Adam Cimarosti

Just trying out the newly released GOT (General OCR Theory) on Hugging Face Does anyone know any trick to improve the output quality? I just used the code on HuggingFace https://lnkd.in/eacHpGuX

3

1 Comment

Several years ago at Form3, I spent 18 months performance testing a payments processing system to try to scale the system to very high throughput (Mark Howard you may remember this! 🙂 ) In the process of doing so, I realised that often we were following a methodology similar to the scientific method: we were devising hypotheses about the possible causes of problems, and then testing them. After many years, I've finally gotten round to writing up my thoughts on the idea of hypothesis-driven debugging, and--in particular--the importance of using falsifiable hypotheses. If you'd like to find out more, take a look at my blog post here: https://lnkd.in/eH7Vv5-p

55

6 Comments

The 2024 StackOverflow survey is out, and for the 8th year in a row, Clojure has ranked near the podium in the Top Paying Technologies category—often occupying rank 1. I don't talk much about why I choose to use a weird Lisp—mostly because nobody cares—but the results speak for themselves. If you can hold your nose long enough to learn something very different, you'll find yourself in a high-productivity environment that features: 🔥 Hot code reloading 🦆 Dynamic typing ♻️ Best-in-class garbage collection 📂 Highly performant immutable data structures 📚 Access to the entire JVM's ecosystem of libraries 🧩 Cross-platform compilation to JavaScript 🧠 An unbelievably smart and humble community Sure, you'll pay a high mental price to learn the language, but as a reward, employers will pay a high price for *you*.

72

47 Comments

Seeing is believing, so here's how to demo Speculation Rules API to colleagues, bosses, clients, friends and family. All client side, 100% console. I aimed to show: 1) how the browser behaves 2) how simple it is to implement 3) new devtools toys https://lnkd.in/gvJ9KQxw (this is my first Frontend Masters blog post, thanks to Chris Coyier's editorial support!)

22

1 Comment

Have you used Cursor? How I use Cursor to fine tune my first LLM https://lnkd.in/eGFKcT8d

18

Opinion: Good prompt engineers need NOT be linguists - they need to be critical thinkers who anticipate different scenarios and craft conditions in their prompts accordingly. Grammar or spelling errors in prompts may not break them, but overlooking key edge cases could. For example, a prompt asking "Is this text about stock price movement?" might catch explicit statements like, "The stock rose 5% today", but miss somewhat implicit ones like, "Investors reacted positively to the earnings report". Worse, it might incorrectly flag unrelated statements as relevant such as, "The company saw a significant rise in customer satisfaction this quarter". Challenge lies in refining prompts to capture nuances while avoiding false positives.

14

It seems that with C/C++ -> Rust, specifically, you could do this by comparing the output compiled code as one of your verifications of correctness for ML learning This is something that e.g. copilot doesn't do - but it'd be straightforward to do so when doing training The expectation would be that the compiled rust code should be equivalent to the compiled C code, i.e. the assembly should be semantically no different As the ML learns, it may end up doing some funky rust code simply to get that to match - but that's probably a good thing, as this is the part that would require some manual intervention & you'd at least be guaranteed equivalent behavior

1

The best article on #llm #tokenization I've encountered in probably the last year.

15

If you want to see what #Clojure brings to the table, you should follow this series by Kapil Reddy. My firm belief, when it comes to programming languages, is to learn those which add a new perspective and change the way you think about code. Clojure most definitely does that. What's your list of languages to learn which change the way you think and make you better? Share in the comments!

13

In this article we look at how we can expand on probabilistic procedures based on Reed Solomon fingerprinting for the purposes of reducing time complexity of certain algorithms. #zkp #zeroknowledge #rust

16

1 Comment

📰Introducing Miniflux: A Minimalist Open-Source News Reader🚀 Ever felt overwhelmed by information overload while managing multiple news feeds? Allow me to introduce you to Miniflux - an opinionated, lightweight feed reader that's going to change your life! 😊 Miniflux is written in Go and PostgreSQL, designed for simplicity, speed, and ease-of-use. Its minimalist approach avoids unnecessary features, focusing on the core functionalities of feed management. Why go for Miniflux? - Zero dependencies: Standalone binary ensures hassle-free installation 💻 - Single binary compiled statically without dependency - No ORM or complicated framework: Clean, efficient codebase for developers 🧑‍💻 - Modern Javascript (ES6 and Fetch API): Smooth user experience 🎯 - Extensive documentation: A beginner's best friend 📚 Personally, I've fallen in love with its simplicity. With Miniflux, it's now effortless to browse through different sources and enjoy the news without distraction. I am sure you will find it incredibly useful for your projects too! 💡 Now that you know about Miniflux, I'd love to hear your thoughts on the tool. Have you tried it already? How has it streamlined your workflow? Let's discuss the beauty of minimalism in software development! Feel free to share your feedback or ask questions about Miniflux here 👇 #Miniflux #FeedReaders #OpenSource #GoLang #TechCommunity

