Sunny Hien Nguyen

LinkedIn team (Juan Pablo Bottaro Karthik Ramgopal) shared a helpful deep dive into their deployment of an LLM-powered job search assistant. Highly recommend this read! Few highlights:  💡 Relatively easy to setup RAG-based pipeline. Achieved 80% of the basic experience in 1 month, then 4 more months to reach 95% mark. 💡 Evaluating the quality of application outputs is hard. I hear this from many enterprises developing LLM applications. (check out Okareo by Matthew Wyman Boris Selitser) 💡 Balancing latency with quality is important. While certain techniques like Chain of Thought could improve accuracy and reduce hallucinations, it will increase response times, which can affect user experience. https://lnkd.in/gtGGwf5P #largelanguagemodels #RAG #EBR #LLM #ml #ai #hallucinations #evaluations #

4

Must learn agentic ai topic by Andrew Ng!

2

👉 Howdy LinkedIn! Tomorrow, on "Building Products with Generative AI" - I will be hosting 🚘 Nitin Kartik at 11:30AM CST/12:30 PM EST! We're going to be digging into the intersection of Product Marketing & Generative AI, as well as other topics that will branch from Nitin's experience in the space. I'm excited for another interesting episode with a great thought leader in the space - we hope to see you join for tomorrow's episode! #GenerativeAI #AI #ProductMarketing

15

2 Comments

I enjoyed learning about the Reflection technique from Andrew Ng, where you ask AI to do a task ("Summarize the Acme deal"), then ask it to reflect on its answer ("Check your answer carefully and highlight any potential hidden challenges"). This extra "Reflection" step can dramatically improve the AI answers -- even allowing older models to outperform newer models. I'd be very interested to see someone apply this technique to a  #BuiltWithEinstein example with an Einstein Copilot Action and an extra Reflection Action that improves the original answer. Full talk from Andrew Ng can be watched here ->

13

2 Comments

How do you pick the perfect AI model for our business needs ? "Every leader faces this challenge. Rapid value, mounting pressures. How do we navigate it?" "Choosing the right AI model isn't just technical—it's strategic. 1. Model flexibility is key. 2. Budget decisions drive scope. 3. Criteria for evaluation are critical. Leaders from AWS and Arthur AI share innovative strategies to guide your choice, ensuring alignment with business needs and technical specifications." "Mastering AI selection transforms outcomes." I did enjoy watching the webinar and so will you i guess , Here →

1

Are Generative AI Drawbacks Your Big Differentiation Opportunity? Struggling to find your differentiation in the generative AI space? What if the so-called 'flaws' in AI are actually your secret weapon for differentiation? Let's flip the script on three common AI challenges: 1. Prompt Engineering: Embrace the User’s Burden Ever feel like you're playing a guessing game with AI, needing to experiment with the inputs till you get a satisfying response? That's prompt engineering at work. While essential for quality outputs from your LLMs, it can be incredibly frustrating for users. Your move? Develop smart systems that intuitively decode user intent, even from the vaguest prompts. Think multi-prompt strategies and context-aware responses that learn from user history. Suddenly, you're not just another AI tool – you're the one that "gets" your users. 2. Hallucination: From Glitch to Unlocking Creativity AI hallucinations – those intriguing yet sometimes annoying fabrications – are usually viewed as liabilities. But what if they could be a hidden wellspring of creativity? For example, imagine you tapped into your model’s hallucination to develop an Ethical Dilemma Simulator for workplace training, or a Strategy Game with "what-if" market scenarios. The key? Clear labeling AI generated content, user controls  for adjusting the level of "creativity" or unexpectedness, and rock-solid ethical guidelines. You're not just sidestepping pitfalls; you're advancing interactive learning by embracing AI hallucinations. 3. Small Language Models: Less Can Be More In AI, bigger models isn't always better. Small Language Models (SLMs) with the right pruning and quantization, can provide superior efficiency, privacy, and specialized performance without sacrificing on accuracy. An AI that runs locally on devices offers ironclad data security and lightning-fast responses. For enterprises paranoid about data privacy (aren't we all?), this could be the clincher. Invest in optimizing these compact powerhouses to meet critical user considerations like latency and battery performance, and you might just corner the market with your privacy friendly, "snappy and secure" AI solutions. The bottom line? In the AI gold rush, your differentiator might just be hiding in plain sight – in the very challenges others are scrambling to overcome. I'd love to hear from you: How are you planning to differentiate your product or service with generative AI?

