Kale Miller

Major news flash: leveraging part of Google’s $1B Digital Futures Initiative we just signed a 1-plus-1-equals-15 deal. Bringing together the two AI and Science powerhouses CSIRO and Google, this partnership will be an inflection point in AI-powered scientific discovery. Press releases with more info here: Google: https://lnkd.in/gDxbsR4b CSIRO: https://lnkd.in/gG2EyqMz Professionally speaking, for my whole career I’ve been passionate about two things more than about anything else- and by far at that: #science and #AI. Can you imagine the enthusiasm I’m feeling at this moment about getting to marry the two so spectacularly…we only got here because of so many others who share that enthusiasm and determination - just calling out a few names on both sides of the aisle: Jon Whittle Scott Riddle Astitva Chopra Heather Tompkins LINDA ALLARD Dr Rebecca Michael Eliza Keck Alison Donnellan Grace Chung Melanie Silva - and thank you to many more whose relentless work has made this possible, you know who you are. To all scientists out there: don’t just watch this space: get in touch and check out the links in the press releases - you do NOT need to be an AI expert or seasoned data scientist to have fun at this party…this is for all scientists and we’re just getting started!! Nicholas Thompson Cade Metz Melissa Heikkilä Mike Cherney https://lnkd.in/giitUt7E

136

23 Comments

‘Use Cases are sexy. Governance? Not so much. And yet in a fascinating twist…mature AI adopters have already learned their lesson: ➡️ We're at the start of an AI marathon; not a sprint. ➡️ AI Governance is the fuel for sustainable value; the antidote to short term AI productivity hype. 🌟 In the long term, responsible AI and ROI are more linked than you think…’ Great post from Sam Burrett. #ai #roi #aiusecases #responsibleai #aigovernance

16

7 Comments

Really well illustrated diagram on Growth Loops - an essential part of product led growth. Your key takeaways: -Reduce friction to getting to an Aha moment -Make it easy to share and add others to your product in a free/viewer mode -Make it easy to pay -Make sure your product gets more valuable the more people use it What are your thoughts? #product #design #plg #quantana

6

1 Comment

At click speed...

1

We recently crossed $19M in ARR at Buffer… for the second time. It's been a journey, to say the least. We first crossed $19M in ARR in September 2018. It's hard to believe that's six years ago. The chart tells the story well - it's been a rollercoaster and the hardest thing I've worked through as a Founder CEO. There's a lot of pain represented in that curve. To be declining for multiple years is demoralizing and exhausting. But I always planned to build a long-term business and was determined to lead us back to thriving, so I remained optimistic despite the fact it took some searching to find the path to new growth. In many ways, this is a failure for me as CEO and for us as a business. It truly is. I fully own that as a fact. And yet, I see it as a success too. We plateaued, declined, and rediscovered growth as a business in those years. I grew significantly as a leader, and we'll soon cross 14 years since I started the company. We went through this cycle without doing layoffs, thanks to the fact we entered our decline with profitability and a healthy cash balance. Most importantly, we figured out what we are for as a business; who we're serving. For me, that involved reminding myself why I even started Buffer. I love building for and serving entrepreneurs, and I love helping creators and small businesses get off the ground and thrive.We re-committed to the business model that serves these folks, and drove our early growth, which is a highly generous freemium SaaS offering. I get a kick out of delivering immense value compared to what we charge for the product, and challenging myself to continually improve our tools. Through the low lows of our multi-year decline, we rediscovered our DNA and the vital components of our culture. We found a way to blend that with what the world and our space looks like today, to arrive at a strategy and way of working that's helping us thrive again and ensure we can exist for another decade and beyond. I found a new level of passion and conviction for what we're doing, for the potential we have as a business, and the ways I can lead and do work that fuels me to help us fulfill those ambitions. The first time we crossed $19M, the writing was on the wall for our looming decline. Pace of product improvement and innovation had stalled; we were debilitated by tech debt we didn't yet know how to manage. We succumbed to squeezing revenue out of existing customers, and were already seeing a steep decline in new paying customers. This time around, we are moving faster and more boldly than we have in years, and we've re-centered ourselves around growing by serving more customers and continually adding real value, rather than through price increases and short-term growth hacks. There's no doubt we will run into challenging cycles again in our future. I'm confident we'll find our way through those too. To me, that's a natural part of building a long-term business. Onwards to $20M, again.

