Hacker News

We provided our harness with a single objective - 2048 with a twist - and 2 days later it delivered.

https://github.com/awesoftsolutions/idea_meta-muse-2048-game

The run is recorded end to end, so the code can be checked against the process that produced it. Our app is drivable via replay here: https://favur.dev/go/drive/the2048

Comments URL: https://news.ycombinator.com/item?id=48979593

Points: 1

# Comments: 0

Article URL: https://elonlit.com/scrivings/your-transformer-is-secretly-an-eot-solver/

Comments URL: https://news.ycombinator.com/item?id=48979555

Points: 1

# Comments: 0

Article URL: https://github.com/bonushora/surgical-dev-ops/blob/main/README_EN.md

Comments URL: https://news.ycombinator.com/item?id=48979553

Points: 1

# Comments: 0

Article URL: https://unstack.io/what-is-ai-good-at

Comments URL: https://news.ycombinator.com/item?id=48979551

Points: 1

# Comments: 0

A functional programming language, designed and optimized especially for cross-platform numerical and geometry computation. I'd like to know if there are other domains relevant to you that this could be useful for: what is missing, or could be improved? Let's discuss thanks!

Comments URL: https://news.ycombinator.com/item?id=48979540

Points: 1

# Comments: 0

TL;DR version: I made a macOS journaling app that use Foundation Models to analyze entries for mood + written themes, and also has voice transcription, because I couldn't find any other journaling app that does those things.

OK, now for the longer, I-rambled-too-much version:

For the past few weeks I've been working on a macOS journaling app called VoxThermic.

The idea for it came from my own journaling habits, where my mood tends to shift a lot depending on whatever I'm dealing with. I've tried some journaling apps, but I've never found one that can analyze mood or see how my mood changes depending on what I'm writing about.

So I just figured I'd build it myself. I've been experimenting a lot with Apple's Foundation Models and Natural Language frameworks, and those all work on-device and offline (which is good because there's no way I'm sending journal entries to some AI company's data centers!)

But there's one problem with using Foundation Models: they're tiny. About 3 billion parameters, with a context window of 4096 tokens. That's fine for analyzing ONE journal entry. But how would I get it to analyze multiple entries?

It took a lot of experiments and the Foundation Model just outright refusing to work for several of them. Turns out it just refuses to work if you send in too much data... who would have guessed? Eventually I came up with one solution: summarizing multiple entries and then passing those into the model. Sometimes the summaries get summarized too, depending on the amount of data to process.

The approach works, but summarization loses detail and nuance. The analysis for a single journal entry can have a lot of detail and depth, because there's no summarizing involved and thus no data loss. The summary for a year's worth of entries... that's not so detailed. I still haven't found a better approach that stays entirely on-device.

Oh yeah, on an unrelated note: I also like to ramble a lot (you could have guessed, huh?), but I like to ramble out loud, rather than type. So I made voice transcription a big feature of the app, though this was the easy part.

Anyway, everything here works on-device, and it can run offline. And I'm aware this is probably a niche app, but it's something I wanted for myself, and I hope it comes in handy for someone else too. The app is free to download and use, with optional features like AI summaries and PDF exports.

Lastly, I also wonder if anyone's figured a better way of doing long-term analysis with Foundation Models? My "summarize the data and pass in a collection of summaries" technique works, but I don't think it's elegant, and yet at the same time I haven't been able to figure out anything better.

Comments URL: https://news.ycombinator.com/item?id=48979539

Points: 1

# Comments: 0

Article URL: https://iretq.com/inside-getofflinedeviceuniqueid-how-windows-derives-its-offline-device-id/

Comments URL: https://news.ycombinator.com/item?id=48979508

Points: 1

# Comments: 0

Article URL: https://4playerchess.plus

Comments URL: https://news.ycombinator.com/item?id=48979496

Points: 1

# Comments: 1

Article URL: https://magazine.sebastianraschka.com/p/controlling-reasoning-effort-in-llms

Comments URL: https://news.ycombinator.com/item?id=48979475

Points: 2

# Comments: 0

I am a frontend engineer and actively use AI. I caught myself that i never write code manually, only if i have to tweak some styling. I feel lazy and dumb compared to pre-AI era. Even if i have to write some small changes in code that take like 2 minutes - it is easier to spend those 2 minutes explaining everything in details to AI and point out what and how it needs to change. Because i dont have to think about the process, i only explain the idea and desired solution - then wait until its done to review.

Thats some kind of degradation. But the thing is that the quality is amazing. My productivity and speed increased significantly.

But there is a frustrating thing. The laziness started affect even review process. I feel like i dont want to spend time to read and check all the changes - i will better ask another AI to do it. And it works! I use Claude for like 95% of work, then switch to GPT to review everything properly. They have different personalities so you have different points of view. I consider GPT as a nerd - it catches a lot of gaps and inconsistencies that Claude misses. So i like to close them both in one room to debate.

What i really do care a lot is a comprehensive documentation. I spend a huge amount of time for planning and brainstorming with AI, then i document everything properly. All those MD files about architecture, composition recipes, cheat sheet… All of it should lie on surface, not to be buried somewhere deep in code. Every cold session starts with all the docs where everything is explained properly with references and examples to prevent AI to skim or give lazy answers. When i need to implement some big feature i structure a proper ready-to-execute plan so every next agent with fresh context reads it - analyzes what was done - figures out what is the task for this session - and goes ahead with all the changes, then report. It can be dozens of chats working sequentially on some big thing; in the end i will run some workflow to review all the implementation, then ask GPT to audit changes, and i will look myself only when AI approves that everything looks great. Then i can tell what and why has to be redone because i dont like how we have it now.

