Hacker News Reader

guardiangodon A warning about 'model welfare'

The Torment Nexus ain't gonna build itself.

Well actually, it might.

guardiangodon Decompiling a Nintendo 64 game in 84 days

Yeah I am decompiling a PS2 game. It's not easy but I think I finally have the right setup.

guardiangodon When Genius Fails: The Intellectual Arrogance of the AI Labs

You missed the years long work at FTX under Sam Bank Friedman all the way to FTX's bankruptcy.

I am running DS v4 flash 0731 lossless at 80t/s right now. It really is not at Opus 4.5 level (for my workload). I would say it's around 3.7 Sonnet, which is still pretty good, but other models such as GLM 5.2 are still leaps better. Of course I run DSv4 flash over GLM 5.2 for a few very good reasons, but intelligence is not 1 of them.

guardiangodon Qwen3.8-2.4T

Extremely large 1 bit models are usually within 50-60% of KV divergence to lossless models. In this case I think the comparison to Opus 4.5 is a fair assessment.

Extremely large models don't suffer as much from quantization due to its weight topology also contains encoded information, so the loss of info from any one weight is somewhat mitigated.

guardiangodon Qwen3.8-2.4T

https://unsloth.ai/docs/models/qwen3.8

The 1bit quant model is at an astonishing 397GB with 95B active per MOE. This literally puts Opus 4.5 performance level into a machine a normal person could buy, and still gets usable tokens/second.

The full lossless model BF16 is clocking at 4.9TB. The model card claims the model to be between Opus 4.8 and Fable 5. Again that's astonishing as getting a machine with 7TB RAM (with context + KV cache) is still within the realm of medium size companies.

Bad things: The open source version has its vision capability removed, and the context capped at 250k . I expect someone to bolt a Kimi 2.6 vision tower to it to restore the vision capability (at less performance of course). For context, I played around with extending the context to 600k for Qwen 3.5 397b, and the context remained stable up to around 480k. It'd be interesting to see if the same can be done to Q3.8 .

Also no out of the box DSpark/DFlash support. MTP is present so we should at least get some boost in TP speed.

guardiangodon [dead]

The article is

One line after another

Hard to read it is

guardiangodon Judge Rejects Google's Attempt to DMCA Its Way Out of Being Scraped

“You are trying to kidnap what I have rightfully stolen, and I think it quite ungentlemanly."

guardiangodon OpenAI and Hugging Face address security incident during model evaluation

Don't ever ask GPT Sol on how to LARP Fallout games, thanks.

guardiangodon Towards a harness that can do anything

I disagree with the idea that file is a good metaphor for LLM. Files have seek and byte streams, which is just an unneeded abstraction for LLM. The LLM doesn't need to seek or jump to the middle of a file, if you store and organize your data properly.

Why force the LLM to use files over vector database or key-value stores, just because it's a design principal for UNIX (which is designed for human users, not LLMs.)

guardiangodon Zig Creator Calls Spade a Spade, Anthropic Blows Smoke

If your objective is just to hack something together quickly to test an idea, and doesn't care if the application segfault 6 hours later (eg. a prototype game engine), C++ is definitely the superior choice to Rust.

Rust is like the instructor demanding that you build your house up to the national code, down to the choice of nail for your floor board. Your house will be perfect, but building it is extremely difficult and high-friction. In many cases this is unnecessary.

guardiangodon Zig Creator Calls Spade a Spade, Anthropic Blows Smoke

Rust does not have a bigger training set than say Python/JS/TypeScript/C++ .

guardiangodon My thoughts on the Bun Rust rewrite

Linus always backs up his insults with technical reasonings on why the other side is wrong though. You might not agree with his reasons, but at least he provided them.

guardiangodon My thoughts on the Bun Rust rewrite

>Posting it in this sprawling, mean form was dumb.

Thank you for putting my thoughts in words I can never string together so well.

I too would just stash the letter in my ZFS dropbox and never let it see the light of day again.

The letter reads like a husband who was just served a divorce notice from his wife. The man is angry, and wants everyone to know that he is not angry, and he is very much not bothered by the whole affair even though he has misgivings since the first day of marriage.

The letter would be much more convincing if it has any technical rebuttal against the decision.

guardiangodon Every postcard tells a story

The 711s in Japan let you print color postcards from memory card for ~$1 USD . I would take photos with my camera, go to the copier machine in a 711 (they are everywhere), print the postcards, and sometimes buy the stamps from the cashier. Some stores don't have postcard specific stamps. Just ask for stamps in equivalent value 100 yen.

guardiangodon The AI Superforecasters Are Here

Given a powerful enough AI with enough outlays (eg. a Playwright instant), can an AI cheat by influencing the real world and self-fulfil its prediction?

Eg. Is corp ABC's stock price going to go down by 10% in 30 days?

Orchestrate a disinformation campaign on the 29th day to tank ABC's stock price.

guardiangodon 60% Fable cost cut by converting code to images and having the model OCR it

Truly a picture is worth a thousand words.

guardiangodon Scanning malicious websites with 'infinite' number of VPN tunnels (Part 1)

Not sure but I'd assume so. You can also setup your own Wireguard exit nodes on VPS/cloud.....or in your friends' homes.

guardiangodon Cybersecurity researchers aren't happy about the guardrails on Anthropic's Fable

I am using LLM to build some security tool, and I ran into this a few times. I have to come up with a reasoning to convince (?!!) Fable to continue the work without downgrading.

I assume Anthropic will continue to tune the model, so I am not too bothered by this.

guardiangodon Corrupting a ZFS File on Purpose

I ran 5 external USB + SMR hard disks in RAIDZ 5 for 10 years. The only thing I had to change was to use Highpoint's enterprise level USB controllers- commercial USB controllers from Realtek and Renasus are junk and will drop the drives after a while.