381

49 Comments

Worth reposting here. 😅 https://lnkd.in/g4Kktqn6 I’m personally really excited for the soon approaching back half of the hype cycle where we can spend time adding some actual software engineering and scalability expertise to the decades of poorly written Python code from the ML world and start implementing features users want without slapping the widely abused phrase “AI” on everything. 🙌🏻 If you think it needs a chat bot, it doesn’t. If you think you’ll get more engagement or find market fit from a “✨ Try it with AI” button slapped on everything, it doesn't and you won't. The best AI-driven features are specific, solve real problems and are so seamless in your product that users don’t need a desperate “LOOK WE HAVE AI TOO” billboard to find them.

8

2 Comments

There are lots of awesome works (research papers, tools, libraries,...) right now around Large and Small Language Models (LLMs and SLMs). In you're looking for ways to : * have more control about LLMs output * better ouputs (avoid hallucination) * faster execution  * lower cost I really recommend to look at {guidance} python library which allow you to write guidance program instead of language text prompt (prompt engineering is not a dev thing right :-) The idea with {guidance} is that under the hood, Large Language Models get back the next token and the next token and so on (4 characters each) based on the input prompt (system, user …) based on intensive maths calculation (token probability distribution) which can eventually be change with the temperature parameter. So, what about techniques to avoid generating tokens (because you already know you want it), constrain tokens with possible values and so on... I really like the analogy mentioned by Harsha Nori taking about {guidance} is like a code compiler that transform high level language (python, .net) to specific hardware optimized instructions. here is Microsoft BUILD session talking about {guidance} https://lnkd.in/evpSmbsa #SLM #LLM #AI #MicrosoftAI

13

1 Comment

💡 Do you know how much is your data worth? A lot. And big AI companies are taking them without asking for permission. Happening right now: ⚠️ Meta’s AI is scraping users’ photos and posts. Europeans can opt out, but Australians cannot : https://lnkd.in/e9RWC54F ⚠️ LinkedIn scraped user data for training before updating its terms of service : https://lnkd.in/eAJjYcRT ⚠️AI start-up Anthropic accused of ‘egregious’ data scraping : https://lnkd.in/e6-ZsRWD And that's only the tip of the iceberg! Don't fool yourself: *everything* you ever put on Facebook, Instagram, GMail, LinkedIn, etc. is somewhere inside AI models. Anything privacy-related from big companies boils down to a nice "trust us", from cookie requests to insidious TOS that discreetly allow data collection. *The best part*: companies not only get data for AI in questionable ways, but they also use the AIs themselves to get data in a questionable way! Like for price-fixing in real estate: https://lnkd.in/e7iXs4rc Privacy is dying. #AI #privacy

4

update: Oratori AI I have added OCR recognition. Users can click photos. Crop the part. Pass questions to the chatGPT API. ChatGPT will solve that. Added features for: Maths, Physics, General and Coding.

36

5 Comments

Day 13: Case Study: The Analog Ecosystem Powering Dynamic Neural Networks Imagine a supply chain that organizes itself, aligned to shared goals in real-time. That’s what we accomplished in the Australian fresh produce industry—a system once weighed down by inefficiencies, waste, and unpredictability. It was fragmented, reactive, and out of sync. The challenge was clear: How do we fix it? The solution wasn’t just to tweak the system, but to reimagine it completely. We stopped thinking of it as a traditional supply chain and envisioned it as a living ecosystem. An analog ecosystem powered by collaboration. We introduced ‘One Touch,’ a simple yet revolutionary system that connected growers, manufacturers, wholesalers, and retailers in real-time. They could see exactly what was happening in the supply chain, at every moment. Inventory, demand, availability—all dynamically adjusting, like neurons responding to live data. The impact? Over a billion dollars saved. Waste slashed. Inventory moved faster. What was once a broken system is now self-organizing, adapting, and thriving. This is the power of thinking differently. When you create an ecosystem that evolves with the world, it transforms how we live, work, and trade. Now, we’ve taken that analog ecosystem and digitally twinned it with a Dynamic Neural Network. And this is just the beginning of what’s possible. #SupplyChain #Sustainability #Innovation #ai #machine #NeuralNetworks #BalancedEcosystemScorecard

2

Anthropic launches tool to connect AI systems directly to datasets. This seems like a really interesting new method for connecting AI to tools, function calling or discovering possible application intents. https://lnkd.in/gUZqgq4Z

5

This is more of a "talking to myself in public" thing than any kind of fully fleshed out hypothesis, so take this with the appropriate grain(s) of salt. But if I were interested in AI and were looking for an angle to explore that might yield something interesting, I'd start working on learning "Multiscale Mathematics" (also known as "Multiscale Modeling"). This is rooted in my personal belief that there has to be a "multi-scale" aspect to intelligence because there is such an aspect inherent in our underlying physical reality.

