Instagram story viewer> @sherpa_ai> Posts
sherpa.ai(@sherpa_ai)'s profile avatar HD Profile

sherpa.ai

@sherpa_ai

8.4K
followers
250
following
We research and build #ArtificialIntelligence technology and products.
POSTS STORIES REELS
Vita ya está disponible en librerías. Si alguna vez te has preguntado hacia dónde vamos, este libro tiene mucho que decirte. #Vita #SherpaAI by @sherpa_ai
En Sherpa llevamos años trabajando para que la IA esté al servicio de las personas. Vita es, en cierto modo, la versión más profunda de esa misma pregunta: ¿qué significa vivir en la era de la inteligencia artificial? #Vita #SherpaAI by @sherpa_ai
Nuestro CEO y fundador, Xabi Uribe-Etxebarria, acaba de publicar su primer libro: Vita. Una reflexión sobre cómo la inteligencia artificial y la ciencia están redefiniendo lo que entendemos por vida. Un tema que, como podréis imaginar, nos toca muy de cerca. 🌱 #Vita #SherpaAI by @sherpa_ai
Generative AI answers. Agentic AI acts. An AI agent receives a goal, breaks it into steps, uses tools and carries out tasks with a degree of autonomy: investigating an incident, cross-checking data across systems, preparing a decision. That's what makes it so useful. It's also what changes the questions companies need to ask. To act, an agent needs access to documents, databases and internal systems. What can it reach? What information leaves the organization? Who approves its actions? Is there a record of what it did? With agents, the question isn't just what AI can do. It's what it can access, and who has the final say. #AgenticAI #AIAgents #AIGovernance by @sherpa_ai
AI is a big journey, but jumping in is not hard." Andrew Ng His advice to companies: start with a small pilot and aim for quick wins. In regulated industries, that first step often stalls for a different reason: the data can't be moved. With Federated Learning, it doesn't have to. by @sherpa_ai
One subsidiary detects a phishing campaign in March. Six weeks later, another subsidiary in the same group is hit by the same campaign, unprepared. The intelligence existed inside the organization. It just never travelled. The obvious fix is to centralize all security logs, and it tends to stall: employee personal data under GDPR, regulators that don't allow certain data to leave the country, vendor contracts that forbid sharing it. Our new article walks through an illustrative scenario: each SOC trains the model locally and only shares what it learned. When one subsidiary spots an attack pattern, the rest of the group picks it up without a single log crossing a border. We also cover what Federated Learning doesn't solve: data quality still matters, and the decision still belongs to the analyst. by @sherpa_ai
When a condition affects very few people, no single hospital has enough cases to train a reliable diagnostic model. The data exists, but it's spread across institutions and countries that, rightly, don't allow patient records to leave their systems. That's the challenge NIH (US) and University College London (UK) are tackling with http://Sherpa.ai for Collagen VI-related dystrophies. Each institution trains the model on its own microscopy images and only shares what it learned. The result: a model trained on larger, more diverse data, with significantly improved diagnostic accuracy. It's the first global application of Federated Learning to this diagnosis, with no patient data transferred. by @sherpa_ai
Schneier wrote this ten years ago and it still lands.​ Data isn't only an asset. It's a liability that grows the longer you hold it and the more places you copy it to. Every centralised dataset is a target.​ Most of the industry answered by building higher walls around the same pile. The other route is to train the model where the data already lives, so the pile never gets built.​ by @sherpa_ai
Federated learning, without the jargon.​ The usual way to train a model is to bring all the data into one place. That stops working when the data is patient records, transactions or security telemetry.​ Federated learning brings the model to the data instead. Each node trains on what it holds and returns only what it learned.​ Not a privacy layer added at the end. An architecture decision taken at the start.​ by @sherpa_ai
Each subsidiary detects threats the others will face weeks later. And security telemetry can't be centralised: personal data, residency requirements, third-party contracts.​ So every SOC trains its detection on what it sees locally. Always less than it needs.​ Federated learning changes the direction. The model travels; the data doesn't. Each site trains on its own telemetry and only what the model learned is shared.​ The bottleneck was never a lack of data. It's that the data sits where it can't be brought together.​ by @sherpa_ai
The defects that cost the most are the ones you've seen the least. A defect that appears on every tenth unit gets caught by any inspection system. A defect that appears three times a year is the one that reaches the customer and triggers a recall and it's the one no AI model can learn, because no single plant has enough examples of it. Across a supply chain, those examples exist. But process data is IP, defect data is liability, and no supplier is sending either to anyone. Sherpa.ai's Federated Learning platform lets manufacturers and suppliers train a shared quality model without a single record leaving the organization that owns it. ◆ Learn from defects seen across the entire chain, not one line ◆ Process know-how and defect history stay private ◆ Every participant gets the improved model back Better product quality doesn't come from collecting more data in one place. It comes from learning from data you'll never be allowed to see. #SherpaAI #FederatedLearning #Manufacturing #Industry40 #QualityControl #PrivateAI by @sherpa_ai
At CES 2026, Samsung convened a panel on what actually builds trust in AI. The conclusion: it isn't earned through promises, but through consistent, understandable behavior. Allie Miller made it concrete: people want to know whether a model runs locally or in the cloud, whether their data is secure, and what is AI-powered and what isn't. She was talking about consumers and devices. But the demand is identical in a boardroom, with more at stake: a hospital, a bank or a manufacturer needs to know where the model trains, what leaves their perimeter and who can see what. That's the principle Federated Learning is built on. Control isn't about protecting data well once it leaves. It's about never having to send it. 💬 In your AI projects, do you know with certainty where each model runs and what data leaves your systems? Source: Samsung Newsroom, "In Tech We Trust?" panel, CES 2026." by @sherpa_ai

Download All

Save the photos and videos currently loaded on this page.

0Photos
0Videos
Ready to download 0 / 0

Keep this tab open until your download is ready.
back to up