9

1 Comment

If you have been wondering about Agentic AI or AutoGen, join Andrew Ng Qingyun Wu and myself for this introductory course of "AI Agentic Design Patterns with AutoGen". You'll find plenty of examples and code in action to help you better understand the building blocks that you can use to construct your own AI agents and agentic applications. Check the words from Andrew:

156

2 Comments

We're on a mission to draw attention to the many real existing use cases for AI-enhanced services in digital transformation today, rather than spend the next few years transfixed by the theoretical magic of LLMs. My learned colleague Cerys Hearsey has just shared a full capability map and recipe for using basic AI services to create an intelligent prototyping system for both products and services - even internal services within the enterprise. This is the kind of comprehensive learning material we are sharing through Shift*Academy every week 💪: https://lnkd.in/dMWJBjQP This is a great answer to the question I posed last week: are LLMs the atoms, organs or creatures in the AI evolution story? https://lnkd.in/ddWxV949

24

1 Comment

Among the AI wars taking place right now (e.g. GPU rich v poor, Data quality), a war over margins is one of them. Daniel Gross compared NVIDIA's healthy 85%+ margins to a 'kind of interchange rate for tokens', and that all layers of the AI stack are incentivised to invest and architect their infrastructure to lower their interchange rate. One of the key incentives at work is to commoditise the cost of complements. This has been the impetus for PyTorch, TensorFlow, Android and now Llama for Meta. We can see this at work in high resolution with Snowflake launching Arctic and Databricks launching DBRX. As Apoorv A. wrote in his excellent piece last week, at this point in time value has disproportionately accrued to Semis and Infra, but if mobile and cloud are any indication then this will invert eventually and margins will expand for applications closest to the end consumers. They'll be helped by Wright's law, which is driving down the cost of every generation of frontier models within months of its release. The cutting edge will remain the province of a handful of companies, but the capabilities are diffusing at unprecedented rates. Application builders can unlock 10x better experiences at diluted margins by leveraging expensive SOTA models, before graduating to higher margins as inference costs inevitably collapse. We're still in the early innings, but the incentive to commoditise complements and the margin wars at large will continue to develop.

54

3 Comments

I think Apple's generative AI features will significantly accelerate the technology's mainstreaming and usefulness. "AI for the rest of us," as Apple's Craig Federighi says. I can't possibly summarize everything here that they announced at WWDC today, and you can find such summaries elsewhere. But I think the most important things to note are: • Generative AI will be everywhere on Apple devices. Literally, everywhere. But in smart ways, I think. They're focused on useful and fun use cases. • Personal context is a key differentiator for Apple. Its generative AI can use all of your personal information—in a privacy safe way (see below)—to generate more relevant outputs. This will make the technology much more useful. • Apple's approach to privacy is a major differentiator. They do whatever they can on device, and offload to cloud-based servers when needed, but in a privacy-safe way. • Siri will indeed get much more powerful. You'll be able to talk to it naturally, it can see your screen, and it can take actions within apps. This will unlock the potential of generative AI models to do things on your behalf. • Other AI tools, like ChatGPT, will be integrated into Apple's experience. Siri will be able to suggest using ChatGPT to answer a question, for example. Users will be able to access ChatGPT free, or sign in to get access to paid features if they're a subscriber. I think this will all go a long way to mainstreaming the technology, making it accessible to average users. By this time next year, my guess is that the majority of iPhone users will be using generative AI in some way, entrenching the technology into people's lives in a way that probably wouldn't have happened otherwise.

