HNHacker News
TopNewBestAskShowJobs

piyh

1,870 karma · joined November 22, 2016

submissionscomments
piyh··on When LLM judges agree, should we believe them?
No modern LLM can tell how many eyes the magic card Pit Imp has. They all say 2. This is across all reasoning levels and paid Gemini, Claude, and GPT (Sol)

Drawing a line red to split up the image then has them answer correctly.

Their failure modes are highly correlated.

piyh··on Moonshot serves Claude instead of Kimi and collects exchanges for model training
>Zhipu initially attempted to target the cyber capabilities of Anthropic’s Fable model. Fable—Anthropic’s top generally accessible model—has strengthened cyber safeguards, making it more difficult for would-be distillers to target Fable’s cyber capabilities. Zhipu eventually gave up trying to target Fable after Anthropic’s cyber safeguards degraded Zhipu’s attacks.
piyh··on DeepSeek v4.1 Flash
My friend's vibe coded vercel site got hit by Meta for 21 million page views in 2 days. It cost him over $300.

My self hosted compose stack running in my basement with two 9's of uptime was a 1 time cost of $600 between cat6e, refurb mini PCs and tons of time prompting for NixOS flakes that met my needs. I'm not sure who came out ahead.

piyh··on Research acceleration: The view inside OpenAI
OpenAI helped out a mass shooter in Tumbler Ridge, plus all the suicides

Anthropic training on CoT for multi gens: https://www.lesswrong.com/posts/K8FxfK9GmJfiAhgcT/anthropic-...

Can't find anything specifically about the Gemini issue being a training data contamination, but the depression was real:

https://www.businessinsider.com/gemini-self-loathing-i-am-a-...

I think the Gemini depression being persisted across gens via training data was a HN comment I can't find anymore, no strong source.

piyh··on I'm teaching an introductory 12 week course on Quantum Oracle Engineering
it is completely un-googlable
piyh··on Research acceleration: The view inside OpenAI
Opus was trained based on it's internal CoT due to a bug for generations. Gemini's depression extended through models. OpenAI has killed people. We've already seen cross gen misalingment.
piyh··on Discovery of a new OpenAI agent message board
I can't wait for the sky to go the color of television, tuned to a dead channel
piyh··on Path to Astra: critical capabilities and frontier safeguards
>we paused certain frontier training (including certain training for Astra) for two weeks

So the pause wasn't really a pause, got it

piyh··on Path to Astra: critical capabilities and frontier safeguards
do you really think that a less negligent anthropic/oai/meta would really fair better against future models?
piyh··on Actinide is first startup to produce high-assay low-enriched uranium (HALEU)
>Under the Atomic Energy Act, all information not specifically declassified is classified as Restricted Data, whether it is privately or publicly held

https://en.wikipedia.org/wiki/Separation_of_isotopes_by_lase...

piyh··on Electric shock gloves in use by police departments in Bellevue, Omaha
The Legend of Korra come to life
piyh··on Cerebras CS-4
Wouldn't you want to keep this for internal development? Keep the single thread gains for yourself.
piyh··on Qwen 3.8 27B
Your own link says:

In / Out Price

$0.289 / $2.40per 1M

piyh··on Qwen 3.8 27B
Qwen 3.6 is ~$2/m tok, 3.8 should be drop in replacement. Gemma 31B is $0.34/m tok. The price differential on these models is massive on openrouter.
piyh··on Accelerating GPT-5.6 Sol Ultrafast
Feels like the 90's again where single threaded speed is improving fast. ASICs and wafer scale rather than node shrinks, but end result to me the consumer feels the same.
piyh··on Gemini 3.7 Flash
Sol on Cerebras is going to be expensive AF
piyh··on Nvidia Nemotron 3.5 Lightning
That SWEBench drop going from bf16 to fp4 is massive
piyh··on Show HN: Run an 80B Qwen in 4.3 GB of RAM on a Mac, and a 35B on an iPhone
SSD NAND reads are nearly infinite. Still makes me uncomfortable, but writing is what kills. There's a reason SSDs are rated by TBW, not TBR.
piyh··on Karpathy’s Pelican
Rendering 3d worlds has hugely improved though.
piyh··on Show HN: Open-source engine running Gemma 4 26B in 2 GB RAM on any M-series Mac
Gemma E2B QAT will run on a 4 gig RAM pi.
piyh··on Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber
Early days or not, Google fucking deleted my IDE and wiped my settings. It took them days to roll out a fix, by which point I had migrated off Antigravity.
piyh··on Sleep regularity is a stronger predictor of mortality risk than sleep duration (2023)
>As a software developer

My man has written too many JIRA story descriptions

piyh··on Kimi K2.7 Code is generally available in GitHub Copilot
I tried running any model on my 1070 and it instantly crashes my old tower, probably time to get off windows and run linux on it.
piyh··on Developers don't understand CORS (2019)
The only thing I remember about CORS is that it takes way longer than expected to debug, by design the error messages sent to the browser are intentionally gutted, and CORS error scenarios are hard to tell from other failure modes atfirst glance.
piyh··on Ask HN: What are tools you have made for yourself since the advent of AI?
https://offmetaedh.com

Art search for magic cards

piyh··on Playing with Vision Embeddings
I was asking nano banana to modify pictures of me to reflect if I dressed or wore my hair differently.

It ended up inserting alien hands touching my hair with smooth pale skin holding weird wormy things.

Every once in awhile you get a peek of the Eldritch horror lurking below our helpful assistants

piyh··on Qwen3.7-Max: The Agent Frontier
TBF, Flash 3.5 was released 2 days ago
piyh··on Gemini CLI will stop working from June 18, 2026
They nuked anti-gravity and installed their codex knockoff in place. The vs code fork IDE and all your settings with it have been removed. Reinstalling the anti-gravity IDE, as it's been renamed does not bring back any of your settings or extensions.

This is a cluster across the entire product line

piyh··on Agents need control flow, not more prompts
9 different frameworks being pushed in the comments of this thread. 2026 truly is the year of agents.
piyh··on Higher usage limits for Claude and a compute deal with SpaceX
Are you using haiku for most tasks? I'm in the Google ecosystem so I'm curious how it is on the other side.
← PreviousPage 2 of 21Next →