Erik Stenman

Both Ilya Sutskever and Jan Leike left OpenAI yesterday, with Ilya announcing that he will soon be revealing a new project. OpenAI estimated that Superalignment will be solved within four years. This means either they have successfully solved it, or else have realised that they are on the wrong path, and OpenAI won't be able to solve since focus has now shifted towards product and distribution. As for what superalignment or alignment means in general? In layman terms, whenever you enter some prompt in ChatGPT, you expect a particular answer in return. But often ChatGPT misunderstands and gives you the wrong answer. In some cases it might give you an approach that you didn't ask or is different from what you expected. Ex: Asking ChatGPT to calculate the area of a triangle from the given dimensions using Heron's formula. ChatGPT finds that the triangle is equilateral, and instead uses √3/4*a². Therefore the goal of superalignment is to ensure that Artificial Intelligence models stay within the bounds of human value, and their goal aligns with that of the requirements of a human, thus avoiding any unnecessary harm. Basically risk mitigation.

9

What software can you develop with a 50k budget vs 10k budget? At QikBuild, we've seen it firsthand. Innovative MVPs using modern development approaches cost about 10-15k EUR. Traditional development starts around 50k EUR. This cost difference isn't just about savings; it's about what you can achieve with your budget. With no-code, we've launched MVPs in weeks instead of months. This speed allows for quick market testing and iterative improvements based on real user feedback. Our digitalization strategy comes with unique characteristics: 1. It leverages pre-built components for rapid assembly 2. It scales efficiently as your user base grows 3. It relies on robust platforms for ongoing support For many SMEs, these features align perfectly with their needs, allowing them to move fast and adapt quickly in the early stages of their product journey. Curious about how our development approach can fit into your digitalization strategy?  Let's discuss your specific needs and find the most effective approach for your goals. #DigitalInnovation #MVP #SoftwareDevelopment #digitalization #innovation

8

Next week (Thursday 26/9) I'll once again be haunting the Göteborg React Meetup with silly ideas and dark thoughts—more specifically, I'll be there presenting "Entropy, Rot, and Decay in IT and Software". You'll learn how entropy is a quite tangible, handy concept to help understand why organizations deteriorate and why technology orgs specifically rot, and eventually, decay into obsolescence. Like in the "whisper game", meaning breaks down when there are too many connections—big organizations struggle to be relevant, fast, and innovative: But they could be! If you ever thought the universe is against you, I will explain that "yes, it is". But that's OK, as you will also learn.

27

4 kommentarer

Am I getting old? 👴 TL;DR: Apply for jobs with the best possible AI assistance 🤖 Some context: We recently started a recruiting campaign for a Senior Developer based in one of the Scandinavian countries and a Junior Developer. (We are still interviewing, so be quick! Check the comment below for a link to our career page 👇.) I've been recruiting for over a decade now and have seen a few application texts, all of which, in some way, tell me something about the candidate. It could be the tone of voice, structure (or lack thereof), or emphasis on different topics and parts of the text. You get the picture. Something is learned about the candidate. I've observed a change in the Application texts in our current campaign. Quite a few of them, in many ways, look and "feel" alike, and given that I have used GPT services since their paradigm-changing launch in November 2022, I recognise the content as "highly likely to be AI-generated." Now for the "Am I getting old?" part. When reading and evaluating (possibly) AI-generated Application texts, I start wondering about the candidates' motivation. Why take the shortcut of autogenerating the Application instead of writing it in your Language and Words? Also, how come so many candidates match the desired competencies in the Role description to the point? I can understand that there are legitimate reasons for getting GPT assistance, such as reading and writing challenges, using it to emphasise or improve already-written text, and proofreading. Also, I applaud the constructive use of AI and generative services in augmenting skills and encouraging us to perform and solve tasks and responsibilities more efficiently. Am I getting old and need to realign my expectations and accept that I need to be even more diligent when reading application texts to find the nuances of the candidates' edits in AI texts? I hope not, but I will embrace the change if it is here to stay 🙂

12

16 kommentarer

I discovered this library while trying to create a bridge between Pinescript and MQL5 in EricTraderSDK. Simply put, suppose you have a piece of Pinescript that generates profit on TradingView and you want to convert it into an Expert Advisor (EA) in MT5 to place orders automatically. This process will involve three challenging issues: - The system needs to understand the Pinescript language (the scripting language of TradingView) and parse it into an Abstract Syntax Tree (AST) data structure. - Implement a technical analysis (TA) library similar to how TradingView does in MQL5. - A module to convert AST data into MQL5. You can find a library that solves the first issue by the author Yunseong Hwang here: https://lnkd.in/gNANtxVJ

