Hacker News Reader

Again? This happened before, if you remember this: https://www.tomshardware.com/software/search-engines/archive...

Previous HN thread:

https://news.ycombinator.com/item?id=46624740

lxeon Why is AI generated language so hard to understand?

https://gist.github.com/lxe/4cd2ca0fd3b714b25f138227c8c650d3

I've been iterating on this writing clarity prompt. The term we came up with was "Asyndetic parataxis with referential ellipsis and repeated rhetorical closure." It doesn't fully describe the exact issue I have with modern AI writing, but it's close.

lxeon Kimi K3.1 Pro and K3.1 Pro Highspeed

That's because every single model release after GPT-3.5 has been, at least in part, been trained on data synthesized by GPT-3.5 and then the other OpenAI models, as well as Anthropic models later.

Also the distillation attacks and other distillation methods that gather chain of thought information from frontier models are also used to generate vast amounts of data in order to generate synthetic datasets to post-train what I believe is the vast majority of models that are being released.

If you look at any information about how these models are trained, the details about how the data has been synthesized, where the agentic traces come from that are used in post-training, etc., they almost never tell you about exactly what model was used to generate them.

lxeon A farewell to Arc browser

https://helium.computer/ this is the browser I use now. it's basically ungoogled chromium without all the nonsense. It can run chrome extensions, unblock, doesn't have random unnecessary features, etc

lxeon Nordwand Mono – A monospace font for coders and adventurers

Which I don't understand. We have wide screens. I want to stack a lot of columns on them.

lxeon The Other Epidemic: Trying to get off benzodiazepines

I was on low-dose Ativan for a few weeks casually prescribed for anxiety. What followed is an absolute year of hell. The most unimaginably difficult time in my life. Usually they tell you that short-term low dose poses lower risk or protracted withdrawal, but it's not a purely statistical risk. There are potentially pharmacological, physiological, and other factors that govern how these drugs affect each individual and those are largely unknowable.

lxeon The Other Epidemic: Trying to get off benzodiazepines

Site seems down. Here's archive: https://archive.ph/Tb1SZ

lxeon Nordwand Mono – A monospace font for coders and adventurers

After Iosevka came out, no other coding font has replaced it for me

lxeon DeepInfer: High-performance inference framework for DeepSeek models

Interference optimization is a solved problem with no moat. Anybody can achieve this result, optimize for a particular hardware, optimize for a particular problem. In general, software is no longer a moat.

The moat is custom inference hardware, lower power, higher throughput, etc.

lxeon DeepSeek V4.1 Flash 1017 (Max)

I specifically tuned it for maximum heat and noise

lxeon DeepSeek V4.1 Flash 1017 (Max)

Hello it is I, lxe

Perplexity looks bad and unsloth just did some good quants but they don’t perform as good on my asymmetric old dual 3090 box.

However, actual quality evals show literally zero degradation in “hard math” questions against full hosted model at a cost of 1.5x - 2x CoT tokens. I expect other quants to suffer from this too. This behavior seems to decrease as precision increases. I also suspect it’s the specific tuning a what experts I’m picking.

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

Nothing is stopping it. This model is woefully bad at accurate creative red teaming however. GLM finetunes on the other hand are pretty good. And I'd bet they are already deployed and doing all sorts of deeds.

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

Why isn't this type of expert caching in the native llama.cpp yet? Why do we need a separate codebase?

lxeon Getting the most out of Opus 5.5 in Claude and Claude Code

Hah, but of course the official documentation will state to launch it in "set it and forget it" mode by defining the ultimate goal. This way it's guaranteed to waste maximum amount of tokens.

lxeon Launch HN: Magnitude (YC S25) – Self-optimizing inference engine for agents

On my local inference box I have a perpetual codex thread open in my llama.cpp checkout that I periodically ask to take a look at currently pending llama.cpp PRs, do some research on latest MTP, Dflash and other prediction or attention optimizations, do research on the latest model quants and finetunes, take a look at localLlama Reddit threads and just do essentially a sweep of the frontier.

Then it rebuilds latest llama.cpp, grabs the PRs it finds relevant to test against, and then it performs a benchmark and finalizes the upgrade and verifies what model, variant, or even a separate finetune that we should be running.

