Ned Batchelder

Had a chat with Eric Meyer and Brian Kardell about CSS masonry, does it belong in the grid spec or not? With a bit of a discussion about how and why we ask for developer feedback on CSS. https://lnkd.in/d3C-_dwq #css

33

I'm so incredibly excited to share something amazing I've created over the past few months, Writeprint.ai. If you've explored the amazing world of generative AI that tools like ChatGPT unlocks, you've learned that ChatGPT doesn't write like you. It sounds like.. ChatGPT. What if you could generate AI content that sounded like you wrote it? That's what Writeprint.ai unlocks. Upload some samples of your writing and Writeprint.ai will start generating content that sounds like these samples. It's super easy to add these samples and even easier to start generating content. While I haven't formally kicked this out to the public, I wanted to share it with you, my friends and family. If this sounds useful to you, I hope you'll give it a try. If you know someone that uses ChatGPT, I hope you'll share it with them. I also would love to have a conversation with anyone using ChatGPT to generate content to better understand how you're doing things. https://www.writeprint.ai

43

6 Comments

We love hearing from people who use Time Travel Debugging (TTD) to dramatically reduce the time spent diagnosing issues. Our security team recently worked with the compiler team to solve a nasty deadlock when using ASAN and shared their journey. In the narrative below they make use of the reproducibility that trace files offer along with hardware breakpoints to solve the problem. If you already know how to debug with Windbg the barrier of entry to TTD is low. As a bonus tip, you can use the timeline view to construct call queries or memory queries instead of using breakpoints and stepping through execution. Or, you can use “dx -g” with a call or memory query to generate a table of information about the call parameters / memory values / etc (my favorite!) Do you have a good story to tell? Please share as inspiration for all of us! ----- When rolling ASAN out across Windows, we hit an interesting bug on a particular build target. Without ASAN the program worked as expected, however when we turned on ASAN we experienced a deadlock every time. I threw the program into TTD and the problem reproduced immediately. From the trace, I could see all threads were blocked on a lock. From there, I rewound the trace to the start, setting a breakpoint on the locking functions, conditional on my specific blocking lock, then executed forward. Very quickly I could see that one location locked the lock, but its epilog ended up calling unlock with a different point. Returning to the trace with the compiler team later, we were able to narrow down the issue to a specific place where, under ASAN combined with a tailcall, the stack would not be appropriately adjusted causing the saved registers to be clobbered. These clobbered registers then got fed into the unlock routine, leading to the inconsistent lock state and eventual deadlock. The fix for this issue is in VS 2022 17.10 Preview 3. Start to finish, from reproducing outside of the debugger to identifying the offending codegen took around four hours, with another hour spent later with the compiler team to identify the specific behavior that was at fault. Without a TTD trace, root causing this would have taken much, much longer. While I find TTD valuable in day-to-day debugging, especially in instances of toolchain issues, the time saved by the replayability is incredible.

45

4 Comments

Its about solving the problem, not 'having AI' in the system. This is from last March, but the Wave 1 vs. Wave 2 distinction does seem to be playing out. Less 'generate me content', more 'synthesize vast amounts of relevant data and aid in decisonmaking or signal extraction'. https://lnkd.in/ehFBFC32 They don't talk about multi-agent systems here, but looking forward to more startups with a focus on leveraging agentic design patterns (https://lnkd.in/ed5rMmFe) to streamline real world workflows.

4

1 Comment

Having a ton of fun writing an ORM for V2 of Case Kit. Typescript has such an incredible type system. I have autocompletion for column names, the ability to include relations, relations of relations, etc etc etc... all returning properly typed data (no classes in sight). It's been really satisfying seeing it coming together, and it's been great getting my hands dirty with some (relatively) complex generics. And to head off the obvious question… why aren’t I just using Prisma? Well there’s a couple of things I’m doing differently that I think are marginally better: there’s no code generation involved here, it’s just types (inspired by Drizzle); there’s no Rust binary you need to deploy, and it hews much closer to Postgres behaviour and syntax (the aim here is not to abstract away which database you’re using). But the real blocker to me using Prisma was that its extensions don’t support modifying queries of nested relations - a hard requirement for me due to the way I have multitenancy and soft deletes set up. It’s not shown in the video but the Case Kit ORM lets you write simple middleware that will be applied to every level of a nested query. Perfect for setting timestamps or adding clauses on tenant id or things like that.

21

This is a fun blog post about first steps with Rust -- and what looks like a legitimately useful tool!