3

Klarna announced that they're filing for a U.S. IPO - https://lnkd.in/dd_dUpBN As Fintech professionals, why should we be paying attention (especially those of us in startups and growth-stage companies): 1. Market Validation, Investor Confidence, and Valuation - Klarna’s move to go public underscores the growing maturity and acceptance of fintech solutions. A successful IPO could boost investor confidence, potentially leading to increased funding and opportunities for other fintech startups. Lastly, Klarna’s IPO valuation will be closely watched as an indicator of market sentiment toward fintech companies. A strong performance could lead to higher valuations for other fintech firms, while a lackluster debut might prompt more cautious investment approaches. 2. Regulatory Scrutiny and Compliance - As a publicly traded entity, Klarna will face heightened regulatory oversight. This may prompt stricter compliance standards across the fintech industry, encouraging companies to invest more in regulatory frameworks and transparent operations. 3. Innovation - With the introduction of shareholders and growth expectations, Klarna might diversify its offerings beyond traditional BNPL services. This could set a precedent, inspiring other fintech firms to innovate and expand their service portfolios to attract and retain customers. #FinTechBrainTrust #Fintech #Payments #Startups

11

Today, Plural is doubling down on our team’s early investments in Helsing, a new type of defence AI company that believes that software, in particular AI, will be the key to protecting our democracies. We are fully behind their core mission to attain technological leadership so that democratic societies are free to make sovereign decisions and control their ethical standards. Having experienced first hand how ill equipped most democracies (old & new alike) were to ascertain their sovereignty I decided to back the Helsing team at the seed round following my partner Ian Hogarth who was involved from inception. We, alas correctly, expected for conflicts to escalate & that Europe’s only path towards protecting its sovereignty lied in technological leadership with ethics at its heart. We brought on board a number of European founders and operators equally committed to protecting European sovereignty and supporting the team as they navigated the escalation of conflict on the continent. Protection of European sovereignty, democratic values and open societies is not something we take lightly at Plural. A number of our team members have grown up on the repressive side of the Iron Curtain during the Cold War. Today 100M Europeans live on the Eastern flank of NATO, right next to the full scale war Russia is waging against its sovereign neighbour Ukraine on European soil. There is a critical and urgent need for investment in and development of world-class defence from Europe in teams with European values at the core. Having already backed some of the most ambitious founders addressing the world’s most critical challenges, we are proud to back Helsing’s world-class team, led by co-founders Torsten Reil, Dr. Gundbert Scherf & Niklas Köhler, that brings the highest ethical standards to defending democracy and freedom from Europe. They founded Helsing to help protect our democratic values and open societies, and they are taking a software-first approach to partnering with industry and governments to connect existing and new hardware platforms with advanced AI to keep our societies from harm. Read Bloomberg's coverage here: https://lnkd.in/gdTB-vap

108

5 kommentarer

Does AI (and AI dev) threaten the open web? Mathias Biilmann Christensen, CEO & Co-Founder of Netlify, makes a strong case why that threat is real, and why it’s important to defend it. It was a pleasure hosting Matt on The AI Native Dev podcast, going deep on many topics, such as: - AI agents have become a new type of Netlify user - how do you adapt to that? - How AI use of a non-AI dev tool (Codemod) boosted their productivity the most - The connecting lines between AI music generation and AI dev New tech paradigms are easier to grasp in closed, opinionated environments, controlled by one entity, than in an open and composable ecosystem. AI is clearly following that path, but where is it leading? Are we ok with all AI dev capabilities coming from our one vendor of choice, or insist on being able to assemble it from best of breed? These are just a few bits of a great episode. Matt’s authenticity and clear passion for AI and the open web meld perfectly with his deep insights, so much that the episode is a bit longer than planned - I just couldn’t stop the conversation! #Netlify #AIWeb #AINativeWeb #AINativeDev #MattBiilmann

