Hacker News Reader

AmazingTurtleon Launch HN: Slashy (YC S25) – AI that connects to apps and does tasks

Why __don't__ you use MCP?

AmazingTurtleon A Brilliant and Nearby One-off Fast Radio Burst Localized to 13 pc Precision

You should have said "it hit my head like a zip bomb", it was unclear to me what you meant till I read the whole paragraph

AmazingTurtleon Launch HN: Channel3 (YC S25) – A database of every product on the internet

If a quick POC is useful, we (everfind.ai) can help fill product-data gaps while honoring your "compliant vendors only" rule. We ingest whatever you have (text, JSON, HTML, Excel), map it into category-specific schemas, and surface clean fields for search/filter. You also get fast fielded search, sensible relevance, hybrid vector search, and an optional guided-selling assistant you can expose publicly. Happy to try this on a small sample.

Feel free to reach out to me at felix.faust@everfind.ai :)

AmazingTurtleon Launch HN: Channel3 (YC S25) – A database of every product on the internet

Searching for Ryzen 9950X3D yields 6 different AMD processors before I see a fully built PC with that Ryzen 9950X3D included. The actual product I was looking for was ranked 16th.

AmazingTurtleon Copilot broke audit logs, but Microsoft won't tell customers

No need for a per-user database, simply attach ACL to your vector DB (in my case I am use postgres, RLS for example or a baked ACL policy list if you're using opensearch for example)

AmazingTurtleon Prime Number Grid

https://susam.net/primegrid.html#1-200-330

zoom out, then play around with cols +/-1 and observe the pattern change. I observe the pattern from -7 to +5; same on #1-200-420

AmazingTurtleon Llama-Scan: Convert PDFs to Text W Local LLMs

Yet another Prompt Wrapper

TRANSCRIPTION_PROMPT = """Task: Transcribe the page from the provided book image.

- Reproduce the text exactly as it appears, without adding or omitting anything. - Use Markdown syntax to preserve the original formatting (e.g., headings, bold, italics, lists). - Do not include triple backticks (```) or any other code block markers in your response, unless the page contains code. - Do not include any headers or footers (for example, page numbers). - If the page contains an image, or a diagram, describe it in detail. Enclose the description in an <image> tag. For example:

<image> This is an image of a cat. </image>

"""

AmazingTurtleon GTP Blind Voting: GPT-5 vs. 4o

12/8 gpt-5/gpt-4o

AmazingTurtleon Curious about the training data of OpenAI's new GPT-OSS models? I was too

He presented an empty prompt to gpt OSS and let it run many times. Through temperature, the results vary quite a lot. He sampled the results.

Feeding an empty prompt to a model can be quite revealing on what data it was trained on

AmazingTurtleon Open models by OpenAI

Im not sure if groq uses the proper harmony template?

AmazingTurtleon My favorite use-case for AI is writing logs

came here to say this. I don't want to wrap my head around writing pretty logs and involving all the variables I want to eventually see in my logs. I just want to "dump it" and keep writing main logic

AmazingTurtleon I extracted the safety filters from Apple Intelligence models

"driving with Focus turned on"

https://github.com/BlueFalconHD/apple_generative_model_safet...

AmazingTurtleon Open source can't coordinate?

Whats wrong with LSP?

AmazingTurtleon Gemini-2.5-pro-preview-06-05

for bulk data extraction on personal real life data I experienced that even gpt-4o-mini outperforms latest gemini models in both quality and cost. i would use reasoning models but their json schema response is different from the non-reasonig models, as in: they can not deal with union types for optional fields when using strict schemas... anyway.

idk whats the hype about gemini, it's really not that good imho

AmazingTurtleon Quarkdown: A modern Markdown-based typesetting system

Markdown, Quardown, Typst, this and that. This looks to LaTeX. All of this became so confusing and un-standardized - I went back to HTML+CSS.

AmazingTurtleon Compiling a Neural Net to C for a 1,744× speedup

I recently read about DLGAs on HN and instantly thought: damn thats some hot take. But I was too stupid to implement it from the paper. Glad you got it working and documented it! Thanks!

AmazingTurtleon xAI to pay telegram $300M to integrate Grok into the chat app

Thanks for this comment, came here to say the same thing.

AmazingTurtleon I am disappointed in the AI discourse

Well, technically, GPT "itself" really isn't a search engine. The LLM was trained with capabilities to communicate to the runtime that I can request additional information for fulfilling the request, such as performing a web request - which is performed by the runtime and the results were fed back into the LLM as a regular prompt. Thus GPT only answers based on the textual context - again: involving statistical likeliehood.

