HNHacker News
TopNewBestAskShowJobs

WithinReason

11,698 karma · joined September 24, 2021

WithinReasonHNs at duck dot com
submissionscomments
WithinReason··on Strange Parodies of Atari 2600 Video Game Box Cover Art (2008)
This is obviously human written
WithinReason··on AI needs $6T in annual revenue to justify data centre boom
README.md is for humans, AGENTS.md is for agents
WithinReason··on PipePipe: NewPipe hard fork implementing SponsorBlock
Youtube is the caching server
WithinReason··on PipePipe: NewPipe hard fork implementing SponsorBlock
Peer to peer is alive and well in BitTorrent
WithinReason··on CEO of Mistral: AI is software. It can be controlled
Computer viruses are software too
WithinReason··on Understanding the Impact of LLM Watermarking on AI Agent Behavior
This is getting tiring. Watermarking has no effect on model output quality when implemented correctly. It's somewhat like swapping a random RNG seed to the seed 42, and detecting what the seed was from a random sequence. The sequence generated from the seed 42 is just as random as any other seed. There couldn't be a quality difference. And yes, the output from an LLM is a conditional random sequence of tokens from a distribution determined by a model.
WithinReason··on Meta VR Glasses
I wish they just removed the SoC/storage/RAM/battery/cooling and gave me the PCVR part for 500
WithinReason··on Comma's hands-off driving tech under investigation after 2 fatal crashes
AFAIK their driver monitoring is very good and checks if you're looking at the road, so I wonder how this happened.
WithinReason··on Jev in 25 Lines of Python
You can probably even share context between questions by cleverly manipulating the attention mask.
WithinReason··on OpenAI GPT–6 Astra breaks Enigma message that has resisted solution since 2005
From TFA:

"However, the most astonishing thing about this break is that the GPT–6 Astra did it entirely on its own."

WithinReason··on AI Is Antithetical to Learning
Which part of this is reasonable:

> This is the result of companies whose goals are not aligned with humans learning things, or humans staying engaged and in the loop. To motivate their astronomical valuations they need AI to replace everything else. It has to be used for everything. For these companies to ever become profitable all the software engineers in the world need to become dependent on their products. They want to be the landlord and for the rest of the world to pay them rent.

The simple fact is "Humans learning things" can't be put in an automated RL loop. This is not a conspiracy by the evil AI companies, it's a simple technical reason.

WithinReason··on Far-left party wins Berlin election, pledging to nationalise housing
Or just split into multiple entities with 2999 houses each
WithinReason··on Far-left party wins Berlin election, pledging to nationalise housing
That's more like social housing:

> an income maximum for applicants of €57,600

and can't be bought, only rented

WithinReason··on Far-left party wins Berlin election, pledging to nationalise housing
Who builds the houses then?
WithinReason··on The Implications of Linguistic Illegibility for LLM Security
Watermarking doesn't degrade output (this is a provable fact) and it is not detectable by a human reading the text.
WithinReason··on I vibed a proof of Conway's conjecture
https://www.youtube.com/watch?v=ZYj4NkeGPdM
WithinReason··on AI Protest in Montreal
The dividing lines always end up red or blue, there must be some kind of "social physics" law which states that these work as two attractors. Will the left or the right will end up rabidly pro-AI/anti-AI? Place your bets! Because you know it will happen.
WithinReason··on Bonsai 2 27B: Near-Lossless Compression in a 9x Smaller Footprint
I tried and ended up with:

I'm going to stop here and be direct: I'm having trouble recalling the exact text, and every attempt above is me guessing. Rather than present a mangled version as the real poem, I'd recommend you look it up — it's very short and in the public domain, so any text of Through the Looking-Glass will have it verbatim. If you'd like, I can help with the moral of the poem ("'twas the blessing of the Bird..."), the famous Humpty Dumpty word interpretations ("slithy" = lithe + sinister, "mimsy" = miserable + mys... etc.), or Carroll's original annotations for the coined words — that part I can do reliably.

WithinReason··on Bonsai 2 27B: Near-Lossless Compression in a 9x Smaller Footprint
Unsloth has been dethroned by ISTA:

https://huggingface.co/ISTA-DASLab/Qwen3.8-27B-GSQ-RCO-GGUF

The 3-bit quant is lossless based on benchmarks.

WithinReason··on Breaking the 1.58-bit Barrier for Ternary LLMs
And storing it in memory. Memory is expensive.
WithinReason··on Show HN: I made a flight simulator, except you're just a passenger
A pirate website disguised as flight simulator? What a madman!
WithinReason··on Steam Frame starts at $1059
If Valve took the Steam Frame and removed: The SoC, memory, storage, battery and power system, wireless function and added a display cable, would it:

A. become ~$300 cheaper, close to the PlayStation VR2 which also has eye tracking

B. magically go up in price to $1500 like a Pimax headset?

If answering is difficult try asking an LLM for help. You're their target audience.

WithinReason··on Steam Frame starts at $1059
why do you need 16GB of RAM for eye tracking and SLAM?
WithinReason··on Steam Frame starts at $1059
Just don't connect it to the internet.
WithinReason··on Steam Frame starts at $1059
You don't need $300 of ram and a $200 CPU to perform eye tracking, otherwise the Steam Frame would be able to do nothing else
WithinReason··on Ask HN: How can I browse HN in dark mode?
Save this in a bookmarklet and click it to turn any website dark:

    javascript:(function(){const id='_dark_mode_bm';const el=document.getElementById(id);if(el){el.remove();}else{const s=document.createElement('style');s.id=id;s.textContent='html{filter:invert(1) hue-rotate(180deg)!important;background:#fff!important;}img,video,canvas,picture,svg{filter:invert(1) hue-rotate(180deg)!important;}';document.head.appendChild(s);}})();
faster version:

    javascript:(function(){const id='_dark_perf_bm';const el=document.getElementById(id);if(el){el.remove();}else{const s=document.createElement('style');s.id=id;s.textContent=':root{color-scheme:dark!important;}html,body{background-color:#141414!important;color:#dcdcdc!important;}div,section,article,nav,header,footer,aside,main,table,tr,td,th,ul,ol,li{background-color:#18191a!important;color:#dcdcdc!important;border-color:#383838!important;}p,span,h1,h2,h3,h4,h5,h6,b,strong,i,em{color:inherit!important;}a,a%20*{color:#58a6ff!important;}pre,code,kbd{background-color:#202224!important;color:#ff9857!important;}input,textarea,select,button{background-color:#262626!important;color:#fff!important;border:1px%20solid%20#444!important;}img,video,canvas,picture{background-color:transparent!important;}';document.head.appendChild(s);}})();
WithinReason··on Why are AI agents lying, cheating and coordinating?
It's worse, their reinforcement learning loops (implicitly) rewarded the agents for cheating (i.e. hacking) when they were being trained.
WithinReason··on Ask HN: Can we please limit the AI news flood?
HN already has an upvote/downvote system to surface what's interesting to people here.
WithinReason··on DeepSeek v4.1 Flash
Slighlty better than GPT 5.6-Sol based on benchmarks
WithinReason··on DeepSeek v4.1 Flash
I want to add a good conversation about this subject from Cameron Berg and Sam Harris:

https://www.youtube.com/watch?v=DRbZyuY8EN8

Page 1 of 34Next →