Gregor Hohpe

🚀 Partitioning an LLM between cloud and edge: Using edge systems to run elements of generative AI could be a game-changer. While it requires careful planning and expertise, this hybrid approach could represent the future of AI deployment. Both the TROM Small Language Model™ and AI Agents fit seamlessly into an edge LLM system, enhancing the hybrid AI approach. 🔧 Autonomous AI Agents for Operations: TROM.AI™ offers a revolutionary solution with autonomous AI agents. These intelligent agents operate independently, handling tasks and making decisions without constant human oversight, streamlining operations, boosting efficiency, and reducing reliance on traditional workforce models. 🌐 TROM Small Language Model™: Develop custom generative AI models tailored to your organization’s unique needs.

16

🚀 Want to run your very own trained LLM, privately and securely? 🤖🔒 Imagine harnessing the power of a Language Model tailored just for you without worrying about third-party servers! 💼✨ Try, LM Studio. Jest for few click . LLM running in your PC. With LM Studio, you can ... 🤖 • Run LLMs on your laptop, entirely offline 📚 • Chat with your local documents (new in 0.3) 👾 • Use models through the in-app Chat UI or an OpenAI compatible local server 📂 • Download any compatible model files from Hugging Face 🤗 repositories 🔭 • Discover new & noteworthy LLMs right inside the app's Discover page #LLM #private #chatgpt #AI

9

Here's a talk I did recently at the ServerlessDays ANZ conference: https://lnkd.in/e7YHZEfZ ☑ Key points: ☑ 1️⃣ It's very hard to use AI to solve real business problems. Most business problems need repeatable, predictable, high quality solutions, and AI is non-deterministic, so it's hard to beat into submission. 🏏 2️⃣ You need to choose the level of control you have over the LLMs. You can have less control but have a very simple system, or you can have more control with a humungous amount of effort. 3️⃣ Start with 3rd party inference (e.g. OpenAI, Groq, Fireworks AI, Amazon Web Services (AWS)), and only move onto your own infrastructure after you hit performance or capability limitations. I took way too long to make this call - compress it down to 1 month. 🐌

18

2 Comments

I have been leveraging GCP for TrussHealth's AI and data storage, along with addressing customer data related issues in previous roles. Google's transparency in addressing and resolving recent data loss incidents instills confidence in their commitment to customer support and functional operational processes. Although data loss is a significant concern for enterprise storage, it is commendable to see cloud providers like Google openly acknowledge such challenges and demonstrate dedication to continuous improvement. Learn more about how Google Cloud responded to accidental data deletions: [Link to the article](https://lnkd.in/gTiRkg6X)

19

There are citations, and there are CITATIONS. This seems to be the latter for LLM-Modulo (https://lnkd.in/gyAyKx4E).. 😅

64

7 Comments

AI could lead to job losses in higher-skilled occupations, we are going towards a new era which we aren't yet prepared for. More builders / doers will outperform the managers / leaders with AI in near future. That brings a lot of potential job losses in the top more than the bottom of the chain

3

Shameless rip from Austin Parker’s BlueSky. But this whole thread is just too good! If you are passionate about #observability you’ll love this! Find the rest here! https://lnkd.in/gsvS_Jpz “”” 1. Metrics are good, actually, but most of you use them very poorly. … 4. there are no silver bullets when it comes to telemetry. you're trading between speed of access, data resolution, and retention … 5. you are not going to build a better datadog on-prem, they have a twelve year and twenty billion dollar head start … 18. the actual goal of observability is to have the data you need to find the answer rather than to have the data that contains the answer … 32. if you have thousands of dashboards, you have no useful ones … 44. single panes of glass are for glass cleaner, not o11y “””

40

5 Comments

I recently read Ethan Mollick’s CO- INTELLIGENCE: Living and Working with AI. He brings a very practical perspective on how collective intelligence can be efficiently harnessed. Here are some ideas that really resonated with me- ✅AI cannot be easily used for mission critical tasks. It requires careful analysis and substantial testing before applying in such areas. ✅ AI always tries to please you. This is more so in the context of Gen AI where LLMs will often provide an answer even if there is no information available. That’s where prompt engineering comes into play. ✅ Novelty is often interpreted as originality. However linking disparate ideas from multiple domains and inserting some random creativity can lead to creating something new! ✅ When AI provides great results, humans let it take over instead of using it as a tool. This could potentially impact human learning and skill development. The idea of collective intelligence implies leveraging AI to help with decision making not let the machine make decisions for you. ✅ Most importantly, without a fundamental restructuring of how organizations work, it will be difficult to recognize the benefits of AI! Overall a fantastic book and highly recommended for anyone interested in understanding the potential this new technology brings to us and how best to leverage it!!!