2

How Dovetail grew to a $970M (Australian) valuation in 5 years—and created a new SaaS category: I chatted with 4 members of their team, and 4 UXR leaders, to learn the details. 1. Product-led growth, design-led company Dovetail's business model is decidedly product-led growth. The company relies on the product, and inherently viral features like research reports, to acquire, convert, monetize, and retain users. But design-led is its culture. CEO and co-founder Benjamin Humphrey has been in the details of the design since its beginning. And the product "feels" like it as a result. 2. Creating a hub for a neglected persona Rewind to 2016. While engineers had JIRA, Designers had Sketch, PMs had Google Docs, and analysts had Tableau, User Researchers had... nothing? Their workflow was a mess of hopping between different software. Ben had lived it as a product designer at Atlassian. There was a plethora of software focused on how to build, but there was almost nothing about what to build. So, after proving out the idea with an MVP, he recruited his Atlassian Engineer friend, Bradley Ayers, to join him. 3. Mastering the PLG motion Dovetail excels in all layers of PLG: → Acquisition: they don't use any outbound sales, focusing on product acquisition → Core problem communication: the home page emphasizes improving how you make product decisions → Reduced friction: its onboarding is focused on a standard walkthrough, but it is 10 steps → Activation: Dovetail has an intuitive product, examples, and templates to take you from setup to habit → Retention: as a system of record, the product gets stickier as more (qualitative) data is put in it → Monetization: the free plan has unlimited functionality, but only 1 project → Expansion: it has a natural growth of free users, who eventually become editors 4. Creating powerful, reinforcing growth loops Dovetail 2 two core loops: • More paid seats helps them build a better product, which helps UXRs succeed, which drives more UXRs and paid seats                                 • UXRs succeed, driving collaborators to join, which go from free to paid seats As well as 2 ancillary loops: • The more paid seats, the more it can invest in inbound sales • The more free seats, the more expansion products like Channels (to analyze customer support data) it can build 5. Building towards a category of one While there are competitors in the research insights space, Dovetail is building out its own category. It's not quite transcription. It's not quite a hub for research. It's that and more. This has helped it create its own, new category in SaaS. As Benjamin, still the CEO, told me: "We're building a system of record for all insights... It's a friggin' long, 20-year journey. You need to be a little nuts to have that level of ambition. "

427

42 Comments

My thumb rules for generating better results from LLMs A.K.A. "Communicating with kids" 👶 how working with language models is surprisingly similar to talking to your kids: 1. Give Clear Instructions : Just like with kids, less is more. Say only what's relevant. 2. Give Time to React : Fancy prompt tricks are like asking deep questions—sometimes you just need to slow down and wait a bit for the genius answer. 3. Be Patient with Repeats : Sometimes you have to ask more than once. LLMs, like kids, might not get it right the first time. Or the second. Or the third...

3

"How does a vision LLM outperform OCR, and why?" I got this question from our customers a lot and we distill the answers into the blog: https://lnkd.in/guv8waCs Overall, Vision language models (VLM) address key OCR shortcomings by: 1. Accurately interpreting low-quality images and complex layouts 2. Understanding context and semantic meaning in documents 3. Handling multiple languages and scripts seamlessly 4. Reducing post-processing overhead #pdf #word #ocr #llm