61

3 kommentarer

Society in general is starting to take bad software seriously. Is your company? Hint: Let's play "bad software practices bingo" and find out! [  ] Like a microwave oven, you let the clock decide if the work's done [  ] The general sentiment is that testing is a vanity metric... [  ] ...So someone other than the developer writes tests... [  ] ...Maybe [  ] You do PI planning and other fika activities at a rate and cost that would bankrupt a small nation [  ] You have several layers of IT roles with inspecific responsibilities [  ] You have a lot of people talking about work, not doing so much of it [  ] Your management does not understand IT, so you have to do "special" (yes, read it in THAT voice) slides for them [  ] You do "quick and dirty" more than once in a lifetime; actually, you do it on a very regular basis [  ] Once per year all teams get to cut down infrastructure spend with 5% because your procurement thinks they can wrench money from AWS' cold, inanimate hands [  ] You estimate stuff to someone who estimates stuff who sometimes nods in blank-eyed agreement [  ] You have no way of knowing if things break before customers/users complain [  ] No one can -actually- say what the hell their software does if [X] leaves [ ] When someone is fired, you get to do/listen to "KT" sessions because -nothing- is documented according to any meaningful practices [  ] No developer has ever spent more than a few hours on quality-enhancing activities like security, observability, or solving known technical debt (go ahead, ask them) [ ] "Design" and "architecture" are things that the team read about in university, and at work these are represented by men in Birkenstocks with oddly specific opinions which can all be safely discarded (like in Rock-Paper-Scissors, PM always wins against Architect) There are so many varieties! What score did you get?

27

2 kommentarer

The AI landscape is changing so fast that it's important not to write code that ties you to a specific AI provider. Here is a real-world example: We're running an exciting AI project at Gazella AB and yesterday we ran into some issues where Anthropic's 3.5 Sonnet didn't perform well enough. With a single line of code changed, we changed to OpenAI's GPT 4o and it was able to solve the task perfectly. All this because we used Vercel's AI SDK which is AMAZING! The developer experience is awesome and the portability between AI models is as easy as changing a single line of code, which means we'll never build ourselves into a corner with any AI model. Highly recommended. Check it out: https://lnkd.in/dzR-ZvfE

35

5 kommentarer

Safe Superintelligence (SSI), co-founded by Ilya Sutskever, former Chief Scientist at OpenAI, is making waves in the AI landscape. With a staggering $1 billion in investment, this funding round propels its valuation to $5 billion (Link shared in comments) But this isn't just about financial milestones, it's about a critical mission: ensuring the safe development of artificial intelligence. As we witness AI's transformative power, the need to prioritize safety is more urgent than ever. At AssetCues we have integrated AI across multiple layers of physical asset tracking and management, and we recognize the far-reaching implications of these advancements, not just for AI, but for the broader SaaS landscape. We are witnessing the beginning of what Jason M. Lemkin of SaaStr aptly calls, the “Hyperfunctional SaaS” era. (Link shared in comments) Traditional SaaS models, once focused on providing generalized, one-size-fits-all solutions, are evolving into industry-specific platforms packed with deep, integrated functionalities. Here's what’s driving this shift: Automation at Scale: Many companies are automating up to 50% of their workforce, particularly in sectors like customer service. AI is the engine behind this shift, enhancing productivity and creating operational efficiencies that were once unimaginable. Instant Data Structuring: The expectation now is real-time data processing. AI enables instant structuring and searching of unstructured data, allowing businesses to tap into insights that were previously out of reach, transforming decision-making processes. Natural Language Interactions: The rise of intuitive, conversational interfaces is changing the way users engage with software. Customers expect products to understand plain language commands, anything less risks irrelevance in today’s fast-paced market. Automated Onboarding: The complexity of onboarding processes is no longer a tolerable friction point for customers. They expect immediate, seamless automation, eliminating manual bottlenecks and improving time-to-value. Predictive Analytics for Proactive Decision-Making: Predictive analytics is playing a critical role in helping businesses anticipate needs and trends before they arise. AI models are being used to forecast demand, identify potential issues, and recommend optimal actions, allowing companies to stay ahead of the curve. Unified Solutions: Customers no longer want a suite of disconnected solutions. They're looking for comprehensive systems that integrate multiple functionalities—think payroll systems that also handle onboarding, benefits, and more. As SSI takes this significant leap forward, it serves as a reminder that the SaaS landscape is evolving. The future belongs to those who adapt to the demands of hyperfunctional SaaS. I’d love to hear how you are adapting to the rise of hyperfunctional SaaS, and what role is AI playing in your journey? #AI #Superintelligence #HyperfunctionalSaaS #SaaSInnovation #Automation

