Salient eDiscovery’s Post

Salient eDiscovery’s case study is a testament to the transformative power of AI in legal review. With Reveal’s continuous active learning (CAL), their client reduced 9 million documents to just 55,000 relevant ones—completing the review in only 90 days! 🚀 It’s always exciting to see our technology enabling smarter, faster, and more confident decision-making. Check out their post to learn how Reveal is revolutionizing legal tech. #eDiscovery #LegalTech #AI

I've been completing work on the DAVID™ operating system preparing to retrain it. One of the core structures of the OS enables identifying and learning Semantic Definitions (dropped patent filing). From that, DAVID™ creates the native code which it then executes. It isn't abstract machine learning stored in data structures, it learns to resolve conflicts and adapt. The fun part is that I no longer need to be the "expert" - humans often can't see beyond the analog extension of tech choosing terrible data partitioning and crippling algorithms. America First for 2025 in less than 256KB of OS.

In a clear, step-by-step tutorial, Maria Rosario Mestre shows how you can use DSPy and Haystack to save time and resources by automating your prompt-engineering workflow.

As you may have heard, Hamel H. and Dan Becker are starting a new course on LLM fine-tuning with a rockstar cast of guest speakers (quite literally a list of my favorite people in ML), including me! I'll be talking about the latest and greatest with Accelerate, of course If you're interested in LLM Fine Tuning, evals, or just from hearing from an amazing set of people and their thoughts, you can sign up here: https://lnkd.in/eBgTjCkR

Predictions for the future of LLMOps are varied.   But what is clear from speaking to Clemens Rawert, Co-Founder at Langfuse;   Is that users have become a lot more thoughtful and systematic as LLMs have matured.   And building golden datasets are going to be even more important for quality benchmarks.   Just 2 points of importance at Langfuse (YC W23), Take a look at the video to find out what else.... 💡

What should we make of OpenAI's stance on patents? In Law360, MBHB Partners Mike Borella and Joshua Rich share their thoughts on the broad exceptions and lack of enforceability, calling it "public relations virtue-signaling to the tech community and regulators." Read more here. https://lnkd.in/evNpVKFU

The self-correction process you’ve highlighted is a fascinating shift in AI behavior. It’s not just about the final answer anymore, but the reasoning path taken to get there. This approach of showing internal "thoughts" adds transparency and a sense of caution, reducing the tendency to hallucinate. It’s a major leap towards building trust in AI responses, especially in cases where accuracy is crucial. What's really interesting is how this could change how users interact with AI—less about getting fast answers and more about understanding the reasoning behind them. #AITransparency #SelfCorrection #EthicalAI

DSPy streamlines prompt and parameter optimization for large language models by automating adjustments, enabling developers to concentrate on impactful system building. This hands-on guide demonstrates how DSPy’s LM-driven optimizers simplify the traditionally complex process of prompt engineering, making LLM pipelines more reliable and adaptable. Dive into how DSPy can elevate your development process by reducing the need for manual, iterative prompt tuning. Read the full article here: https://lnkd.in/g5xdQKpy #generartiveai #llm #dspy #promptengineering

Build AI applications that have long-term agentic memory! Our short course “LLMs as Operating Systems: Agent Memory” is based on insights from the MemGPT paper and taught by two of its coauthors. Learn how to implement persistent, efficient memory management for applications based on large language models. Enroll for free: https://hubs.la/Q02-BC7R0