36

3 Comments

Some thoughts on Alignment: - It's just fine tuning (basically)! - It's all about the labeled data - Some recent work Snorkel AI on what we call "enterprise alignment" + a mini rant on what we need to focus on to actually make AI work :) There have been a plethora of terms to describe some form of tuning an LLM with respect to a labeled dataset: - "Fine tuning": Tuning on labeled training data - "Instruction tuning": Tuning on prompt, response pairs. - "Alignment": Tuning on some form of preference data (e.g. LLM A vs. B, yes/no, written feedback, etc.) Conceptually though: it's just tuning LLMs for a labeled dataset either way! For the above reasons- Snorkel AI we just prefer to call the above all "fine-tuning" for simplicity; and, believe that this (+ prompting) will all eventually converge and become lower-level details of LLM adaptation. What will matter then? And what matters most now, as the algorithms and infrastructure for doing fine-tuning has already standardized and commoditized? The labeled data! This is evinced by BigTech and LLM providers spending $ Bs on labeled data for fine-tuning and alignment, and organizations like OAI beginning to research more scalable approaches for future superalignment objectives, like weak-to-strong generalization (which we started working on 8+ years ago, https://lnkd.in/gnBqS422). The problem for enterprises: the use cases, objectives, policies, regulations, tone of voice, etc. that they need their AI models to align with are *not* the same as the generic ones these LLM providers are tuning/aligning to! We call the challenge of aligning to enterprises' specific objectives/policies enterprise alignment. It's difficult especially because private data & internal subject matter experts are usually required to do the data labeling (i.e. can't just be outsourced, even if there were $ Bs to spend!) Snorkel AI we've show that our programmatic approach to data labeling & development can solve this challenge, leading to custom alignment in a few hours of development- for ex., aligning a finance chatbot to be compliant with financial advisory policies (+20.7 pts. above baseline) in several hours. Check out our work here: https://lnkd.in/gW5hbqJF As for the "mini rant" - mostly bait to read a long post on a niche technical topic :). However, many of us in AI have a strong view that: - (A) Superhuman AGIs are not right around the corner. - (B) Enterprises are struggling to deploy AI to impactful production settings- and the party's over for all of us (and all the positive impact that we *know* AI can have) if we don't deliver value ASAP. - (C) Safe systems at scale are usually built through years/decades of incremental and collective engineering (e.g. think air travel), vs. secretive leaps forward. That is: let's all learn to walk safely first- then we'll get to flying saucer seatbelts!

147

4 Comments

For our current defending LLMs project I’ve been doing a bit of ye olde model training. This is the output from a recent model for detecting SQL injections, fine tuning on top of DistilBERT (https://lnkd.in/g_Dh8KDJ). Happy with those results. I really do believe that more of us need to be exploring how to properly secure our LLMs as opposed to just treating them like cheap labour. #splunk #splunksecurity #AI #ML #Steve #pirates

45

3 Comments

How Dovetail created a new SaaS category while growing to a $970M (Australian) valuation in 5 years: (I learned the details by having a chat with 4 members of their team, and 4 UXR leaders.) 1. Product-led growth, design-led company While Dovetail's culture is design-led, its business model is decidedly product-led growth. The company relies on the product, & inherently viral features like research reports, to: • Acquire • Convert • Monetize • And retain users 2. Creating a hub for a neglected persona Let's go back to 2016. While: • Engineers had JIRA • Designers had Sketch • PMs had Confluence • Analysts had Tableau • User Researchers had... nothing? Their workflow was a chaos of switching between different software and tools. Ben had lived it as a product designer at Atlassian. While there were many softwares focused on how to build, there was almost nothing about what to build. So, after validating the idea with an MVP, he recruited his Atlassian Engineer friend, Bradley Ayers, to join him. 3. Mastering the PLG motion Dovetail excels in all layers of PLG: → Acquisition: they don't use any outbound sales → Core problem: the home page emphasizes on improving your product decision-making → Reduced friction: its onboarding is focused on a standard walkthrough → Activation: an intuitive product, examples & templates → Retention: as a system of record, the product gets stickier as more data is put in → Monetization: free plan has unlimited functionality, but 1 project → Expansion: natural growth of free users, who become editors 4. Creating powerful, reinforcing growth loops Dovetail has 2 primary flywheels: • More paid seats help them build a better product, which helps UXRs succeed, which drives more UXRs & paid seats • UXRs succeed, driving collaborators to join, which go from free to paid seats As well as 2 secondary flywheels: • The more paid seats, the more it can invest in inbound sales • The more free seats, the more expansion products like Channels (to analyze customer support data) it can build 5. Building towards a category of one Dovetail is building out its own category even though there are competitors in the research insights space It's not quite transcription or a hub for research. It's that and more. This has helped it create a new category in SaaS. To learn about the complete story and breakdown, check out the deep dive here: https://buff.ly/4dBAdHs

61

4 Comments