57

4 kommentarer

The newest hype for Generative AI created also a lot of void. Where? In our heads. The dilemma is simple: there is hype for it, so let us use it. Immediately, before the others will 😃 The thinking of change through optimization is well justified, as this allows to make things better, faster, cheaper or at least give it a try. However there needs to be a business goal behind it, which justifies this effort. My consideration does not go into the direction, of how much will this cost or how long will it last. These are the key KPIs of a well defined business case. I mean ANY justified business goal at all in the end ☝️☝️☝️ I had the opportunity to work at Microsoft , when this hype started and I observed many companies around the world (desperately!) looking for (at least) one use case fitting into the frames of #genai. There is a saying the cultivate in Microsoft, which I support and shall quote as a well wrap up of this post: PAUSE. THINK. ASK. I am here to support your efforts, so is SoftwareOne Have a great rest of the week ahead of you! #business #technology #software #AI #xai

17

Kamel: How about coming to the Tech Supreme Court AI Edition in Nice? You'll be a witness. Me: "Yes! Awesome. Let me get that approved first." A few days later... Me: "Let's go!" That's how my Tech Supreme Court - AI Edition in Nice adventure kicked off. I thought it would only put Artificial Intelligence on trial, but it was also the occasion to get inspired by what the French Mediterranean AI Hub is all about. Thanks to Yannis Nakos (he/him), I discovered the IBM AI Labs. We discussed at length the combined strengths of Generative AI and Predictive AI leveraging Business Rules. Back in the golden era of Business Rules Management Systems, Ilog Rules and TIBCO BusinessEvents reigned supreme. We may witness these powerful technologies making a comeback. Incredible people here, and I found out that beyond #WatsonX, IBM is still a major #AI player. I just wonder why they don't advertise their open source models and tech platform more. If I had to define the Tech Supreme Court, it sits at the intersection of Political Debate, Theatre and Technology Think Tank. But even then, that's an understatement. You have to experience it to grasp the power of such an expressive form. So hats off to Kamel Amroune, Guillaume Tanferri, and Axelle Hering. I was the witness on the challenge of AI on Copyrights and Intellectual Property. Even though the prosecutor, Laurent Kratz, and the defense attorney, Djamila Aouada, asked thought-provoking questions, 10 minutes are short to debate the full extent of what it means for creators and businesses operating in the creation economy. A few key points: 🔹 It's a new paradigm, therefore requiring new regulation. 🔹It's a new paradigm, so we also need to establish new business models that are fair by design. Music streaming is a potential model, but it's very hard to make a living off it. I'm looking at you NFTs. 🔹 We could ban models that perform "Art Laundering", and create "Anti-Art Laundering" Systems, akin to AML in the Financial System, but then there are misalignments - see Japan's draft resolution and competitive posture about AI and copyrights. Geopolitic is a parameter in this equation. 🔹 I think, in this case, the solution lies in protecting the "Artistic Style" (or voice signature for singers), instead of the artistic product. It was also a chance to discover the Maison de l'Intelligence Artificielle. Its mission is educating on and promoting AI, plus it showcases French #AI startups in fields like robotics, computer vision, etc. Definitely worth a visit. Special thanks to Marco LANDI and his wife - I finally had the honor of meeting them. Word is there'll be a Paris Edition. If I were you, I'd get in touch with thedots right away ;) Beyond that, I'm just happy to have met such great people. Incredible leaders by day, incredibly funny, accessible, relatable and knowledgeable. The Tech Supreme Court is more than an event, it's an innovative experience. #IADATES #TechSupremeCourt

47

2 kommentarer

𝗕𝗿𝗶𝗻𝗴𝗶𝗻𝗴 𝗣𝗫 𝗳𝗶𝗹𝗲𝘀 𝘁𝗼 𝗥 PX files, which are widely used by national statistics offices to publish data, have been tricky to work with in R—until now. 😎 For the past two years, Lars Pedersen and I have been working on 𝚙𝚡𝚖𝚊𝚔𝚎, an R package that lets you 𝗿𝗲𝗮𝗱, 𝘄𝗿𝗶𝘁𝗲 𝗮𝗻𝗱 𝗺𝗼𝗱𝗶𝗳𝘆 𝗣𝗫 𝗳𝗶𝗹𝗲𝘀 𝗱𝗶𝗿𝗲𝗰𝘁𝗹𝘆 𝗶𝗻 𝗥. Here’s a quick demo of what 𝚙𝚡𝚖𝚊𝚔𝚎 can do. If you work with PX files, we’d love for you to try it out and share your feedback. 👉 Check out the package and documentation on GitHub to get started: https://lnkd.in/dRcBkAu9

