HNHacker News
TopNewBestAskShowJobs

sznio

1,189 karma · joined August 5, 2022

meet.hn/city/53.132398,23.1591679/Białystok

Socials: - t.me/szniok

Interests: Cybersecurity, Programming, Web Development

---

submissionscomments
sznio··on Microsoft Teams is a red flag
a shared drive is much better than sharepoint. I don't see how sharepoint can make it easier. the best alternative is using google workspace instead. it doesn't make you sign in four times a minute.
sznio··on Show HN: Gemini 4 Argon plays StarCraft
MoE doesn't work that way. It's not about increasing performance of a task, but speeding it up. To get a faster model, you need to make it smaller. If you make it smaller, the less it can remember. MoE tries to get faster responses without reducing the capabilities by running a smaller part of the model, and constantly swapping the parts used, so that the whole model gets to influence the output collectively, but each token is created by a smaller set of weights.

Anyone else feels like they can tell when the underlying model is MoE? Dense models are more concise in their answers. MoE seems to stall for time, writing vaugely until a right set of weights gets picked and commits to an answer. MoE also backtracks a whole lot more.

sznio··on Steam Client Is Now Faster Than Epic's New Vibe Coded Launcher
If I had a performance improvement ready for release, and my competitor started bragging about being faster, I'd for sure jump at the opportunity to release it sooner.
sznio··on Hackers activate automatic defrosting in supermarket fridges across Europe
whoever works at the hacking department of the FSB needs to hack something, anything, because they want to keep a job.
sznio··on Do Frameworks Matter Anymore?
they matter even more.

the ai agent is forced to follow some kind of structure you can comprehend. without a framework you don't even have an idea how to refactor this LLM ball of mud

sznio··on Attack of the Vibe Coded Photoshop Clones
i can imagine reading something similar about Linux 30 years ago.

> oh nooooo, what's going to happen to all the Unix vendors with the systems they sell for thousands of dollars?

adobe has fucked their customers for over a decade, is time they're fucked back. I won't miss them. Hope their employees can find jobs at better companies quickly.

also:

>They’re spending minimum $2K to vibe code a Photoshop competitor when Affinity Photo is free and Photoshop costs $200 a year. Not very smart economics.

very smart economics, because they do it once and get to have it forever, for free, for everyone. it's charity.

sznio··on Yandex bot traffic down 99% after data center attack
I don't use X myself, but my friends told me there's a noticable reduction in hateful tweets, at least in the Polish sphere. Much less comments blaming Ukrainians for everything wrong with the world.
sznio··on Recursive self-improvement through collective intelligence
>In late 2026 OpenAI/Anthropic will release a tiny model that outperforms their previously-best 2026 model (Astra/Fable), that is priced 10x cheaper, and they will market fine-tuning for it. (Lol.)

Opus 5.5 might not be 10x cheaper but it finally made me not hit the limits constantly.

sznio··on Terence Tao: Math 2.0 [pdf]
>We don’t know where this technology advance will lead or settle, so it’s absurd to try to establish a working paradigm at this point.

why not try? we'll learn something from it.

sznio··on ADHD as a circadian rhythm disorder: evidence and implications for chronotherapy (2025)
in my case I need to take it at a perfect time where it stops working just after I go to bed. Take it too early and it will stop working before I go to bed and I won't fall asleep because my mind will distract me from it. Take it too late and I technically will sleep, but I won't actually feel rested in the morning.
sznio··on ADHD as a circadian rhythm disorder: evidence and implications for chronotherapy (2025)
I noticed that my bedtime drifts because after 11PM it's finally quiet enough to do anything productive. Before that it's just impossible to work. After I started living alone, I finally could go to sleep at a sane time since I didn't have to wait hours for everyone else to calm down enough to let me think.
sznio··on ADHD as a circadian rhythm disorder: evidence and implications for chronotherapy (2025)
got this same issue. as a teen I'd just sleep 13 hours and then be groggy and in pain for the rest of the day. now I'm just incapable of sleeping more than 7 hours, I have to wake up, do something else for an hour, then go back to sleep for 1-2 hours to be finally rested.
sznio··on Claude Code’s suggested message feature: I think the real customer is the model
>Customers pay for things that are useful.

