Manish Shah

This weekend's reading: DORA Accelerate State of DevOps report. Pretty lengthy but very interesting if you're working in this space, the sections on AI and the leadership qualities required to drive positive change are good reading in particular. https://lnkd.in/ecqPcf_H

28

1 Comment

I believe in encouraging a culture where taking calculated risks is encouraged because it’s vital for driving innovation and progress. The lessons we learn from failure are invaluable. In my leadership roles, I've experienced this firsthand with initiatives like AppBench, a developer self-service platform, and a Business Event Mesh, which provides real-time, context-aware business events available for next best action. By pushing boundaries, we can improve efficiency and enhance the overall experience for our internal and external customers. Adopting a blitzscaling mindset, where rapid growth and bold risk-taking allow us to seize opportunities, enables us to adapt quickly and stay ahead of the competition. Creating an environment where people feel safe to take risks without the fear of failure drives innovation, builds resilience, and ensures long-term growth, keeping the organization flexible and ready for whatever comes next.

1

Interesting how old threads can weave back into your life. I’ve just received ethics approval for my Master’s study, which means I’ll be launching the first questionnaire soon! 👩‍💻 What fascinates me is how this new research takes me back to my Honours project in 2003, when I built JavanOwl - a web-based tool that allowed people to write and run Java code right in their browser. It was innovative for its time, using the internet to make programming more accessible to those without the usual resources. It looks a bit dated now though 😂 Two decades later, AI coding assistants are lowering the barriers in a similar way - making it easier to learn new programming languages and reshaping which skills matter most in software development. My Master’s study explores the impact of these tools on how we build software and the skills we rely on. I didn’t expect to revisit this theme, but here I am - still fascinated by the ways technology reshapes how we learn and build software. For anyone curious about my original Honours project, here’s a link to my paper on JavanOwl published in IEEE Symposium on HCC Languages and Environments, 2003: https://lnkd.in/gbyyqxu6 / https://lnkd.in/gxQG5qyH. I’ll be sharing the questionnaire to my study on the impact of AI on software engineering on LinkedIn and other professional platforms shortly. If you’re a software engineer, I really hope you'll consider participating as your insights will be invaluable!

74

7 Comments

Read a round-up of the latest financial services updates in our new edition of UK Regulatory Radar: https://lnkd.in/etSaaVW3

4

I came across this study from McKinsey that states that AI adoption is surging to a record high, and Gen AI is performing well, but data barriers and risk need to be addressed (maybe not such a big surprise). What I like in this survey is the message that AI and Gen AI have begun to create real business value and don't only exist as garage projects in an IT organisation; I notice that Risk, Legal, and Compliance reports increased revenue and that HR reports the most significant cost decrease. I'm realistic, but this is a good way to start the value conversation. That is a strong signal that business adoption is moving in the right direction. Unsurprisingly, the industry technology is most likely to invest in AI. I think financial services are forced to adopt AI to be cost-effective and comply with regulations. Interestingly, energy and materials are more prominent than media. Given the insights from the McKinsey study, I'm keen to explore the AI landscape in the Nordics. Localised research could shed light on any potential lagging of AI adoption in our region.

8

A novel which clearly explains the Devops principles, LEAN IT and Agile way of delivery. This is a short summary of the novel, if you have chance, recommend to go through it! https://lnkd.in/gzp2xg2M

8

1 Comment

Again Every posting news but not telling what happened same story :) 𝐖𝐡𝐚𝐭 𝐇𝐚𝐩𝐩𝐞𝐧𝐞𝐝 𝐢𝐧 𝐂𝐫𝐨𝐰𝐝𝐬𝐭𝐫𝐢𝐤𝐞 The driver file is severely bugged, causing frequent system crashes during startup. This results in a boot loop and requires a manual workaround (booting into safe mode and deleting the problematic file). There was a faulty update to CrowdStrike's Falcon Sensor, a core component of their cybersecurity suite, which triggered a compatibility issue with the Windows kernel, leading to system instability and BSODs. CrowdStrike develops and distributes the Falcon Sensor, a cybersecurity application that integrates with Windows. 𝐈𝐧𝐭𝐞𝐠𝐫𝐚𝐭𝐢𝐨𝐧 1)Microsoft is responsible for the car's overall performance and safety. 2)CrowdStrike is responsible for the functionality and compatibility of the alarm system with the car. 3)The BSOD incident is akin to the alarm system malfunctioning and causing the car to stallr. 𝐀𝐜𝐜𝐞𝐩𝐭 𝐭𝐡𝐞 𝐅𝐚𝐜𝐭 𝐨𝐟 𝐂𝐫𝐨𝐰𝐝𝐬𝐭𝐫𝐢𝐤 1)Exposed a major single point of failure in essential and important services ! 2) More than half the world is down, aeroplanes grounded, banking systems The most serious issue is that we cannot push an update. Each machine requires manual intervention to boot into safe mode and delete the update file. 𝐋𝐞𝐬𝐬𝐨𝐧 𝐟𝐨𝐫 𝐂𝐈𝐒𝐎 1) Critical server must run on both Technologies (Linux and Windows) 2) Prepare PR for How to Handle Customer Trust :) 3) Technology Stack solid 4) Utilized SCA Tools Thank You Prabh Nair for this insight.

