Igor Royzis

A lot of comments were made about Stackoverflow's decline in traffic as ChatGPT grew, as shown in the attached image. Yes, it's an undeniable fact - LLM feed off of human content with no credit or attribution, and profit off of others' labor, with no current recourse. We're in uncharted territory, and just like with Uber, or AirBnB, laws and regulation will be slow to catch up. But catch up they will. There's a reason why AirBnB didn't bankrupt the hotel industry. There's a reason Uber and competitors are not operating in some countries, and leaving others. People, as represented by capable sovereign states, understand you're playing on the field, but by different rules. So, they either change the rules for everyone, or you adjust your play. None of this detracts from the potential of machine learning, large language models, etc. It should not however, allow us to get carried away into thinking that everything will be replaced by artificial general intelligence. The brightest minds on Earth are still that. They evolve off of meaningful work made by others, progress it, and allow others to do the same. That is the future. At this point in time, this amalgamation of knowledge can be summarized and exploited for profit, with LLMs that freely enjoy access, legally defined or otherwise, to it. That faucet isn't endless, and humanity isn't done innovating. Will a more powerful AI play a role in evolving humanity? Sure, in some fields. Alongside humans. Dario Amodei from Anthropic explains this in detail in a recent podcast with Lex Fridman https://lnkd.in/dtTXpQJz

4

4 Comments

#DB2 for #IBMi conversation with Charlie Guarino and Ryan Moeller. Check out the future of #SQL and #SQE (Sequel Query Engine) https://hubs.ly/Q02NPJps0 Thanks to TechChannel for hosting!

5

Despite the very large number of data lineage solutions out there, getting *full* visibility to your data pipelines remains a common challenge for growing data platforms. And if you have some Spark-based pipelines in your data stack, whether it's Databricks, Amazon EMR or others, you should prepare to work a lot harder... ! However, there's a class of new solutions, such as OpenLineage and Foundational that go beyond traditional SQL parsing. Read more on our blog: https://lnkd.in/gCEdAXSE #datalineage #spark #datagovernance

12

Will Ahmed has fantastic advice here that goes beyond hardware/software startups. Here are my takeaways: It’s crucial to have a strong long-term vision for what you’re building, but equally important is the ability to distill that vision into a Minimum Viable Product (MVP) and have a clear plan for scaling. I’ve learned firsthand that trying to scale too quickly without the necessary infrastructure can lead to a loss of quality or burnout. I highly agree with point #4 on CAC and referrals. Securing high-value influencers who are genuinely invested in your product can significantly drive referrals. This isn’t as easy as it sounds, and it takes time to find the right partners. The key is to focus on 2-5 partners who believe in your vision, can provide valuable insights to refine your MVP, and will drive referrals through authentic recommendations. Will’s insights are a reminder that building a successful company is about balancing vision with execution. You need to have a great big idea, but know how to break it down, scale it smartly, and build genuine relationships that fuel your growth. The right focus, the right people, and the right timing can make all the difference. What are your thoughts on this approach?

1

This doesn’t mean coding is dead (sorry, Jensen Huang). Instead, what we are seeing is how deeply polarized the tech market has become in 2024 and how wide the gap is growing. Fewer tech jobs = fewer tech companies hiring. And traditional SaaS companies are at the center of this shift. Because: 1) VC money isn’t flowing like it used to. Historically, most of the VC capital raised by tech startups went to hiring and customer acquisition. Now, with funding rounds slowing, there’s less cash to fuel headcount growth. 2) Growth has stalled. Many SaaS companies are paying the price for their “growth at all costs” phase, especially now that the ZIRP era is over. There's a ton of zombie companies out there—still walking, but essentially dead. Some are going for “life raft” options: selling at lower multiples, turning to private equity, acqui-hires, or raising funds at steeply diluted valuations. And of course, this means no expansion, hiring freezes, or worse—mass layoffs after the 2022 bubble burst. Now, on the other side of the gap is AI, grabbing all the money, headlines, and triple-digit growth. But even within AI, there are two distinct markets. 1) Core AI infrastructure – AI-focused hardware, software, and services. 2) AI-embedded companies – Businesses that natively integrate AI (like LLMs) into their applications. In 2024, every company is calling itself an “AI company,” slapping AI into their apps or pitch decks to ride the hype wave. This approach works for now—they’re raising capital, iterating, and searching for product-market fit. But in 5–10 years? Many of these companies might not be sustainable. Their AI offerings are often easy to replicate, and their competitive edge fades quickly. Moat-less. The good news is, we are starting to see more enterprise spending in the application layer and I think more value is being created. Agentic applications are really moving the needle now. Which is critical - the entire AI movement rests on the future of application builders. AI infrastructure, on the other hand, has more staying power IMO. These companies—especially those working on hardware—are playing the long game. Yes, they’re capital-intensive, but the opportunity is massive. The proverbial diamond mine of AI is far from fully explored. And as history shows, the ones selling the tools to dig the mine will often be the biggest winners. What do you think? Let me know your thoughts. P.S. Great data from Bigdata.com and Linas Beliūnas