Occasionally, it performs its own optimizations and commits, which then gets superseded by pull requests and merged code that essentially validates the model's own optimization directionality.

lxeon 13 Months Sober (2025)

Yes, I was also surprised with this narrative because that's exactly how I have treated my drinking years, And I very rarely see this type of honest look back in the community. A lot of self-reflections are either self-flagellating or vilifying alcohol.

I find it very positive for myself to basically look back at any period in my life and be honest about it without re-purposing it into something that it wasn't.

lxeon Radicle: Disclosure of Vulnerability in the Network Protocol

> The network protocol used by Radicle does not give the confidentiality it was expected to give. Anyone who can observe the network path between two nodes can read the data they exchange as the data is sent in plain text.

Is this a... design choice? This feels like too egregious of an omission to be a regular vulnerability here.

lxeon Qwen 3.8 Omni Flash

Looks like the harness repo is already removed?

lxeon Thelio Mira AI Linux Workstation: 192 GB GPU Memory

I thought it was a pricing mistake. But no, you just have to select the variations. $42,412.00

lxeon When fruit is scarce, these monkeys hunt animals

Related somewhat: ground squirrels in Bay Area park started actively started hunting voles.

https://www.ucdavis.edu/climate/news/carnivorous-feeding-squ...

lxeon Tether: iMessage, SMS, etc. on Linux

Oh, fantastic, thank you. If this allows me to drop files onto an iPhone without me having to open an app, that would solve a lot of my issues. I use localsend right now and although I really love it, it still requires me to always keep the app open on the phone.

lxeon The Twelve-Factor App (2025)

I know that all of this is super relevant, but it's extremely aspirational, and I can pick apart pretty much every one of these factors on how it doesn't fully hold up when it comes to the reality of production applications.

lxeon GLM-5.3-Flash

Is the actual Z.AI ecosystem good enough to replace the main drivers like Codex and Claude? Because it looks like Z Code is just a Codex fork. Just like the Kimi Code one is.

What irks me about this is that the harnesses seem to be just an afterthought here.

Don't get me wrong, I love messing around with installing Pi, getting it hooked up with OpenRouter, and just trying all kinds of different stuff, local models, etc... but when it comes to literally just setting up a productivity environment and trusting my entire machine with it, I just run Codex.

I have heard from anecdotes where people have indeed replaced their main drivers with DeepSek V4 Flash or GLM and state that "it's almost as good as... [claude/gpt]" but I never hear anyone say "yeah, this is the model/harness that I now run on my machine and don't mess with it"

lxeon Digging the grave of my skills: Hollywood creatives training AI to do their jobs

If you haven't been following the current explosion of video gen models, including all the content you see on social media, you should pay attention.

I have no doubt in my mind that right now, in August of 2026, given a couple of weeks, an amateur with a high-end last-gen GPU or somebody with maybe a thousand bucks worth of AI credits, given a good amount of exploration/tinkering to fill the gaps that GenAI cannot currently do, (which is storyboarding, screenwriting, characters, and all the actual creative stuff), can crank out a high quality 1h+ feature film with only minor uncanny artifacts.

The reason why this isn't happening en masse right now is that most creatives are probably not focused on generating feature films, but rather content that's more digestible by social media users.

I would expect to start seeing a LOT of development in the upcoming months as some form of commodification where generated content can be both high quality and tailored to the user.

lxeon Claudette: Make Claude stop talking like a BuzzFeed article

I'm sorry, what? This just pipes the output to another LLM. You can just use a sane system prompt to do the job.

lxeon Felony Bench

Yup, I guess I should have added choice 5...

lxeon Felony Bench

Then who gets prosecuted?

lxeon AI boosted homework scores, then exam scores dropped: study

If you structure learning in such a way that makes learning just a means to some end, and overindex on that end being the ultimate goal, that's what you get.

This is a pedagogical problem that AI merely exposed. Educators need to figure out How to make students choose the scenic route instead of having them optimize for the most efficient completion of a task.

lxeon Felony Bench

Let's say I am "User". I subscribe through a "Third Party" to use "AI Agent" allowing an "LLM" to run.

I want to accomplish some legal non-nefarious task, and run the agent. The agentic loop causes a CFAA-violating behavior.

Who gets prosecuted?

1. User

2. The third party model host with whom I have the account

3. The developer of the harness /agent software

4. The developer of the LLM model

lxeon Meta's blockbuster trial draws parallels to big tobacco

Yup and this is what discovery is looking for / has already found.