Amr Awadallah

New Mindset Needed for Large Language Models - https://buff.ly/3XbN4vp #AI #ML #LLMs #genAI

If you're connected with me, you know I scream about how necessary integrations are. Check out this piece where I explain more for insideBIGDATA!

21

A good read on building with LLMs https://applied-llms.org/

11

New on the Review: Conventional wisdom teaches managers to avoid “anti-patterns,” general practices that are deemed unproductive, like micromanaging. But Will Larson, a veteran engineering leader and CTO at Carta, argues that not every anti-pattern is bad, and makes a case for the three that eng execs should explore: 🔍 1) Shying away from micromanaging. “New engineering managers are often advised to ‘step away from the code.’ But an extremely high-functioning exec understands the domain they are operating in at some level of detail,” he says. “As you get too far out of the details, you just become a bureaucrat. Too many well-meaning engineering managers end up as bureaucrats.” 📏 2) Pushing back on flawed metrics. “Often when people start measuring, there's always a concern that their measurement is flawed. And that's true for all functions,” he says. “But I’m a believer in measuring something imperfect, but useful, versus holding out for the perfect metric while measuring nothing in the interim.” ☔ 3) Serving as the umbrella for your team. “In the past, I used to think I was energizing my team by sparing them the details. But now it feels like lying to them,” he says. On a tactical level, Larson adjusted this habit by buffering less information. “That means even if it’s disappointing for folks, I’d rather them process news bit by bit, rather than deal with a huge ocean of mess all in one moment.” Link to the full article in the comments.

196

8 Comments

Our paper, "QGen: On the Ability to Generalize in Quantization Aware Training," (https://lnkd.in/gXXvS_9Q) is now available on Arxiv. This research delved into the generalization capabilities of quantized models, a relatively unexplored domain, examining them through the lens of sharpness in the loss landscape during optimization. I found the experiments and mathematical proofs particularly fulfilling. We're hopeful for acceptance at the conference to which it's been submitted. 🤞

47

Today's article is the second installment of a two-part series on streaming data landscape. The initial piece, titled "What Exactly is a Unified Real-time Platform," delineates the fundamental features of URPs and elucidates distinctions between event stream processing (ESP) platforms, streaming DBMSs, and stream-enabled analytical DBMSs. This article aims to assist organizations in identifying the most suitable URP for their requirements should they conclude that a URP is pertinent to their solution needs. Please send us feedback at urpfeedback@gmail.com (cc: Roy Schulte, Sanjeev Mohan) #realtimedata #streamprocessing #streamingdata #streamingdbms #streaminganalytics RTInsights

14

2 Comments

Thanks Joanne Chen and Jaya G. from Foundation Capital for the nod! Great to be in the cohort of top notch AI companies. Observability data is untapped goldmine of information, but when the time comes (in the middle of incident), it still requires "expert" skills to find the needle in the haystack. We at Deductive AI are democratizing the expertise to everyone in the organization.

33

4 Comments

KV-cache quantization is super useful when you have a large batch-size and/or longer context-windows. This requires on-the-fly quantization. Happy to see HQQ used in this context: https://lnkd.in/eyxyPnhP

70

Check out our latest post on Clappy.ai, "Leverage Layout-Aware RAG for More Accurate Retrieval of Knowledge from Documents." Discover how to harness the power of Retrieval-Augmented Generation (RAG) for documents in PDF, Doc, and Markdown formats. Dive into the specifics of implementation and explore our reference repository to get started with advanced RAG for your knowledge base. Don't miss out on this essential read! Continue reading here for all the details: The Power of Hierarchical Indexing with RAG. (https://lnkd.in/gbiVHfcH)

2

How did we end up with systems aligned with the strategic interests of Big Tech and misaligned with the needs of customers and society? By 2022, Amazon, Alphabet, Meta, Apple and Microsoft collectively spent $222B in R&D, with Amazon leading at over $73B, followed by Alphabet at nearly $40B. The next five biggest R&D spenders listed on NASDAQ spent only $20B, which includes Nvidia at $7.3B For perspective, the entire R&D budget for the U.S. Federal Government in 2022 was about $180B across all agencies. 45.7% of the U.S. R&D budget was for national defense. The next largest focus area was health at 24.2%. I'm not suggesting the USG R&D budget is too small -- I'm suggesting that Big Tech's R&D budget is too large, concentrated, and conflicted. This is how we wound up with LLMs rather than responsible and accurate AI.

17

21 Comments

Can’t agree more on this as we are seeing more early stage startups building more useful AI-native agents instead of copilots on existing workflow.