In my setup the documentation for AI outweighs the code (instruction files on every level, rule files that load by file type, subagents with roles, hooks around every edit). The most weightful part:

- A catalog of silent failures. 36 numbered traps that compile fine but render broken. Every entry is a real failure that shipped at least once: cause, symptom, fix.

- A "polish" rule. When you tell AI "make it better", its reflex is to ADD decorative stuff. So there is a written rule that polish means subtraction, with a list of patterns i rejected in real reviews.

- CI that fails when docs drift from code. The component registry is checked against source files. Because docs that lie are worse than no docs.

- A runtime verifier (WebMCP). An agent drives a real Chrome browser, clicks through the page, watches the console, takes screenshots at 3 viewports.

Basically I stopped writing code and started engineering the environment that writes the code. Every file exists because AI failed at something once, and i wrote it down so it wont happen again.

The biggest problem for me is motivation. I remember that feeling how you got stuck - cant figure out something for hours, searching right solution, trying, and then FUCK YEAH FINALLY, and the satisfaction: YOU NAILED IT. I dont have it anymore. I feel confident even in things where i have no expertise at all. Thats like: meh AI can handle it.

I think my system solves the quality question, but increases the laziness/motivation problem. I made it safe to be lazy.

So two questions:

1. How does your AI setup look like? I mean real files and rules, not just how you prompt.

2. Has anyone actually solved the motivation part?

Comments URL: https://news.ycombinator.com/item?id=48979474

Points: 2

# Comments: 0

Article URL: https://bartoszmilewski.com/2026/07/19/profunctor-optics/

Comments URL: https://news.ycombinator.com/item?id=48979470

Points: 1

# Comments: 0

Article URL: https://theconsensus.dev/p/2026/07/11/of-forks-and-clones-and-package-management.html

Comments URL: https://news.ycombinator.com/item?id=48979469

Points: 1

# Comments: 0

Article URL: https://www.reuters.com/business/world-at-work/asml-offer-employees-20000-retention-bonus-staying-2027-2030-2026-07-20/

Comments URL: https://news.ycombinator.com/item?id=48979461

Points: 1

# Comments: 0

Article URL: https://www.reuters.com/business/healthcare-pharmaceuticals/fda-says-taylor-farms-cyclospora-finding-was-false-positive-2026-07-20/

Comments URL: https://news.ycombinator.com/item?id=48979432

Points: 1

# Comments: 0

Article URL: https://worldholidays.me

Comments URL: https://news.ycombinator.com/item?id=48979428

Points: 1

# Comments: 0

I feel like from day one, LLMs was talked about in relation to how it will change the world, how it's good or bad for business and the environment. But then yesterday, I was out walking with the dog, and it hit me how freaking cool this technology is.

I don't want to talk about where I am on the spectrum between booster and denier. I'm so sick and tired of hearing about it, and it doesn't matter in this context.

I spent the first 8 years of my career trying to make the computer understand reality through images - scanning receipts, finding defects on wind turbine blades, and other niche-but-important things. The amount of work it took to just read a receipt and get the text somewhat right.. It was insane. It was really fun work, but it was so hard to make the real world fit into boxes.

I feel like I completely missed being excited about LLMs - instead I got dragged into discussions about what it means. And it's not because I've lost the ability to care; I can get excited about a new hashing algorithm and other weird tech, but LLMs? it didn't even register as a technology. Maybe it's an uncanny valley kind of thing - where it doesn't look like a classic computer science thing, so my brain didn't register it as such?

During my walk I started thinking of the possibilities. And I know your mind might immediately jump to fully-fledged big-scope solutions, but just try to give it space in that other part of the mind - the part where it's cool and not a tool to change our society. We have no idea what's gonna happen - and that's fine - we've never known that. But what we do know is that we have this really cool technology that you can run on your own computer now.

You can actually literally IRL for real, make a python program yourself that you can talk to and it understands you and do things. All on your own machine. No nothing! IT'S LIKE A TOTALLY NORMAL TECHNOLOGY! It's so cool, I can't really believe it. It can even view images and explain it you, and I just.. I'm so excited now.

Don't let them take the fun out of this technology, and spend a few seconds thinking about that cool little trick that made it possible for us to talk to sand.

Comments URL: https://news.ycombinator.com/item?id=48979417

Points: 1

# Comments: 0

Article URL: https://www.wsj.com/world/europe/stagnant-germany-considers-the-unthinkable-sunday-shopping-13b4651c

Comments URL: https://news.ycombinator.com/item?id=48979413

Points: 1

# Comments: 0

Article URL: https://pyrefly.org/blog/pyrefly-attrs/

Comments URL: https://news.ycombinator.com/item?id=48978616

Points: 1

# Comments: 0

Article URL: https://news.risky.biz/risky-bulletin-hacker-wipes-romanias-entire-land-registry-database/

Comments URL: https://news.ycombinator.com/item?id=48978605

Points: 1

# Comments: 0

Article URL: https://golemui.com/blog/the-age-of-token-efficiency/

Comments URL: https://news.ycombinator.com/item?id=48978587

Points: 1

# Comments: 0