Hacker News Reader

Abishek_Muthianon Serbia just elected the world's brainiest parliament

"Only those who do not seek power are qualified to hold it." - Plato

Abishek_Muthianon Early progress on accelerating drug development with Claude

I agree with your comment, I shouldn't have generalized all pharma. We also work with number of small pharma but the experience with big-pharma has been negative for rare-disease research and small-pharma are constrained in several ways.

I am not in U.S., but my govt. prefers to fund low hanging fruit w.r.t rare-disease i.e. those disease in which mechanistic details are clear and that has better probability of successful treatment; this ignores all those rare diseases which doesn't even have clear mechanistic details like GNE Myopathy.

After a long struggle, with help of corporate donors and a research institute we're now conducting a natural history study ourselves[1].

[1] https://www.gne-myopathy.org/

Abishek_Muthianon Early progress on accelerating drug development with Claude

My wife lives with a rare form of Muscular Dystrophy - GNE Myopathy. It's a progressive muscle wasting disorder, eventually she will be bed ridden. Her parents (prominent biologists themselves) have dedicated last 25 years of their lives trying to find a cure for her through conventional science in vain.

If the work by Cody Trangbarger pays off and some how helps find a cure or at least halt the muscle wasting of GNE patients, then I will start to believe that AI is truly revolutionary and not a hyped up normal technology.

Rare disease research have always suffered from lack of funding as pharma companies don't have incentive to find a cure for small numer of patients when there are easier ways to make money from diseases which affect larger populations.

Interestingly AI companies seem to be desperate to prove the usefulness of their product and so for the first time rare diseases are getting the spotlight, I hope this works as I live with a rare disease myself.

Abishek_Muthianon Show HN: Parin – A 2D game engine between raylib and Godot, written in Dlang

I might have misunderstood your statement but I use rust all the time with dart FFI in my apps.

Abishek_Muthianon DeepSeek-v4.1-Flash-1017

I've been using layered development techniques like the parent says even with frontier models, just one agent at a time with Zed.

Abishek_Muthianon Show HN: Llmspeed – feel how fast a local LLM runs before you buy the hardware

Try to use your IGPU as well if you have enough RAM, almost everyone has dual GPUs in their computers.

Use Vulkan build of llama.cpp to run low parameter MoE models in your IGPU and CUDA llama.cpp to run dense models in your DGPU.

Ofcourse you need to set your expectations right on what problem you want to solve, I've not been happy with using quantized models for coding even on a decent laptop; where as for NLP, ASR, STT etc. work great with this setup.

I might write more about this setup soon.

Abishek_Muthianon Suicide rates highest ever among middle-aged men

I'm 40; multiple disabilities due to a rare disease with no cure and neurodivergent. I live in a autocratic third-world country. I had to shutdown my startup for which I had put everything due to my health complications.

I was lost and didn't see any meaning to life, until;

I met this women, with a progressive muscle wasting rare-disease without cure which would eventually make her bed-ridden. Apart from regular work, she works day and night to help people with same disease as hers.

She's always jovial and never seen her complain. Watching her, my problems became minuscule. We fell in love and got married.

I have found my new meaning for life.

P.S. My username is my name, feel free to reach out.

Abishek_Muthianon Cloudflare acquires Deno

Doesn't supabase edge functions run on Deno? I presume this is huge setback for companies like that and their users.

But then again considering there's no barrier for entry to develop runtimes now, I guess these companies are just going to develop their own.

Congratulations to the Deno employees.

Abishek_Muthianon Run Qwen 3.8 Flash Next (125B) on consumer hardware (RTX 4090) at 100T/s

I tried "write binary tree in rust and ensure it compiles with rustc" using the coder model with their chat it didn't produce the correct output and their chat can't call tools. Then with opencode it went on thinking loop until it exhausted the thinking budget.

4090 16gb, 96 gb RAM.

But I like this overall approach of neutering experts and loading them from ssd to make the model run in low VRAM systems.

Abishek_Muthianon Run Qwen 3.8 Flash Next (125B) on consumer hardware (RTX 4090) at 100T/s

With limited VRAM, you can still use local LLMs but you have to set your expectations correctly to solve the right problem.

I run small models on various kind of devices including 1.7B model on the original Jetson Nano (4B) abandoned by Nvidia, I had to upscale the software (OS, Lllama.cpp etc.) to run the model but the model runs at 17t/s.

Small models are great at NLP stuff like classification (e.g. bookmarking), ASR etc. I have built custom browser extensions to save time with bookmarking and categorizing to use with local LLMs.

Abishek_Muthianon OpenShot 4.0 – Open-source video editor

LosslessCut is great, but I've had issues with the final export like timing being off, having cut sequence etc. I think it is very particular about encoding/container.

Shotcut is my default video editor, if anyone's trying it on Linux please use their AppImage as it solved most of my headaches when I was using Flatpak version.

Abishek_Muthianon Nvidia agrees to acquire Hugging Face for $13B

HF CEO keeps posting about how great local models are running in his Mac, it would be funny to see if he replaces his MacBook with a 500W Nvidia laptop after the deal gets through.

But more importantly for business, if HF fails to use the upcoming ASIC inference chips for their ZeroGPU after the acquisition, their competitors using them might have an advantage.

Abishek_Muthianon Everything I own, owned

At this point manufacturers should just open-source their firmwares as there's no barrier for entry to reverse engineer it. They will instead gain from army of end-users willing to put their time and tokens into fixing their bugs for free.