7

With Redpanda Data's recent acquisition of Benthos I've gotten some questions about how it compares to Arroyo. In short: Benthos is *stateless*, while Arroyo (like Flink or RisingWave) is *stateful*. This may sound small, but adds up to a fundamental difference in how these systems are designed, built, and used. In a streaming engine, statefulness combines two inter-related features: * Repartitioning data (GROUP BY in SQL) * Remembering previously seen events With just these two features, we can support most of normal SQL (window functions, joins, aggregation, etc.) plus time-oriented streaming operators (time windows, event-time semantics, watermarks). Stateless engines on the other hand are limited to simple transformations and filters (SELECT and WHERE in SQL terms) that operate on individual events. But all of this is a tradeoff; stateful systems are much harder to operate. So if you don't need joins, aggregates, or windows, a stateless system can be a better choice. They also work well as infra glue, routing data between different systems—Benthos in particular has a huge connector ecosystem. For a deeper dive on this, I wrote down some thoughts on our blog:

57

4 Comments

My original ARR reporting script at Intercom was thousands of lines of Python code. And honestly, it was a pure miracle it worked at all. The script did one thing: Sort customers' charges by billing number, calculate the difference between periods, and return expansion amounts. Simple enough, right? Not quite. It actually turned into a multi-thousand-line nightmare that manually pieced together various pricing models. It had to account for every possible way someone could pay us: - Monthly - Quarterly - Annually …using different methods. The script was brittle— It broke any time we changed something about our business. Worst of all, no one could use it but me. That meant nobody else could do ARR analysis. As we scaled, this became a major problem. What intimidated me most wasn't the complexity— It was not knowing how to turn this Python script into something that would scale for… - Me - Finance - Everyone who needed to report on ARR We needed a new approach… One that wouldn’t break. One that everyone could use. So we moved everything to a database and settled on a specific table format. That table format is now the basis upon which our ARR Reporting offering at Equals is built. It was a complete game changer in how the entire company accessed ARR reporting. The goal was clear: Turn all our defined logic into a source of truth. So, we started with the foundation. A single comprehensive table— One place to answer +90% of the ARR questions we had. The SQL queries were complex — that's unavoidable. But we made them comprehensible: - Keep the written query organized - Add clear comments - Break the query into bite-size components - Make it approachable to others Your ARR table should play nicely with other tables you build around it. It’ll help you answer the majority of questions related to ARR… …but it’s also the centerpiece for other questions you’ll want to ask about the business. That's why design framework matters— It should make it simple to join additional datasets to your ARR build.

20

2 Comments

Check out this article from #gizmodata founder: Philip Moore demonstrating how to use #sql for #olap #hierachical #aggregation! https://lnkd.in/eCYfsKDB In this article, he demonstrates how to use #duckdb's powerful #recursive #cte to build an "exploded hierarchy" table - which makes multi-level and multi-dimensional aggregation a breeze. With this article - you will see how to build a #rolap #cube - with no expensive third-party tools required! Contact #gizmodata today to get a consultation on how to #olap-ify your data - at: [email protected]

14

⏳ Time is almost up to register for Wednesday’s Omni 🎙 Live: Exploring the dbt <> BI workflow Join Christopher Merrick and Jack Sweeney for a brief chat and demo diving into: - What the dbt <>BI workflow *should* be - Tips for deciding what should live in the dbt vs. BI layer - How we’re solving this with our bi-directional dbt integration Register now 🎟 https://bit.ly/443Q9yE

14

Prompting for scale will soon be obsolete... i think...i hope. Prompting is an error-prone, discovery-based process, unsuitable for scalable AI. Many folks rely on black-box LLMs behind API walls (e.g., Claude, GPT-4), lacking direct control over LLM model weights. Programmatically custom-pruned LLMs will be the future of controllable, production AI, not prompting. hint--> #steeringvectors