Even then, I had multiple cases where files were corrupted, and once the whole array refused to be online due to corrupted metadata. I had to make ZFS to replay the journal log with undocumented commands. Sometimes it takes a few days of hair-rising recovery but I always manage to get the array back intact.

The files that are corrupted are always extremely large files (>50 GB) with many small read/writes (eg. iSCSI image files.)

It's pretty impressive how resilient ZFS is, really, given I had what likely to be the worst possible hardware combination.

guardiangodon A Man Who Reads Books for a Living

ESL and I got 512 WPM and 75%. I don't agree with the 1 wrong answer but I digress.

Reading fast means you can take in more info per unit of time. It can be a useful ability, if tedious at times.

guardiangodon Why China got rich and India didn't

Because all the educated Indians moved to US and contributed to the American economy, versus the Chinese who get educated in America, then move back to China to work there.

English is a major reason- Indians are just better at speaking English than Chinese and thrives in corporate America. Whereas mainland Chinese couldn't climb the corporate ladder and have to seek better opportunities back in China.

Find me a tech executive in a Big Tech firm who is from mainland China. You can't find one. Both Lisa Su of AMD and Jensen Huang of Nvidia are Taiwanese immigrants who grew up in US in the 70s and thus speak fluent English.

guardiangodon 20 years on AWS and never not my job

I just want to contrast this article on AWS to its Azure counterpart- https://news.ycombinator.com/item?id=47616242.

2 companies have functionally similar products, but behaves completely different. One company makes technical decisions with security as the fundamental principal, while for the other company, security is not a consideration.

guardiangodon Decisions that eroded trust in Azure – by a former Azure Core engineer

Having it running on host (!), and the metadata for all guest VMs stored and managed by the same memory/service (!!), with no clear security boundary (!!!).

It's like storing all your nuke launch codes in the same vault, right in the middle of Washington DC national mall. Things are okay, until they are not okay.

guardiangodon Decisions that eroded trust in Azure – by a former Azure Core engineer

As someone who had worked adjacent to the functionally-same components (and much more) at your biggest competitor, you have my sympathy.

Running 167 agents in the accelerator? My gawd that would never fly at my previous company. I'd get dragged out in front of a bunch of senior principals/distinguished and drawn and quartered.

And 300k manual interventions per year? If that happened on the monitoring side , many people (including me) would have gotten fired. Our deployment process might be hack-ish, but none of it involved a dedicated 'digital escort' team.

I too have gotten laid off recently from said company after similar situation. Just take a breath, relax, and realize that there's life outside. Go learn some new LLM/AI stuff. The stuff from the last few months are incredible.

We are all going to lose our jobs to LLM soon anyway.

guardiangodon Using local LLM and Ghidra to analyze malware (Part 2)

Thoughts:

- Local LLM, with a powerful debugger as its oracle, is now powerful enough to run rudimentary malware analysis without consulting with external sources.

- More complex malwares are still beyond what local LLMs can handle. The local LLM can see all the behaviors by the malware, but the LLM fails to put the analysis together to deduce the true intention of a binary.

- Local LLM is a very lost-cost way to do malware analysis (about 5 US cents of electricity.)

- The biggest killer-app feature is having the LLM writes its analysis back to Ghidra. The more you interact with the LLM, the more data it will write back to Ghidra. This could potentially saves hours per manual debugging by skipping function/resources/variables labeling.

guardiangodon Nvidia Releases NemoClaw – Enterprise AI Agents, Redefined

The "NemoClaw vs. OpenClaw" section is absolutely ridiculous.

  OpenClaw vs NemoClaw (NVIDIA)
  Developer  Peter Steinberger (individual project) vs NVIDIA Corporation
  Current Status  Acquired by OpenAI (Feb 2026) vs Upcoming release (GTC 2026)
  Target Market  General-purpose consumer AI assistant vs Enterprise AI agent platform
  Core Strength  Rapid deployment, viral adoption vs Security, privacy, enterprise reliability
  Ecosystem  Community-driven (NanoClaw variants) vs NVIDIA NeMo & NIM integration
  Governance  Transitioning to foundation management vs NVIDIA-backed with open-source access
  GPU Acceleration  Not natively optimized vs Native NVIDIA GPU acceleration

NemoClaw is not even out yet, so who knows what it might look like. I guess if you sprinkle the word 'Nvidia' around enough, your product is automatically better than the rest.

I don't even like OpenClaw, but this is just silly.

https://nemoclaw.bot/claw-ecosystem-overview.html

  NemoClaw  NVIDIA (Python/NeMo)  Enterprise-grade Security Platform  Built-in privacy tools, compliance, GPU acceleration  Enterprise server architecture

guardiangodon U.S. Lost 32,000 Private-Sector Jobs in September, Says Payroll Processor

>wasted much of it on insurance or overhead.

Insurance and overhead (eg. safety harass) exist for a reason other than to drain your wallet. Roofing is also a physically difficult job. You won't find many 50+ year old to couch on a rooftop all day, regardless of pays.

guardiangodon Boeing has started working on a 737 MAX replacement

Well since the 787 program will very likely never break even, let alone turn in profit, for Boeing, the 737's replacement will be a do or die project for Boeing. They cannot afford another money-losing product.

guardiangodon When will Jaguar Land Rover restart production? "No one knows."

>The company is also asking UK government officials to provide emergency support for its suppliers to get through this period, according to people close to the talks.

The support is going to suppliers, who are the true victim, but it's privatize the gain, socialize the cost. JFR screwed up, so they should be the first to step up to assist the suppliers.