61

6 kommentarer

-Didn't you get the memo? - eeeeh? What memo? - The CoDeMeMo - Continuous Delivery Metric Model I'm excited and happy to re-publish the Continuous Delivery Model. This model originates from the time when I founded (2007) and operated Praqma. A company specialized in Continuous Delivery implementation of software development processes. Praqma is now gone. The company was swallowed in a merger and acquisition process in 2019 and the model didn’t survive. It was stored away in a GitHub repository and never used again. ...until now! The model was optimized specifically to be used as an assessment tool for helping companies to map their current Value Streams and to build a backlog for implementing Continuous Delivery. The assessment process was based on “The Toyota Way” principles: #ValueStreamMapping, #GenchiGenbutsu (observe deeply), identifying waste (#muda, #mura and #muri), measure #LeadTimes and Continuous Improvement (#kaizen) based on #5S (sort, straighten, shine, standardize, sustain). Lean Software Development! The assessment process we developed would span five intense workdays on-site, with a mix of all-hands-on-deck Value Stream Mappings, interviews, observations, and lean coffee workshops. Some of the larger assessments we did had 120 participants in the first-day Value Stream Mapping workshop. I’m now re-publishing the model. I had to rewrite the code as I didn't have acces to the original codebase. I enhanced readability and responsiveness, but more importantly I’ve turned the model into an open source template you can inherit from in #Jekyll and host it as a static web. Build your own model! ...More on that later! But for now I just want the release this MVP so we can get the model live again. How To Read The Model: Its a matrix with six different areas to consider in becoming black-belt Continuous Delivery - Software as a Factory champion. Each area has 5 levels of expertise ranging from novice to expert. Each card in the matrix represents a certain theme or disciplin. And each card is scored - by the team itself, during the workshops - for effect on three parameters: - Throughput (lead times) - Feedback (data for decision making) and - Payback (...on technical debt) And then scored for complexity as one of: - A low hanging fruit - A team effort - A mountain to climb To show-case the features of the model I have added an anonymized and generalized write-up of some of the typical or illustrative findings and mitigation from the total of 24 assessments that I conducted on the basis of this specific model. I've started out with the three first cards in the "Build" area - more will follow. You can subscribe to the atom feed in the footer on the website to get updates. I would love your feedback on the usefulness and suggestions for improvement. If you and you team would like to do this assessment on your product and process, I'm happy to announce that I'm back in business, charged and ready - just reach out!

39

18 kommentarer

Model training costs are going up not down. So, are we dependent on big AI SaaS companies? I still see an alternative in training for specific purposes models up to 8B parameters at decent costs

8

I don't buy into Klarna's AI transformation story. It feels a bit too, say, fairy-tale-like. Klarna, recently made the news again for ending their fancy SaaS subscriptions in favor of AI-accelerated software development. I've had the "pleasure" of dealing with Klarna AI, and while it's pretty grounded, it's no Robochat-007 they make it sound. AI-generated code, more often than not, slows down the releases rather than accelerating it. Then there is a whole lot of compliance, security, and other mumbo jumbo. I wonder what mystical OpenAI models they have that other lesser mortals don't! Klarna has an IPO on the horizon, and this whole thing feels like a carefully staged PR stunt. By playing the AI card, they look like tech pioneers to future investors. Suddenly, they can justify what really seems like an after-effect of a massive drop in valuation due to overspending, interest rates, and competition. Meanwhile, the customers! Huh, what about them? They should be happy that Klarna AI Bot can tackle their Python questions. Jeez, some people just can’t be pleased, can they? --------------------- Hi, I am Maheep. Follow me for a candid, fun, and cynical perspective on tech and business. #klarna #ai

7

10 kommentarer

A lot of people ask me why I decided to open MINDBANK AI in the US and this graph sums it up. It explains why the US leads the #EU in #deeptech. The EU leads in R&D as a % of GDP, but they lack the early stage support for #entrepreneurs. Startups aren't easy and while Germany does a great job of supporting #startups, the financial support from Angels, VCs, and others does not compare to what you can get in the US. @esmtberlin #Investors wanted: https://lnkd.in/eA7k7Z_a

24

3 kommentarer