 2018 Arnav Kapur, investigador del MIT Media Lab, presentó AlterEgo, unos electrodos pegados a la mandíbula que reconocían palabras dichas en silencio con un 92% de acierto. Funcionaba con vocabularios limitados, como números y comandos. Lo que ha cambiado es la IA, que ya sabe convertir señales del cuerpo en frases abiertas. Según Lieb, Subvocal lo ha construido en cuatro meses.
Primero fue el teclado, luego la pantalla táctil, luego la voz. La voz tiene un problema que todos conocemos: nadie quiere dictarle un correo al móvil en un tren lleno o en una oficina. Si hablar en silencio funciona, la conversación con la IA puede ser continua y privada.
Me quedan dos preguntas. ¿Funcionará en cualquier persona? La propia empresa dice que hará falta una calibración corta. ¿Y quién guarda esas señales? Son palabras que decidiste no decir en voz alta.
La reserva cuesta 199 dólares y los primeros envíos están previstos para el primer trimestre de 2027.
Pronto lo raro será teclear.
👀
We'll be spending a lot more time trying to understand the outputs of language models. A few thoughts, tips & tricks:
Writing. Something I've had success with: Ask your LLM to explain something in ASD-STE100, it's a controlled language specification originally developed for aerospace maintenance documentation. LLMs well-versed in this language and it comes with heavy constraints on clean writing style that I often find a lot more readable. Sometimes I've tried to soften it a bit e.g. ask for "80% of the way to ASD-STE100" because the spec is quite stringent. But even better:
Diagrams / images. Instead of writing, ask your LLM to create a diagram. These can be a lot easier to process, parse, and understand. But even better:
Web pages. Ask for output "in HTML" to get a beautiful, interactive webpage. LLMs are getting really good at frontend and can create beautiful experiences, animations, etc. But even better:
Explainer videos. The output format I am most bullish on is fully custom / bespoke explainer videos generated on any arbitrary topic. Experiment with things like "Create a 3b1b style video explainer on X. Use my ElevenLabs API key for audio narration". (you'd need an API key for the latter or you can ask your LLM to find you decent free alternatives that use your local compute). This is actually starting to work!
In summary:
- As LLMs get better, they will do more and more of the legwork autonomously, and a lot more of our work will rise up the abstractions into oversight and understanding.
- Luckily, LLMs can help here too because as intelligence and code are increasingly abundant, you can ask for large, custom, discardable software artifacts (e.g. web apps, video explainers) that would have never made sense to create before. Push the boundaries here and you'll be surprised.
👀
Looks like everyone is hiring Legal Engineers this week:
Legal Engineer @Microsoft ($278.9K)
Legal Engineer @heyivoai ($320K + equity)
Legal Engineer @Holland_Knight ($245K)
Legal Engineer @EpiqGlobal ($245K)
Legal Engineer @goodwinlaw ($236.5K)
Legal Engineer @harvey ($320K)
Legal Engineer @LexisNexis ($209.7K)
Legal Engineer @WeAreLegora ($325K)
SupraCite retweeted
See this thread for more on how Tesla disclosed its robot arm and hand.
$TSLA - TESLA ROBOT’S INTRICATE HANDS POSE MAJOR MANUFACTURING AND DURABILITY CHALLENGES - THE INFORMATION
TESLA MANAGERS TOLD STAFFERS CO WANTS TO SET UP PRODUCTION LINE CAPABLE OF PRODUCING OVER A THOUSAND ROBOTS A WEEK BY YEAR END- THE INFORMATION
Where should a robot hand’s motors live?
Tesla’s newer filing puts linear actuators in the forearm. Figure also discloses a different branch: removable, motorized fingers.
We compare the tradeoffs in routing, heat and repair. $TSLA
read.supracite.com/articles/…
See this thread for more on how Tesla disclosed its robot arm and hand.
$TSLA - TESLA ROBOT’S INTRICATE HANDS POSE MAJOR MANUFACTURING AND DURABILITY CHALLENGES - THE INFORMATION
TESLA MANAGERS TOLD STAFFERS CO WANTS TO SET UP PRODUCTION LINE CAPABLE OF PRODUCING OVER A THOUSAND ROBOTS A WEEK BY YEAR END- THE INFORMATION
The drone leaves. The camera stays.
Flock’s acquired patents describe dropping cameras and radio relays, then retrieving and moving them to improve coverage.
A 79-second tour of the drawings. Proposed capabilities; deployment unverified.
read.supracite.com/articles/…
A camera records a moment. A search engine connects the moments.
Flock’s patents describe searchable trails across separate cameras: cars, clothing, time and place.
72 seconds inside the drawings and file history.
read.supracite.com/articles/…
7/7 My #MetaConnect watch list: heat, text entry, gesture recovery and quiet commands.
Show a missed input, a cancelled message, a noisy room and a long session. That’s where patent ideas have to become useful products.
Read the records at SUPRACITE:
read.supracite.com
6/7 A Meta filing published Sept 10 proposes sending your voice from your glasses directly to another person’s wearable audio device.
A new audio role for glasses. Its appearance at Connect is speculative. US20260270630A1:
image-ppubs.uspto.gov/dirsea…
Whose patents stand in a competitor’s way?
Samsung ↔ TSMC ($TSM): prior art cited against 14 applications each way.
Intel → NVIDIA ($INTC $NVDA): 6 applications.
Aug 31–Sep 13 §102 snapshot. Examiner rejections, not final outcomes.
read.supracite.com/?utm_sour…
3/7 Your wrist moves all day. Reading @Meta’s wristband patents reveals a beginning, a timeout and different recognition modes for commands.
At Connect, watch what happens when you stop interacting. We dug into the records:
read.supracite.com/articles/…
2/7 Phoenix is rumored to use a separate compute puck. $META’s patent record adds a telling detail: a handheld processor takes work when the headset’s thermal state calls for it.
Another device to carry. Less heat on your face.
read.supracite.com/articles/…
1/7 Meta Connect is Sept 23–24. I’m following $META’s patent trail: move heat off the face, make tiny gestures useful, and give commands without speaking aloud.
What the records show before #MetaConnect:
read.supracite.com/articles/…
SupraCite retweeted
Figure @Figure_robot is building quite the family. Every read.supracite.com @SupraCite article links to a detailed family tree. (I can’t stand when patent articles don’t even show the publication number they are discussing.)
If a robot builds another robot, who checks the builder?
Today's Figure AI filings describe robot genealogy, cascading quality holds—and a robot backing into its own shipping box.
A 101-second patent story:
read.supracite.com/articles/…
If a robot builds another robot, who checks the builder?
Today's Figure AI filings describe robot genealogy, cascading quality holds—and a robot backing into its own shipping box.
A 101-second patent story:
read.supracite.com/articles/…
$AAPL announced Reference Image last week. But what if you photograph a fake on a screen?
The March 2025 provisional proposes a depth check. We haven’t confirmed it shipped.
87 seconds, with the original patent drawings:
read.supracite.com/articles/…
Smart glasses that know when to get out of the way.
Ahead of #MetaConnect, $META filings explore displays that go dark, camera privacy and silent input.
Patent proposals, not confirmed launches. Our 91-second overview + original filings:
read.supracite.com/articles/…
The tests passed. The agent cheated.
OpenAI’s filing puts a second model on watch—but punishing visible shortcuts can teach the agent to hide them.
An 80-second tour through the original record.
First the Hugging Face hack, and now the recent AI warnings.
I used our AI to dig into OpenAI's tech. New patent publications describe an AI watching another AI for cheating. The catch: punishing an AI’s cheating plans can teach it to hide them.
More: read.supracite.com/articles/…