22

After a year of work my book Building A Startup - A Primer For The Individual Contributor is finally available! https://lnkd.in/erThxvpP If you’re working in or around startups as an IC or technical leader this book is for you. It’s based on 20 years of startup work and will help set you up for success to do more with less, move faster and more correct, and spend more time building features and product people love and less time banging your head at roadblocks and bottlenecks.

115

12 Comments

Anthropic released another paper on their ongoing work towards interpretability of large foundational models like their Claude 3 family, built on previous work by many. Their work continues to be enlightening and inspiring. The findings in this work are generally similar to previous findings on smaller models but scaled up to Claude 3 Sonnet: 💡 they find large numbers of "features" corresponding to combinations of neuron activations and these features map to concepts that we recognize/understand: for example, features for transit infrastructure, particular people, concepts (like bias and "objects lacking sentience"). This might sound familiar from classical machine learning, but the interesting thing is that there is no single thing/neuron encoding each of these concepts, it's a particular combination of neuron activations that embeds these concepts in "superposition" in the network. 💡they observe a sort of "distance" of these feature activations that arrange themselves "near" semantically similar features. When they use their sparse autoencoder approach to learn more and more features they observe that the level of abstraction in the detected features becomes more granular, in ways that are interpretable and meaningful. 💡when they artificially fire these activation patterns during inference, it affects the output in ways that relate to the "feature". For example, asking Claude "what's the most interesting science" normally it responds Physics, but if you clamp the feature activations for the feature "Brain Science" to 10x their max values then Claude responds: Neuroscience. Ways to interpret _and_ influence the model. Great readable work [1] along with an interactive explorer of some of the "features" themselves [2] 🎂 There's an easter egg too: if you scroll in the feature explorer tool and try and select one of the "Randomly selected features from 1M" you get a message "if you want to see the rest of the features, we're hiring!" with a link to their recruiting page :) [1] https://lnkd.in/gZbparwp [2] https://lnkd.in/gC9eus53

19

I've been improving the Communal Plot since it went live a couple weeks ago! The features I'm most proud of so far: - The plot updates fully in real-time! No need to refresh the browser to see new responses come in. - Analytics around percentiles, correlation, and clustering that also update in real-time as new responses come in. - Wordle-esque text sharing (this has been fun in my group chats!) - The timelapse feature which shows how today's plot was formed

26

"The core art of Operations Management is observing the messy, human chaos of business and applying order to it." The Operator's handbook is a great guide to the various operations disciplines like #businessoperations or #bizops, #revenueoperations or #revops, #financialops or #finops,#legalops, #marketingops, #peopleops and #productoperations / #productops that have emerged as a result of shift from hardware to software-based value creation (see my previous post too --> https://lnkd.in/gwNWwkNh?). Great introduction, though I find it lacking as it did not include factory operations, supply chain, store operations that rely far more on offline-online interfaces. For any operator out there, process, analytics, metrics, cadence, solutioning and execution are the key concepts to apply.

1

2 Comments

Sharing some thoughts on the reliability of AI-powered solutions and testing approaches: https://lnkd.in/eYsEzTAP

5

We've been making some cool examples of things that can be built with substrate.run lately. Here's a simple chatbot that can help you cook: https://lnkd.in/eEZm-nRJ If you're working on building a chatbot, take a look and see what it might look like to use Substrate 🧑‍🍳

17

2 Comments

Protip: If you use a 3rd-party API in your app, check in frequently. APIs change. One of my favorite projects, my resume map (https://lnkd.in/g8zbx_H), was based on Mapbox's Javascript library and used Stamen's toner map. Stamen decided to shift their tile service to Stadia Maps. Mapbox decided to update its library and rebrand it as MapLibre. Ever break all the things? So, after a few quick updates, it's back online, just as pretty as it ever was. #map #maplibre #stamen #stadia #gis

9

2 Comments

AI is useful, but not as much as it's being hyped. I typically use AI for explainers. Tell me how something works, with references/sources. Writing code with AI is a waste of time. The code is either wrong or it tends to bloat my code (too verbose or repetitive). https://lnkd.in/gDmQNwnk

6

2 Comments

It’s true. We now live in an era where being “Industrial Age-normal” is now considered extraordinary. In a world where anyone can do almost anything at any time, being predictably reliable becomes a superpower. Rather than ‘gramming about how productive you’re being, you actually shipped what you said you’d ship. Rather than retweeting that pic of some guru’s 4:30am watch display, you actually got your @$$ out of bed like you said you would & walked around the block, before coffee. Rather than promising anything or being prompted, you were just proactive. Now people think you’re made of magic. Basically, everything I’ve ever known or been told about being a decent human being is so true now, I could build a multi-million dollar consultancy on it. This is how you know the world is ending. (j/k but kinda not)

1