4

💡 Why do we need so many evaluation tools for LLMs? As engineers, we build 'production-ready' LLM products using these metrics. But what happens next? How do we maintain control and ensure reliability? At Feedback Intelligence, we’ve crafted a cookbook to keep your LLMs reliable and aligned with user expectations. 🍲 📖 Give it a read and let’s chat!

15

BIG product update here at Forwrd.ai! 🔥 Improved business logic editor 🔥 Enhanced usability for data modeling. [link in the comments] Our north star product metric is TIME TO VALUE ⏳ Achieving faster time to value often requires re-writing core product modules. These engineering tasks are complex and can be time-consuming. It’s natural to delay these crucial updates, pushing them to “later” until they become unmanageable. My recommendation: Incorporate these vital tasks into your backlog on a monthly basis. Trust me, the payoff is worth it. Regularly addressing core updates not only enhances performance but also ensures a seamless user experience. #Startup #ProductManagement 

37

4 Comments

Part II of our blog is out! In this installment Abde Tambawala and I explored usage metrics and pricing models across different AI modalities and discussed how traditional User-based subscriptions will need to be supplemented with usage metrics. Take a look and let us know what you think! #pricingstrategy #ai #monetization

27

2 Comments

OpenPipe Introduces a New Family of ‘Mixture of Agents’ MoA Models Optimized for Generating Synthetic Training Data: Outperform GPT-4 at 1/25th the Cost Quick read: https://lnkd.in/gfxdiDT5 OpenPipe’s MoA models have excelled in rigorous benchmarking tests, achieving notable scores on LMSYS’s Arena Hard Auto and AlpacaEval 2.0. The MoA model scored 84.8 on Arena Hard Auto and 68.4 on AlpacaEval 2.0, indicating its superior performance in generating high-quality synthetic data. These benchmarks are critical as they represent challenging user queries that test the robustness and adaptability of AI models. The MoA model has been benchmarked against various GPT-4 variants in real-world scenarios. Results showed that OpenPipe’s MoA model was preferred over GPT-4 in 59.5% of the tasks evaluated by Claude 3 Opus. This is a significant achievement, highlighting the model’s effectiveness and practical applicability in diverse tasks encountered by OpenPipe’s customers...... OpenPipe

42

2 Comments

While switching from academia to industry is often negatively associated with going to the dark side, my experience at Dotmatics is the opposite. One example is Dotmatics' support for the development of free and open tools for standardization of cytometry data annotation. This is critical for FAIR data, especially in the context of big data science. Eliminating data silos and enhance interoperability for good, not evil. This is from the top of the company down. Curious to know more from the view from the top to see why I am so passionate about this vision? Keep reading:

12

Anthropic Raises the Bar with Claude 3.5 Sonnet: A New Era for AI and Data? As someone who's passionate about data and AI services, I'm always on the lookout for advancements that push the boundaries of AI and its real-world applications. Today, Anthropic's launch of Claude 3.5 Sonnet has definitely caught my attention. Here's what's got me excited: Powerhouse Performance: Claude 3.5 Sonnet reportedly surpasses many benchmarks in reasoning, coding, and even visual comprehension. This could be a game-changer for complex data analysis and model development. State-of-the-Art Vision: Improved image analysis and text transcription capabilities have huge potential for automating data extraction from documents and images, streamlining processes for many industries. "Artifacts" Feature: Anthropic's new interface for displaying generated content (code, documents, etc.) seems incredibly user-friendly, promoting easier collaboration and review within teams. Here's where I see the potential impact for data services: Data Science and Analysis: With its ability to navigate unstructured data and leverage internal tools and APIs, Claude 3.5 Sonnet can generate high-quality visualizations, actionable predictions, and valuable insights. This has massive potential for accelerating research in fields like healthcare and education. Streamlined Processes: Improved automation in tasks like data extraction and code generation could improve efficiency and reduce costs. New Service Opportunities: We might explore offering services specifically around the unique capabilities of Claude 3.5 Sonnet, tailoring solutions to clients' needs. I'm eager to hear your thoughts on Claude 3.5 Sonnet, especially from those of you working in AI and data services. How do you see this fitting into the current AI model landscape? What are the potential implications for our industry? Let's discuss! #AI #DataScience #Anthropic #Claude3.5Sonnet #ArtificialIntelligence #Innovation #QuadrantAI

85

I remember learning category theory in undergrad and wondering: when would this ever be useful? Turns out it's very useful for program synthesis and software modeling -- thanks Kristopher for the great talk!

9