Game of Thrones Daily
titsay
hello vonnie
Kaledo Art
Xuebing Du
tannertan36
Sweet Seals For You, Always
pixel skylines
styofa doing anything
Jules of Nature
todays bird
shark vs the universe
Lint Roller? I Barely Know Her
Show & Tell
Claire Keane
❣ Chile in a Photography ❣
No title available
dirt enthusiast
sheepfilms
Misplaced Lens Cap
seen from Malaysia
seen from United States
seen from China
seen from Malaysia
seen from United States
seen from T1
seen from United States
seen from United States
seen from United States
seen from United States
seen from United States
seen from United States
seen from Türkiye
seen from United States
seen from United States
seen from United States
seen from United States
seen from United States
seen from United States
seen from India
@holyjak
Neuro-Symbolic Artificial Intelligence: Towards Improving the Reasoning Abilities of Large Language Models
neuro-symbolic reasoning aims to combine LLM's ability to learn from data with established practices of more human-like symbolic reasoning.
We're adopting Rust as our C++ successor language, and using AI agents to accelerate the transition.
A story of using AI to help translating the WIP Ladybird browser from c++ to Rust. Starting with the JS engine:
The result was about 25,000 lines of Rust, and the entire port took about two weeks. The same work would have taken me multiple months to do by hand.
The limits of Rust, or why you should probably not follow Amazon, Cloudflare and Discord
The latest Intel processors, Core Ultra Series 3 a.k.a. Panther Lake, look awesome. They are fast in benchmarks, can consume very little pow
Some good reasons to not use rust (anemic standard library & very fragmented and constantly changing library landscape, language churn,...), and where it shines.
One of the problems we had to solve when we added multiplayer editing to Figma was supporting simultaneous editing of ordered sequences of o
How Figma solved simultaneous editing of ordered sequences of objects. Instead of Operational Transformation, but "the high implementation cost is only worth it you need the specific benefits it offers for editing text sequences." and "OT was overkill for us because we didn’t need to work with enormous sequences and we didn’t need to avoid interleaving." Instead, they use Fractional indexing - every object has a real number as an index. To insert between two objects, just set the index for the new object to the average index of the two objects on either side.
Claude Spend. Contribute to writetoaniketparihar-collab/claude-spend development by creating an account on GitHub.
This *may* be useful - a tool reading Claude Code session files and displaying a dashboard showing what you spent your credits on. "Surfaces insights like which prompts cost the most and usage patterns"
Datomic knowledge graph for your codebase. Immutable facts, Datalog queries, iterative LLM reasoning. Runs locally — your code stays yours.
Noumenon builds a Datomic knowledge graph from your repository so agents query structured facts instead of raw context windows. It includes data such as git history, file structure, authorship, change frequency, and cross-file import graphs. Explore complexity ratings, safety concerns, architecture, etc.
Aside: A fascinating story of using an autonomous AI agent to iteratively improve the system by proposing a change (to the code, prompt, examples), re-running benchmarks, and keeping or discarding the change. Key points here. (Inspired by autoresearch .)
by cgrand (🦋 🦣 𝕏) There is a modeling technique I’ve used in the past, and the more I use it, the more it feels like the right default in a
When you model with data, you may run into a situation where plain values are no longer enough. Not because you need more structure, but because you need more distinctions, more equivalence classes. People try to solve that with :type tags, defrecord, or metadata. They all have problems.
Another possibility is to (mis)use (tagged-literal a-name {...}) which gives you both the :tag and the nested map as :form. Why?
If a plain 42 and a "user id 42" should not be interchangeable, then they should not be equal, not be confused, and not accidentally flow through the same code paths. This is what tagged values give you: not more structure, but stronger distinction to prevent unknowingly sending specific data through generic paths and its counterpoint avoiding to make specific pipelines accidentally generic.