15

Emerging technologies don’t simply materialize fully formed; they require the establishment of supply chains, the development of manufacturing processes, and the evolution of the technology itself. Along this journey, the cost incurred includes learning from the ten unsuccessful paths before arriving at the eleventh one that proves effective…a perceptive read if you are working in the Spatial Computing space

19

𝗪𝗵𝘆 𝗠𝗼𝗻𝗲𝘆 𝗦𝗵𝗼𝘂𝗹𝗱𝗻'𝘁 𝗗𝗿𝗶𝘃𝗲 𝗬𝗼𝘂𝗿 𝗦𝘁𝗮𝗿𝘁𝘂𝗽 𝗔𝗺𝗯𝗶𝘁𝗶𝗼𝗻𝘀 "Money is the only singular reason not to start a company. The odds of success are low, and if wealth is your sole motivator, you'll likely regret the journey. Instead, start a company because you believe in your idea, are passionate about it, and want to create something extraordinary. Have a unique perspective and be ready to overcome any challenge to make it happen." Focus on your passion, not the payout. See The video Here : https://lnkd.in/gw83NYG7  #StartupAdvice #Entrepreneurship #PassionOverProfit

3

4 Comments

#WorkTrendIndex reports with a focus on AI usage by all of us in Bay Area :  You can find the full global report, along with national stats here: 2024 Annual Work 

3

I haven’t tried this model fine tune and due to issues with gguf quant fine tune conversion, still working on adjustments. That being said llama3 and mixtral are my go to models. I’m using a context tuned version to 32k. Just not practically to run a 70b parameter local (yet) #llm #llama3 #genai #models

4

1 Comment

Interesting paper from the Anthropic team (Claude). This is a good step for #AI agents to be able to one day explain *how* they are making recommendations. Explainability is a current pain point, especially as AI goes upstream towards decision-making for customers: applying for loans, career moves, dating? https://lnkd.in/gUejxKXG

1

Gavin Uberti (CEO Etched) is another talented Harvard drop out, who is betting on a silicon model of accelerated compute which only runs transformers. A great (very narrow) strategy which will have a 20x hardware performance advantage over NVidia Blackwell architecture. No details on their software stack strategy, unlikely to match NVIDIA Cuda. HOWEVER, if the industry move on from Transformers as the central tenet of AI models, Etched will be in a very bad place indeed. The threats and risks to NVIDIA investors starting to emerge. If we also consider that NVIDIA is trading at 21 times forward PS ratio, (Microsoft only 11 times) these are risks to the NASDAQ as a whole.

2

Last night I had the fortune of representing LaunchDarkly at Sourcegraph's CodeAI Summit NYC. An important theme I took from last night's event? Context. Models like specifics. Computers can't read our brains or know what we want to achieve. Technology is kinda useless without the right context. As we move forward in this increasingly AI-generated world (for better and for worse) let us not forget the importance of adding context to the equation. Whether that's providing context to a dataset through how it was collected, sharing context of your repo to better aid a code assist tool, or simply providing context behind the why of our decisions. It's an interesting (and important) place to keep in our minds. #AI #MachineLearning #DeveloperAdvocacy #DeveloperEvents

14

1 Comment

I am glad to share that it is now possible to build reusable custom prompts with GPT-4 in AI Builder. Think of these are your own custom generative AI capabilities, whether you are trying to categorize incoming support tickets, generate automated summary for a sales lead or draft response to customer issues. Prompts can be used across Power Platform, whether you are trying to build workflow automation, intelligent enterprise apps or add new skills to your copilots. Following are the new public preview capabilities: ✨ Model selection between GPT-3.5 and GPT-4 ✨ Temperature selection, to have control over the output ✨ RAG with enterprise data in Dataverse Please give it a spin and let us know your feedback. #AIBuilder #Powerplatform #PowerApps #PowerAutomate #CopilotStudio

147

17 Comments