5

 LLM's Llama/Mistral take context length ✨✨from 8k to 2 million!✨✨ That's 256x longer context! Moving from initial BERT's sin/cos positional encoding to RoPE/LongRoPE. https://lnkd.in/gdJ7Hubn #AI #LLM #NLP #LongRoPE #RoPE

10

1 Comment

Arg! After thousands of lines of code cannibalizing the LLVM support library I’ve run into a fundamental design flaw in my approach. Too technical to talk about. Let’s just say I’ve created a set of cyclic dependencies where a depends on b which depends on c which depends on a. It’s because I made a fundamental change to how the library works (which makes it much easier to use when it has to work on different platforms like MacOS and iPads, and is the main reason I went down this path rather than trying to use the library as is) while trying to stick very close to the original structure and design. Not gonna fly. So either I throw out all that work and give up the ease of use - not happening - or I step back and rethink the big picture design elements. And despite the frustration I am loving digging into this library. Usually. By the way this is pretty much what programming is all about - running into obstacles and figuring out how to overcome them. We call it Software Architecture, but it’s not at all like building cathedrals.

6

5 Comments

Does #OpenSource have a future? Absolutely! This is a rare occasion of Betteridge's Law being wrong. I recently spoke with James Governor of RedMonk about the current open source landscape, recent fears about licensing switches and how Fastly is committed to supporting various open source projects such Scratch, RubyGems, cUrl, and the Linux Foundation. Watch the video and let me know what you think: https://lnkd.in/gsVJWC_p

43

I’m skeptical about Bluesky debunking X in the long term. Two reasons: First, the main motivation for the recent so-called exodus is political, not product-based. It’s very hard to debunk a long-established product with a clone. Second, I’m skeptical that a social network that is both ideologically charged and philosophically good with censorship will result in anything mainstream. The community-driven block lists or the unidirectional “intolerance” tag they apply discretionarily will produce the hell of an echo chamber, like Mastodon is today or like Twitter was not so long ago. This is fine; it is just that the traction an echo chamber can gain in 2025 is capped; this is not 2020 anymore. I’m bad at predictions, and I can’t say I care terribly about the outcome here. I'll leave this written to revisit in a few years.

33

17 Comments

At Tonk Labs we are focussed squarely on helping you build trust with anyone, about anything, on the web. Usually we attack that problem with cryptography: to disintermediate faulty, centralised trust infrastructure. But new tools like the Perspective API from Jigsaw that use AI to detect toxicity are another fruitful approach! https://lnkd.in/eCPKQHKa

10

Here's Chapter 6 of LazyVim for Ambitious Developers, available publicly! https://lnkd.in/eiqQFjyd Neovim 0.10 dropped this week, which resulted in a ton of changes to LazyVim (mostly removing plugins in favour of new native features). LazyVim released version 11 yesterday. This resulted in a marathon editing session in which I updated every released chapter to suit LazyVim's new reality. In addition, Chapter 8 has been edited down and is available on my Patreon.

2

Sad that https://lnkd.in/eAnUjq-i hasn't taken off more -- such a powerful approach. So much energy has gone into the JSONSchema-ification of LLM outputs though, so much less applicable but so much more approachable so it kinda makes sense. For example: https://lnkd.in/e6pW_FNS -- heaven forbid you want an LLM to return both the code and filename to put the code in, you need to take a quality hit to use a structured approach or pray it comes out kinda right if you don't. AICI resolves that super elegantly

5

The teams at BigBox VR and Toast VR recently released updates to Population:ONE and Max Mustard that utilise Meta Haptics Studio and Haptics SDK as their unified tool chain for cross platform haptics development. You can read more about it in our latest blog post. As an added bonus you can directly try out and learn from the Population:ONE haptics in Haptics Studio - we shipped a subset of them as a new example pack in 1.3! https://lnkd.in/ey3tJwJ3

29

3 Comments