Hacker News Reader

iAkashPaulon More annoying macOS 15 Sequoia prompts: Bluetooth

IMO hitting CMD+T thinking it'll open a new tab in my browser but it opens the font window (from Mail) is a user-crime.

iAkashPaulon The Cheating Device (ChatGPT on a TI-84) [video]

A USB MITM board with ESP32 could just connect to your phone & do one-way code/content creation for otherwise software locked devices.

iAkashPaulon Set Up a $4/Mo Hetzner VM to Skip the Serverless Tax

Scaling/cold-boot time is the real concern IMO no the cost

iAkashPaulon Benchmarks show even an old Nvidia RTX 3090 is enough to serve LLMs to thousands

Pretty sure this was never questioned for batched requests, sg-lang/lmdeploy/tensorRT-LLM will have nearly twice as reported speeds with INT8 (fp16 A100 benched here https://github.com/sgl-project/sglang?tab=readme-ov-file#ben...)

iAkashPaulon Ask HN: Visualize Software Architecture/Concepts

Apart from draw.io I like to use isoflow.io for near isometric diagrams with directional flows as well

iAkashPaulon Show HN: Pie Menu – a radial menu for macOS

Hey Marius, cool project! Would this work with just the middle mouse button click too?

iAkashPaulon Taking command of the Context Menu in macOS

More than this pressing spacebar should have gotten some customisation by now, eg. certain mp4s/markdowns/any other file-extension for quick previews

iAkashPaulon Intel N100 Radxa X4 First Thoughts

Can't wait to get this & have TrueNAS+Jellyfin up & running on it, certainly beats having to buy so many addons for the Pi & still not have a super reliable build.

iAkashPaulon Ask HN: How is your organization using LLMs?

Where did you find OpenAI in my reply?

iAkashPaulon Glacier is an open-source web browser with extensive customization capabilities

Thx will check it out

iAkashPaulon Solving the out-of-context chunk problem for RAG

One quick way to improve results greatly is to ask questions with 2/3 chunks & in the lookup for these chunks mention the IDs of the other chunks, qdrant allows for easy metadata addition. So just generate a synthetic question bank & then do vSearch against the same instead of hoping for the chunks to match up with user questions.

iAkashPaulon Glacier is an open-source web browser with extensive customization capabilities

What I wished browsers on Android would do is scroll pages with volume buttons and close pages on double tap or some combination similar to screenshot ring.

iAkashPaulon Ask HN: How is your organization using LLMs?

Mostly Mistral/Mixtral models in addition to Whisper.

iAkashPaulon Ask HN: How is your organization using LLMs?

Heavily,

Call summarisation at scale, for further insights of key topics/sentiments for diarized & fused channel call data

Lead mining quality improvement by asking questions instead of merely classification

Internal simple RAG chatbot across ~1000 PDFs, not yet down the GraphRAG route

Better chatbot experience over RASA

Contract mining for speeding up mundane governance tasks

iAkashPaulon Benchmarking LLM Inference Back Ends: VLLM, LMDeploy, MLC-LLM, TensorRT-LLM, TGI

Nice, I'll bench server.cpp as well

iAkashPaulon The Framework Laptop 13 is about to become one of the first RISC-V laptops

How long until someone makes a harvested Apple Silicon logic board work with this!

iAkashPaulon Every Way to Get Structured Output from LLMs

My point exactly, lot of brigading on this thread

iAkashPaulon Every Way to Get Structured Output from LLMs

JM2C

1. Langchain not being used in production?

> How out of touch is that remark? Hard pressed to find agentic framework implementation outside of Langchain/Llamaindex.

2. Outlines is not expected to work with OpenAI API because it wasn't created to do that.

iAkashPaulon Syzygy: An open standard for low cost, compact and high-performance peripherals

I remember watching Andy Barry/MIT demo of Pushbroom stereo approach on a flying-wing drone, it used stereo 120FPS cameras to dodge trees & branches. It's almost a decade later now & such hardware is still not cheaply available for Pi's/SBCs.

iAkashPaulon Japan enacts law to promote competition in smartphone app stores

Should have emboldened the ecosystem by providing incentives for web-based government services instead of forcing users to only depend on two primary entities in this space.

iAkashPaulon AMD's MI300X Outperforms Nvidia's H100 for LLM Inference

INT8/FP8 benchmarks would've been great, both cards could have loaded them with around 60GB VRAM instead of TP=2 on H100.

iAkashPaulon 38% of webpages that existed in 2013 are no longer accessible a decade later

Not being able to renew domains beyond 10yrs could also be a factor

iAkashPaulon MicroSD cards' SBC days are numbered

Yeah heard about these in the latest Geerling video

iAkashPaulon Project Gameface launches on Android

Off the bat I can see on-device assistants be more capable & allow customized handsfree browser/PDF interaction

iAkashPaulon MicroSD cards' SBC days are numbered

Samsung & Transcend SD/mSD have always errored/failed on me with power outages on Pi without additional power management hardware & software.

Having push to eject NVMEs brings forth a future where you can swap out LLMs/brains in K2SO type droid

iAkashPaulon Apple to Power AI Features with M2 Ultra Servers

Summarisation & SDXL run just fine on M1 & above Macs. Probably the video editing features like Google/Pixel might need cloud compute.

iAkashPaulon OpenAI: Model Spec

Right-clicking to inspect element ain't gonna make it

iAkashPaulon Raspberry Pi Sign Language Spectacles

Realtime translation of text in signage & other things is going to be one of the killer apps for pass-thru AR!