2

Earlier this week I did a talk at JAX London, one of my favourite, mainly Java-centric conferences. I've been speaking there for probably well over a decade, not just London but in Germany too. With a little luck I will be speaking in Munich in November (watch this space). My talk was about the internal workings of an LLM, tokenisation, embedding, attention heads and decoding. I ran a few demos at each stage to show how things work and to demonstrate what LLMs can and can't do well. At then end of the talk, my audience had a good understanding of exactly why an LLM can't naturally do maths, work out how many R's in "strawberry" or reverse sentences. Those were the negatives, they also clearly understood how LLMs excel at certain tasks and how to maximise those "skills". As with everything in mathematics I find LLMs fascinating, they are, after all, just probabilistic and nearly chaotic but driven by simple maths. This week we saw the amazing models of a fruit fly's brain. The neurons in a brain are significantly mode complex than a digital AI neuron. However, the number of neurons in even the smallest LLMs exceed the numbers in a fly's brain. We're a long way from understand how either of them work but it's fascinating to see that we're approaching the same quest from two very different angles.

51

12 Comments

To all of my technical and architect contacts, I have a generic microservices architecture question. Serverless, containers, hybrid, any other - based on your experience which option would you recommend and which would you stay away from in terms of dev pipelines and run cost? And what would be your comms preference - APIs, service mesh, message queues, other? It would be good to have some human input on this one and think greenfield 😄

10

1 Comment

Hey Gemini. What might George Orwell think about people using Large Language Models to generate content and communicate ideas? … George Orwell's perspective on large language models (LLMs) would likely be complex and ambivalent, drawing on themes from his most famous works, "1984" and "Animal Farm." Here's how it might break down: Potential for Manipulation: - Orwell was wary of governments and authorities controlling information. He might see LLMs used by powerful entities to generate propaganda or manipulate public opinion, similar to the way the Thought Police used language in "1984." - LLMs' ability to mimic human writing styles could create a world of "doublethink" (consciously believing contradictory ideas), blurring the lines between fact and fiction. Promise for Communication and Creativity: - Orwell was a prolific writer who valued clear, honest communication. He might appreciate LLMs' potential to democratize content creation, allowing more people to express themselves effectively. - LLMs' ability to generate different creative text formats could be seen as a tool for sparking new ideas and amplifying human creativity. Overall, it's difficult to say definitively. Orwell likely wouldn't be a blanket supporter or opponent of LLMs. He would likely be cautious, urging users to be critical of the information they generate and consume.

13

1 Comment

Should've used Klave: https://lnkd.in/esezB9GS

6

Research paper, 03.2022 "Imperialist appropriation in the world economy: Drain from the global South through unequal exchange, 1990–2015" Rich countries rely on a large net appropriation of resources from the global South. 👉🏾Drain from the South is worth over $10 trillion per year, in Northern prices. 👉🏾The South’s losses outstrip their aid receipts by a factor of 30. 👉🏾Unequal exchange is a major driver of underdevelopment and global inequality. 👉🏾The impact of excess resource consumption in the North is offshored to the South. https://lnkd.in/epdPhh57

3

1 Comment

I absolutely love this summary of good design and incremental refactoring. In my career I have not yet found a better approach to building large scale simple solutions that can handle change.

11

Wonderful episode tackling important themes of costs, sampling and linking to business priorities. This is my new favorite podcast for 2024. #observability #finops #sre

12

3 Comments

I'm creating a curated overview of resources around event-driven architecture. To celebrate great content others have created, and because sometimes people ask me for a reading list.  Today I added an article by Barry O Sullivan that seems to be over six years old, but I think it's still good: https://lnkd.in/eBiQZf52 So if you'd like to suggest an addition, but have doubts because it's old - go ahead. It's about quality, not recency. The list is of course https://lnkd.in/eqHcSsif #eda #eventdrivenarchitecture #softwarearchitecture

18

Interesting to see the introduction of agents into copilot studio today. making agents available in low code has great potential for all our manual processes.

12

2 Comments

Part 4 of Talking with your Data... adding tools to the framework to help the LLM with Question Answering... showing what kind of complex problem solving we can achieve with a single prompt using a state of the art LLM (Google's Gemini 1.5 Flash) without spending any £s. #ai #llm

2

In a follow-up to my last blog post, here are more things you can do on the road to your app modernization efforts.

9

You are Hercules, once again battling the hydra. As always, the hydra can be defeated by cutting off all its heads. But every time you cut one off, it grows more. Today, however, you face a far more terrifying beast—a mathematical hydra. Not just a monster with many heads, this one has a tree-like structure, with heads sprouting from other heads. Worse still, it can regrow entire groups of heads at once! Do you stand a chance? To clarify, the hydra is represented by a rooted tree. When you cut off a head (a leaf in the tree), the entire parent branch gets copied twice, with new branches attached to the grandparent. The only exception is a head directly attached to the root—it doesn’t regrow. Let’s start with a small, six-headed hydra (see the picture), and chop off head #1. Immediately, two new heads grow in its place. Now, cut off head #2. Oh no—eight more heads appear! At this rate, how could you ever win? You can keep experimenting with hydras of different shapes and sizes, but one thing is certain—fighting hydras is tough work. But here’s the twist: no matter how terrifying the hydra seems, you can always win. The catch? It might take a VERY, VERY long time. That’s not even the biggest surprise. The real surprise is in the strategy—or rather, the lack of one. Any strategy you use will eventually win. Just keep chopping off heads, and YOU CANNOT LOSE. What if the hydra grows three new heads each time? Still can’t lose. A hundred heads? Nope. In fact, the hydra can grow ANY number of heads, even decide how many at each turn—and you’ll still win. Surprised? The proof, while simple, is a bit technical. It involves so-called ordinal numbers, which allow us to enumerate infinite sets. Without getting too deep into details, we can assign ordinals to each head and show that no matter how many heads regrow on lower branches, the largest ordinal always decreases when you cut a head off. In other words, a single head on a higher branch is infinitely worse than any number of heads on lower branches. One last fun fact: although each hydra and the number of steps to defeat it are finite, the proof HAS TO involve infinite values. In mathematical terms, the "hydra theorem" cannot be proven within Peano arithmetic.

17