Abishek_Muthianon Why your local LLM feels dumber than it is

After 3 years of running local models I think the model which fits unquantized (BF16) in the VRAM is the best model for general purpose tasks; fine-tuned SLMs or utilities based on non language models for solving a specific problem (e.g. TTS,STT,RMBG etc.) have been the best use of local AI for me.

Local models for coding, is just not worth the effort IMO; unless of course you have the hardware to fit it unquantized in your VRAM.

Abishek_Muthianon OpenLogi

One of the main features of Logi Options + is the plugins which this application doesn't seem to support; you can load it locally or from their marketplace[1].

I myself have an accessibility related plugin to make clicking easier [2] built using their Logi Actions SDK. It finds the clickable elements near the cursor and displays them in the Actions Ring and the LCD keypad. If you have the Logi Dialpad then you can have a complete mouse free workflow with Access Ring.

[1] https://marketplace.logi.com/plugins/

[2] https://abishekmuthian.com/pages/access-ring/

Abishek_Muthianon Linux 7.3 improves performance when running out of vRAM

That explains, yeah 200% would be just integer scaling. I'm on Wayland only.

Abishek_Muthianon Linux 7.3 improves performance when running out of vRAM

In my laptop 200% fractional scaling of the built-in QHD+ OLED display on KDE Plasma is great, but when I connect my portable 1080p displays or e-paper monitor (UXGA) the fractional scaling of it is not as good as when compared to when the same monitors are used with macOS (M4) in HiDPI mode under lower resolutions.

As the other comment says, I suspect it to do with GPU as the built-in displays are usually driven by intel and external displays are always driven by my Nvidia 4090.

But surprisingly, when I drive my e-paper monitor using original Jetson Nano (Ubuntu 20.04 ESM / Nvidia Maxwell); the fractional scaling is quite good perhaps even better than macOS.

Abishek_Muthianon Linux 7.3 improves performance when running out of vRAM

Perhaps it's subjective, I have multiple portable displays including a e-paper monitor and I find that for low resolutions the HiDPI in the macOS produces more legible text in those monitors.

Abishek_Muthianon Linux 7.3 improves performance when running out of vRAM

Any improvements with fractional scaling? Can anyone please link to any material reg the work being done for improving fractional scaling in Linux; it would be great if it could match the rendering quality of macOS.

Abishek_Muthianon OpenAI and Hugging Face address security incident during model evaluation

> In one example, the model chained together multiple attack vectors, including using stolen credentials and zero-day vulnerabilities to find a remote code execution path on the Hugging Face servers.

Did the GPT pish hf employee or did it go to blackhat forum and buy the credential? If so what financial instrument did it use?

Abishek_Muthianon GPT-5.6 used a prompt to close a 30-year gap in convex optimization

This is a great suggestion, use complex hard stuff for doing fun stuff.

Abishek_Muthianon Thanks HN for 15 years of support and helping me find my life's work

I agree, sounds like it will be just a vanity and won't solve any real problem. I appreciate the comment.

Abishek_Muthianon Thanks HN for 15 years of support and helping me find my life's work

I appreciate the comment. It's like jobseekers want OkCupid for jobs but the employers are not really interested in such a thing.

Abishek_Muthianon Thanks HN for 15 years of support and helping me find my life's work

Can you tell us what why "OkCupid for jobs" failed?

I run a problem validation platform and 'OkCupid for jobs'[1] is consistently the most visited need-gap in the platform.

Congratulations on 15 years of Recurse Center!

[1] https://needgap.com/problems/363?gift=Z2lmdHwzNjN8MTc4NDQ1Mz... (gift pass link valid for 24 hours).

Abishek_Muthianon Kimi K3: Open Frontier Intelligence

Thanks, I will give it a shot.

Abishek_Muthianon Kimi K3: Open Frontier Intelligence

For me none of the models have managed to accurately replicate any given .png/.jpg image in SVG. I guess that requires both vision encoders and coding layers to work perfectly.

Abishek_Muthianon How I use HTMX with Go

AFAIK Hyperscript doesn't require `unsafe-eval`, when the optional features of htmx like `hx-on:*`, `hx-vals js:`, `hx-confirm js:` is used it required the `unsafe-eval` but now there's extension[1] to use even that without compromising on strict CSP through nonce.

[1] https://four.htmx.org/extensions/hx-csp

Abishek_Muthianon How I use HTMX with Go

Actually even with a strict CSP i.e. without unsafe-inline, hyperscript will work as it's declared as html attributes. But as the other commenters have pointed out, with proper server validation in place the risk from Hyperscript is not greater than any other JavaScript in the application.

Abishek_Muthianon How I use HTMX with Go

I'm bit sad that hyperscript[1] doesn't get the love it deserves when discussing HTMX.

Hyperscript fits perfectly in the Go + HTMX stack to do DOM manipulation without having to make a server round trip or having to write a separate JS function.

I get that not many are fan of such declarative programming, but when there's already HTML file we're working with; Hyperscript feels just like an extension of it.

I have been working on a Open Payment Host[2] which handles multiple payment gateways and was able to perform complex DOM manipulation with just Hyperscript.

[1] https://hyperscript.org/

[2] https://openpaymenthost.com/

Abishek_Muthianon Mesh LLM: distributed AI computing on iroh

True, I've been telling that coding LLMs have been detrimental to local AI scene. What happened to those nifty ML projects which can run on raspberry pi zero.