33

1 Comment

I'm a month late to the party on this one but better late than never :-) Wanted to send a big shout out to Renata Hakiwai, Eli Pohio, and the HTK Group team for leading this mahi! 🙌 Massive thanks to Judith Dixon (nee Hanna), Craig Hind, Paul Keating, and the Amazon Web Services (AWS) team for your genuine support and engagement in this work. A special mention to our incredible crew - Ngapera Riley, Lee Timutimu, Wade Reweti, and Atawhai Tibble. It was an honour working alongside you all, whānau. 🙏 I encourage everyone to check out the blog below and dive into the Framework document itself before raising questions or comments about: 1. Data locality 2. Onshoring data 3. Māori data sovereignty  4. The role of [insert any global tech giant here] Just to clarify – this framework isn’t a silver bullet. These are practical steps you can consider when engaging with data and with AWS, or take your own approach. Kei a koe te tikanga. Mauriora! #AWS #HTK #NNMD https://lnkd.in/dpGTAvJR

20

2 Comments

The FireHydrant acquisition of Blameless signals a major shift in observability and incident management. My Partner Jaya and I just wrote about the looming "observability crisis"—how the explosion of data and tools has paradoxically made it harder to extract meaningful insights. This deal is a direct response to that crisis. By combining forces, FireHydrant and Blameless aim to build an end-to-end incident management platform to reimagine root cause analysis from the ground up. But to *truly* address the observability crisis, it will take more than consolidation. We need AI-driven systems that: → Provide natural language interfaces to query complex systems → Automatically generate and update incident response playbooks → Offer predictive maintenance by recognizing patterns across heterogeneous data sources → Synthesize post-mortem reports, distilling insights from terabytes of data Most critically, we need to ditch volume-based pricing models that incentivize data hoarding over problem-solving. Expect more moves as the market evolves to meet the demands of increasingly complex enterprise systems.

50

5 Comments

This is a big deal. Even more features you're used to with traditional databases, but with the performance and scalability of NoSQL. The hybrid NoSQL+Relational data modeling and schema capabilities, as well as the long-term vision for Fauna's database, are the number one reason I came to work at Fauna. My favorite part is you can add schema controls over time, as you need them. It's not an all-or-nothing addition.

6

1 Comment

AI's Impact: Then and Now I was listening to a firechat from my new CEO Jason Liu discussing AI. I thought it was the time to dig into the linked-In articles I prepared (but never published) 6 years ago following my course at MIT Sloan School of Management Computer Science Artificial Intelligence Lab (CSAIL) on the impact of AI on business strategy in 2018. At the time, I was aiming to share my learnings in 3 short articles: Part I/III: Will this AI wave be the tidal one? Part II/III: “AI - With great power comes great responsibility” PART III/III: AI in Oil & gas I have kept the article as is, so seeing the evolution since 2018 is also interesting. ------------------------------- Part I/III: Will this AI wave be the tidal one? #AItidalwave; #narrowAI; #generalAI; #AIisthere We hear about AI quite a lot, yet it is still quite opaque for many. First of all, AI is not such a new thing. Without going back to ancient mythology, modern AI really started in the 50s. Alan Turing designed its famous test in 1950! (https://lnkd.in/drWk44Wj). What is interesting to comprehend today is that while AI went through several “waves” of hype, this 3rd wave seems to be better set up to deliver on its promises thanks to the contextual convergence of data proliferation, cheap storage and amazing computing power resulting from the rise of the GPU. With money flowing into AI at an increasing pace and AI being a strategic priority for China, one could think that big changes are underway…but actually AI is already part of our northern hemisphere daily life. Natural Language Processing, aka NLP (not to be confused with Neuro Linguistic Programming) is there when we use Siri, Alexa and chatbots are all over the place. Machine learning is core to Netflix recommendations and Amazon might soon ship you directly what you were thinking to buy before you opened the app to order! Whatever will come first, flying taxi drones or autonomous driving vehicles, (Diego Tejada Evans: bets are still open:) transportation will most likely start soon to undertake massive changes. So, what is coming our way? And what does it mean for each of us? This is the best part but to churn these ideas we must first distinguish between “Narrow AI” (what exists today) and “General AI”. Narrow AI : Think of the AI programs that have been beating Chess, Jeopardy and Go human champions but would not be able to do anything else than what they have been programmed/trained for. Narrow AI will increasingly displace jobs. General AI: picture the type of AI usually portrayed in movies. It does not exist yet and there’s a plethora of opinions about when this will happen if it ever does! Good news being that, if it does not, we can calmly go to Mars without fearing a “HAL like” AI will consciously throw the crew overboard! Additional interesting links: Everything you need to know about AI: https://lnkd.in/dJ5VmT24