18

6 Comments

Unlock the Power of Data Lake Table Formats with Alex Merced, Developer Advocate at Dremio! 📊 Join Alex Merced as he delves into "The Who, What, and Why of Data Lake Table Formats." 📅 Save the Date: https://lnkd.in/eFZ6m5HV 🔍 Key Learnings: - Explore how data lake table formats handle large datasets. - Learn how these formats optimize query performance on extensive data. - Understand the importance of data durability and recoverability. - Discover how these formats ensure data is auditable and compliant. Don't miss out on this opportunity to level up your data analytics game! 📈 #DataAnalytics #DataLake #Dremio #TechTalk

4

AI is being treated as a magic hammer in the development community? Take a look at a report created by Retool on The State of AI. What are your thoughts? #redefiningwork #technews #AINews #talent #copilot #Chatgpt #claudeai

4

Stop (over-) engineering your prompts Generative AI is really cool. I use several tools in this category every day for writing suggestions, image generation and more. The tools can do amazing things, especially if given the right prompt. But there's a limit to how much prompt engineering one should do, especially if you're trying to generate text for mainstream consumption. This topic is already a bit of a bandwagon but this topic has been bugging me for about a week so I'm going to write about it. Generative AI relies on a class of machine learning applications broadly described as deep learning. Deep learning boils down to feeding a neural network a ton of content that has been classified into groups and then iterating over the content to allow the network to identify patterns in the content. Basically the network builds up statistical relationships between the words or other elements in the content. So it knows, for example, that if a piece of content is a recipe the word tomato is often followed by the words sauce or paste and very rarely by the word stem. But in a document about farming tomato is often followed by the word seed, starter or stem. The deep learning model has developed an elaborate matrix of statistical relationships between words in different types of content. If you described CHatGPT as an elaborate word guesser you wouldn't be wrong, just over-generalizing. Large language models are trained on billions of documents. So how, you might ask, are the documents classified in order to feed them into the neural network? Did someone categorize the web by hand? Of course not. The developers of the LLM leveraged Natural Language Processing (another deep learning application, btw) and other tools to categorize documents before they got fed into the model. Okay, but, again, what does that have to do with prompt engineering? When you ask an LLM to generate a piece of content, say a blog post about cycling in Northern California, you are specifying a category of content. If you later asked for an "image in Rubinesque style of a man and a woman engaged in flower picking while overhead rages a space battle between starships populated by photorealistic blob-like aliens carrying professional immersion blenders done in the style of pointillism" you have specified many categories. You're essentially asking the LLM to extract elements of the content that it was trained on that match your descriptors and which have mathematical relationships to other adjacent content elements and to include those in your generated image or blog post. And ... prompt engineering? The categories in the content were identified with Natural Language processing. Your prompt should be in natural language, not some over-specified, highly engineered artifical style. Don't over-think it, work with the model, don't try to control it too much. That's it.

2

🎯 BENCHMARK: Redpanda vs. Confluent: A performance and TCO benchmark It's here — the definitive comparison between Redpanda and #Confluent so you can confidently choose the right fit for your real-time data projects. This report by McKnight Consulting Group compares the two #streamingdata solutions with the insights you need to know about performance, cost efficiency, and scalability 👀 Download the free report here👇 https://lnkd.in/dTFfBFFs #kafka #apachekafka #datastreaming #realtimedata

19

We are LIVE at PASS Data Community Summit! Join us for Day 1 of Postgres Conference Seattle for a "Journey through Postgres" w/ Zak Tedder in Rm 420 or "Tips and Tricks for developers moving to PostgreSQL" w/ Eric Lendvai in Rm 422. Check out the full schedule here https://lnkd.in/estNqRHm #Community #Postgres #Data #peoplepostgresdata

1

Founder: "As you will see in our demo, our product is fully functional, and the initial OpenAI API dependencies you highlighted as a major red flag for the current $4B valuation have been thoroughly addressed." The demo: #RightClickSaveAS

49

12 Comments

Here's a quick way to discern AI hype from real, applicable use cases next time you're reading a product pitch: Substitute every instance of "AI" with "LLM."  That's the current tech - large language models. They are incredibly powerful and disruptive, but "artificial intelligence" is a misnomer most of the time.

14

2 Comments