🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

Online data mixing reduces training costs for foundation models, but faces challenges: ⚠️ Human-defined domains miss semantic nuances ⚠️ Limited eval accessibility ⚠️ Poor scalability Introducing 🎵R&B: first regroup data, then dynamically reweight domains during training!

Scale alone is not enough for AI data. Quality and complexity are equally critical. Excited to support all of these for LLM developers with Snorkel AI Data-as-a-Service, and to share our new leaderboard! — Our decade-plus of research and work in AI data has a simple point:

Foundation models are great at public knowledge, but fall short on domain-specific tasks. 📋 That’s why we’re working with the brightest in their fields to build high-quality training data that actually makes AI useful. Want to learn more? 👉 snorkel.ai/expert-communi…

Our decade of work on AI data development has always been about *accelerating* the subject matter expert - not replacing them! Where automation is possible- saturation has been reached. The key to real AI delta is expert knowledge - which all comes down to the amazing experts!!

Huge thanks to Nasdaq for featuring our Series D! 👏 We’re using this momentum to solve the toughest data challenges in enterprise AI—from evaluation to expert curation. Building AI? Let’s talk data. 📈 #SnorkelAI #AIInfrastructure #LLMs #DataAsAService

Today, Ramp reached a new valuation: $16 billion Let the robots chase your receipts and close your books, so you can use your brain and build things. That's the way AI was meant to be.

A frequent evaluation mistake I see: assuming you need orders of magnitude more data than you actually do. What different evaluation set sizes are good for:

Three days out! 👏 We're going live with AI leaders from Accenture US, QBE, @Comcast, BNY & more. 🔹 Expert data + agentic AI 🔹 Live demos 🔹 Real-world use cases 🔹 Fresh research RSVP: snorkel.ai/events/develop… #AgenticAI #SnorkelAI

Highlights from Henry Kiss Ehrenberg’s theCUBE appearance on the future of AI: 🟦 Data strategy is key 🟦 Expert data drives real advantage 🟦 Trust & compliance are critical 👉 Full convo here: youtube.com/watch?v=Qjt-d9…

How can we close the generation-verification gap when LLMs produce correct answers but fail to select them? 🧵 Introducing Weaver: a framework that combines multiple weak verifiers (reward models + LM judges) to achieve o3-mini-level accuracy with much cheaper non-reasoning

Very exciting work on using weak supervision for RL- closing the “generation-verification gap”!! Once again- principled approaches to labeling/data development are the keys!

LLMs often generate correct answers but struggle to select them. Weaver tackles this by combining many weak verifiers (reward models, LM judges) into a stronger signal using statistical tools from Weak Supervision—matching o3-mini-level accuracy with much cheaper models! 📊

Excited for this!!

Introducing Weaver, a test time scaling method for verification! Weaver shrinks the generation-verification gap through a low-overhead weak-to-strong optimization of a mixture of verifiers (e.g., LM judges and reward models). The Weavered mixture can be distilled into a tiny

We just dropped a benchmark dataset on Hugging Face to test AI agents on real-world insurance underwriting tasks—built with CPCU experts. Most models still struggle. Here’s how to evaluate them right: 🧠 Dataset: huggingface.co/datasets/snork…

Excited for this!

Tokenization kickstarts every Transformer pipeline—shaping how models digest data. Our latest work introduces semantic, grounded video tokenization, leveraging objectness cues to boost efficiency and performance of video understanding models.

Working with Amazon Web Services to push what’s possible in financial services: from AI agents for underwriting to data-driven copilots. When the data’s right, the system works. #SnorkelAI #AgenticAI #GenAI #AIinFinance