8

What challenges do implementers of AI face when hallucinations occur, and how might they be mitigated? 🤔 Join the conversation 👇 https://bit.ly/4bMZg9S #ai #llm #aiethics #safeai

3

2 Comments

AI Skills in Non-Tech Professions Lead to Massive Wage Increases ================= #087852cd ver:0.85 Workers in non-tech professions can benefit from significant wage increases by acquiring artificial intelligence (AI) skills, according to a study by PwC. In the US, workers with AI skills earn a wage premium of up to 25%, surpassing other countries. Job ads in the US for roles requiring AI skills offer wages 53% higher than similar roles without AI skills. The report also highlights the increasing demand for skills involving person-to-person contact. Overall, AI skills are highly valued across industries and can lead to substantial wage increases. #AI #skills #wageincrease https://lnkd.in/gwck6Q37

1

1 Comment

Istio Service Mesh: great to see Ambient moving up to Beta, delta xDS becoming the default, and api-gateway becoming a first class citizen in Istio. https://lnkd.in/gYPmnTuj

34

Dan Cleary and his team at PromptHub are doing really interesting work to enable people to test out and evaluate prompts using different models and learn how prompt engineering impacts results in the real world. Whether your using the models in their native form or creating a RAG architecture he’s covering the bases. He also writes about his work and research in very easy to read language (and best I can tell it really is his words not AI!). He’s clear about what is his own research and when he’s leveraging others work. He’s very transparent. What he writes about is actionable for those getting up to speed on Gen AI. It doesn’t really matter if your a beginner or expert, he’s got lessons for all of us. I encourage you to follow him and check out his platform.

14

1 Comment

Revolutionizing UX Testing: How AI's New Interface Interaction is a Game Changer for All Users The new "computer use" feature of Claude 3.5 Sonnet is poised to revolutionize UX testing, and I believe it will be a game changer across demographics and cultures. H ere’s how I envision it being applied effectively: Automated Usability Testing - With Claude, I can simulate user interactions, such as clicking buttons and navigating menus, to identify usability issues. For instance, I could automate the testing of an e-commerce site by simulating a user's journey from product selection to checkout, pinpointing areas where users might struggle. This ensures that the interface works seamlessly for everyone. Performance Evaluation - By replicating user actions under varying loads, Claude can conduct performance testing that I would find invaluable. Imagine simulating high traffic on a banking app to identify bottlenecks during peak usage times—this capability ensures smooth functionality when it matters most. A/B Testing - Claude can automate A/B tests by interacting with different versions of a UI. I could gather data on user preferences and engagement levels, providing insights into which design resonates better across diverse user groups. Accessibility Testing - This AI can also check for compliance with accessibility standards by navigating through applications and identifying elements that may not be accessible to all users. For example, it could evaluate whether all interactive elements are keyboard-navigable, ensuring inclusivity for everyone. Continuous Feedback Loop - As Claude interacts with the interface, it can provide real-time feedback on user experience issues. This allows me to make iterative improvements based on observed behaviors rather than assumptions, leading to a more refined product. I truly believe that these capabilities will enhance the UX testing process significantly, resulting in improved user satisfaction and engagement across various demographics and cultures. This is an exciting time for UX design! #Claude3.5 #AI #UX #ArtificialIntelligence #UX #TESTING #UsabilityTesting #AI+UX #UXDesign #UserExperience #UIDesign #Tech #DesignThinking #Anthropic #MachineLearning #ComputerUse #AIModels #Automation

30

2 Comments