NanoClaw is a secure, lightweight alternative to OpenClaw. Your personal AI agent that runs in containers, built to be understood and custom
"the secure personal AI agent" in 15 source code files, a single Node.js process running in a container. Inspired by OpenClaw but tiny. You can add new skills to expand its capabilities, e.g. so that it can talk to Telegram. Uses Claude in the background. You can talk to it via a messaging app, it can run scheduled tasks, access the web, etc.
I love simple, boring and reliable tools. In the software world, the two best are without a doubt Rust and PostgreSQL. One example: a backen
A useful article. Tips: Use sqlx instead of ORM. How to do batching inserts and updates. Leader election with PostgreSQL's advisory locks. Replace redis with unlogged tables. Using PostgreSQL as a job queue.
Also, this experience is pretty impressive:
Before, the service was written in Go completed batch jobs in ~30 minutes, and needed to run on instances with at least 4 GB of RAM. After being rewritten in Rust and configuring a high-performance memory allocator, it's now completing batch jobs in less than 5 minutes and run on a single instance with 512 MB of RAM.
How Rust and Its Compiler Have Revolutionized Software Engineering and Reliability
A lot of reasonable people may perceive my enthusiasm for Rust as misguided fanaticism, but it isn't. It's cold pragmatism. Sherlock
"... when you have eliminated all the other languages due to their shortcomings, Rust is the last one standing, so today we are going to study these shortcomings, and how Rust solves them. I hope to convince you that Rust is probably the right choice for your next project or to invest time in learning it to level up your career."
You may disagree but the are certainly some good arguments here.
Deploying Rust to production checklist
While a lot of time is spent on design patterns and low-level tricks such as SIMD accelerations, I'm suprised that very few resources ar
A great checklist for ensuring your production Rust service are small, fast and secure.
Dr. Eliyahu M. Goldratt, in his lecture titled Beyond the Goal: Theory of Constraints, describes the necessity of answering the following…
Dr. Goldratt (of The Goal) tells us to ask 4 question of any new technology:
What is the power of the technology?
What limitation does it diminish?
What rules helped us accommodate the limitation?
What rules should we use now?
The author demonstrates how they applied these questions to their transition from a Node framework to PHP's Laravel. He closes with the insight that "Rather, we must additionally map the technology back to behavior and make a determination based on the full context." Which makes perfect sense, and is something often forgotten.
AWS European Sovereign Cloud
Built, operated, controlled and secured in Europe. With zero operational control outside of EU borders, only AWS employees residing in the EU will control day-to-day operations, including access to data centres, technical support and customer service for the AWS European Sovereign Cloud.
GitButler software development platform
I've switched to using GitButler for my interactions with git, mostly because it allows me to easily spin out separate, small PRs, while keeping all the code present in my local development workspace.
Technically, it gives you a new branch, gitbutler/workspace, and automatically merges into it all the local branches you want. It supports updating (e.g. rebasing) all those branches, resolving conflicts, creating PRs etc.
Typically, I work on a problem, and discover somewhat independent improvements, refactorings, or new functionality that I need but which I'd rather publish as separate, small PRs. With GitButler it is trivial to commit these to separate branches, push them, and get them eventually approved, while having all the code merged and available in my "workspace" so that I can continue working on my problem, without having to switch out of it and/or wait for PR approvals and merges.
Folks’ different perspectives can seem very alien to each other.
Bob Marshall explains how most organisations are much less effective than they believe themselves to be, and much less effective than they could be, and describes the four different mindset along the axis of increasing effectiveness: Adhoc, Analytic, Synergistic and Chaordic.
Don't use flatten unless you really know what you're doing - it indiscriminately flattens all nested sequences. Use e.g. (apply concat) (or a transducer version of that) instead.
Trick Gandalf into revealing information and experience the limitations of large language models firsthand.
Test your LLM hacking skills in this game, where you are trying to trick increasingly more secured LLM to reveal a secret. Or notch it up and try to fool an agent to do you bidding by injection information into a website it is reading.