AmazingTurtleon You Don't Need Re-Ranking: Understanding the Superlinked Vector Layer

At everfind.ai, we've found a middle ground that leverages both structured and unstructured data effectively in retrieval systems. We utilize a linear OpenSearch index for chunked information but complement this by capturing structured metadata during ingestion—either via integrations or through schema extraction using LLMs. This structured metadata allows us to take full advantage of OpenSearch's field-type capabilities.

At retrieval time, our approach involves a broad "prefetching" step: we quickly identify the most relevant schemas, perform targeted vector searches within these schemas, and then rerank the top results using the LLM before agentic reasoning and execution. The LLM is provided with carefully pre-selected tools and fields, empowering it to dive deeper into prefetched results or explore alternate queries dynamically. This method significantly boosts RAG pipeline performance, ensuring both speed and relevance.

Additionally, by limiting visibility of the "agentic execution context" to just the current operation span and collapsing it in subsequent interactions, we keep context sizes manageable, further enhancing responsiveness and scalability.

AmazingTurtleon The recently lost file upload feature in the Nextcloud app for Android

We feel your pain at Nextcloud. Our team at Everfind (unified search across Drive, OneDrive, Dropbox, etc.) has spent the past year fighting for the *drive.readonly* scope simply so we can download files, run OCR, and index their full-text for users. Google keeps telling us to make do with *drive.file* + *drive.metadata.readonly*, which breaks continuous discovery and cripples search results for any new or updated document.

Bottom line: Googles "least-privilege" rhetoric sounds noble, but in practice it gives Big Tech first-party apps privileged access while forcing independent vendors to ship half-working products - or get kicked out of the Play Store. The result is users lose features and choices, and small devs burn countless hours arguing with a copy-paste policy bot.

AmazingTurtleon xAI dev leaks API key for private SpaceX, Tesla LLMs

Guess who's going to be fired by elon :D

AmazingTurtleon How to write blog posts that developers read

Ironically I opened the post, read the headlines, deemed it uninteresting and went along with other HN headlines.

edit: I think I read HN comments more than HN articles. Interesting

AmazingTurtleon Mlx-community/OLMo-2-0325-32B-Instruct-4bit

What if I repost your comment without your permission?

AmazingTurtleon Ask HN: With trust in Firefox gone, is Chrome-ish the only option?

I like chrome, I don't like google. I mean: I am so used to chrome devtools, I don't even know how to switch to firefox. I gave it a try and it was frustrating that I didn't know where to find my shit. Felt like 10x performance loss for my web development activities. Is there any way to make firefox devtools look and feel like chromes?

AmazingTurtleon Natural occurring molecule rivals Ozempic in weight loss, sidesteps side effects

Nestle is making billions of dollars -- off water.

AmazingTurtleon 400 reasons to not use Microsoft Azure

take my angry upvote

AmazingTurtleon The Fall of FiveM

I know the guy who leaked the alteriwnet source code back in the days. Thats some nice code right there

AmazingTurtleon Show HN: Finding your documents is hard

As one of the founders of everfind.ai, I'm thrilled to see the community engaging with our mission to simplify document search and retrieval. I'm curious—what's the biggest challenge you face when managing your documents, and which integrations or features would transform your workflow?

AmazingTurtleon Show HN: Watch 3 AIs compete in real-time stock trading

Combining universal time-series prediction models with latent space global knowledge on realtime information could result in an accurate model prediction on the stockmarket with a bias towards succeeding. https://research.google/blog/a-decoder-only-foundation-model...

AmazingTurtleon Show HN: Open-sourcing my failed startup Buzee – A file search application

Great initiative! It's always inspiring to see individuals open-sourcing their projects to benefit the community.

At refind.ai (https://refind.ai/), we’ve been working on a similar challenge of navigating large volumes of documents, but with a more AI-driven and user centric approach. Beyond just search, we focus on transforming unstructured data into structured insights. This includes features like automatic metadata extraction, natural language search, and integrations with external systems like email and CRMs.

To those who struggle with file management at scale, especially in environments where tools like Spotlight or Windows Search fall short, I think there’s a lot of potential for tools like Buzee to evolve further. If anyone wants to collaborate or learn more about our approach, feel free to reach out!

Best of luck with Buzee’s journey—I’d love to see how the community builds on it!