Depends on the definition of useful. In a corporation "useful" is "checks regulatory boxes" - it doesn't need to actually be useful, it must just pretend to be useful. Hence all the shit customer service processes. They are useful to the corporation as in that they tick the "we have customer service" box. Whether the consultant just hangs up on you instantly after picking up doesn't matter.

sznio··on Mistral Large 4
edit: they didn't, but they did make the first popular open-weight MoE model

didn't mistral invent the MoE architecture?

sznio··on Pi 1.0
the young yearn for X forwarding
sznio··on GPT-6.1 Sol replaces GPT-6 Sol after just 7 days, with near-Astra intelligence
Pipelining.

They probably have 6.2 just about ready, while 7 and 8 are still being cooked up. You can work on multiple releases in parallel.

We're just getting the latest training checkpoints constantly, just to edge out the other lab, while they are trying to come up with something worthy of a new major version number. It's actually worrying, since there's so much pressure to release now.

sznio··on Ask HN: Is Opus 5.5 another step change?
It's the first model that I feel I can just use freely without crashing into the 5 hour limits. I just manage to get my work done and still have tokens left over
sznio··on 'That's so AI ' What gen Alpha's biggest insult tells us
Thankfully anyone who dealt with a teacher knows how much value their opinion holds.
sznio··on AI safety is mostly a sex cult in Berkeley
Nerds like to fuck.

It's nothing surprising to anyone who's been to a furry convention. Every time someone tries to label "non-mainstream sexual behaviors" as something objectively wrong, alarm bells in my brain go off.

We've been too bullied in high school to get laid then, gotta catch up now.

sznio··on Opus 5.5 is good at explainer videos
All of these already have a similar vibe and will become commonly recognised in a few weeks time, making them not good.
sznio··on Claude Opus 5.5
Nothing, because current Haiku is useless. But I use Qwen3.6 35B-A3B and it's both much faster and much more capable than Haiku. And Qwen3.8 27B can do so much more, decently replaces Sonnet, albeit not as fast. Luna doesn't feel more capable nor faster than Qwen3.8. I think Anthropic could surprise us with some more innovation in this niche.
sznio··on AI has no intent and no motivation
>This is bound to eventually lead to a disaster when someone deluded into trusting the "AI" does some stupidity it suggests without passing it through any reality check.

Or, the AI is plugged into a machine because efficiency (aka the shareholders need even more profits), and it does some stupidity. And they will be deployed carelessly, because that makes money. And people will suffer.

sznio··on Claude Opus 5.5
I'm more excited by the Haiku 5.5 announcement buried in this post. I'm wondering if we will finally get a decently capable fast model.
sznio··on Fable 5 – Median thinking declined in August
I don't work at Anthropic, but I would assume they could serve smaller quantizations during peak hours - this effectively controls the "resolution" of the model. They could also control the resolution of the KV cache, which would make the model not necessarily dumber, but worse at understanding the incoming requests. And finally, you could pass off what was "high" effort as "extra", because why not.
sznio··on Fable 5 – Median thinking declined in August
"Your mileage may vary"
sznio··on We are unconsciously becoming someone else
>We are no longer relying on people for advice or questions, cause ChatGPT is just one tab away.

I set the line at "people" questions. I'll ask an LLM about math, programming, or how to better take care of my flowers. But turning to an AI for the human questions, relationship advice, I don't even understand why one would do that. The machine hasn't experienced that. There is no objective truth to these questions. Asking other people these questions lets you connect and bond.

I noticed a long time ago that each personality trends towards an average of the community they interact with - which makes me even more hesitant to use LLMs outside of work. I don't want to start talking like a clanker myself. I go out of my way to interact with weird people.

Thinking back to pre-LLM days, I would have the same line with classical search engines. I would never Google for relationship advice, since it doesn't know me well enough to actually say something worthwhile.

sznio··on Hugging Face is billing OpenAI $100M for hacking it
I wish I could stop.
sznio··on I stress-tested Meta Muse until its agent control plane started timing out
tekst po polsku czyta się niewiele lepiej xd
sznio··on Dario, Please
>I guarantee you the first email the CEO sends to the company is a "pause every LLM project we have - we gotta think about this".

which the government doesnt want since this will stop progress, while other countries will continue to develop LLMs

sznio··on Cops Search Flock Cameras for Reasons of 'LMAO,' 'IDK,' and 'Asdfg'
actual cops or script kiddies that found a way to impersonate them?
← PreviousPage 3 of 14Next →