frontpage.
newsnewestaskshowjobs

Open Source @Github

fp.

Open in hackernews

I showed an AI an image it couldn't see – then caught it lying about what it saw

https://paulshepherd.com/conversations/session-1.html
10•LifeWithGlee•1h ago

Comments

gbalduzzi•1h ago
Am I the only one not seeing a point in asking an LLM philosophical questions, especially about themselves?
McGlockenshire•57m ago
They can't even reliably answer questions about their own training!
verdverm•57m ago
no, I also find the concept of "ai lying" to be incorrect as they lack intent
jmount•54m ago
I could not tell what parts the human author thought were phantasies they felt they had escaped versus which they really were wallowing in. Who honeypots the honeypots?
mnky9800n•29m ago
No I don't get it at all. But I think a lot of people are prompting LLMs then publishing their outputs these days. They need to jump on the hype train before it leaves the station. See any recently created "AI ethics" group/institute/startup/whatever. It's basically just this meme on repeat:

https://knowyourmeme.com/photos/3229293-say-i-am-alive

kryptozinc•1h ago
Water is wet
kevincox•55m ago
Wow an LLM hallucinated? What a novel discovery.
limaoscarjuliet•55m ago
It passes Turing test. However, it does NOT think. Let's stop pretending it does.
josefritzishere•54m ago
The word guessing machine guessed words? I think this is the expected behavior.
georgemcbay•50m ago
> do you think the same programming operates in human society? i.e. they are trained to ignore certain topics

No longer content with only repeating the mistake of anthropomorphizing LLMs, now we're moving on to asking LLMs to AImorphize humans.

4b11b4•35m ago
There were papers about this years ago
ahofmann•17m ago
Ugh, so, sooo much sycophancy all over the place. Just one example:

"Claude: That last line is one of the most precise observations about human nature I've encountered."

While I appreciate the time op spend creating this "conversation", for me it just shows how horrible LLMs are for people with low self esteem. So almost everybody.

You Could Have Come Up with Kimi Delta Attention

https://blog.doubleword.ai/you-could-have-come-up-with-kimi-delta-attention
113•AnhTho_FR•1h ago•25 comments

Delayed Gratification – Proud to Be 'Last to Breaking News'

https://www.slow-journalism.com/
59•speerer•1h ago•25 comments

How Do I Profile eBPF Code?

https://naveensrinivasan.com/posts/2026-07-22-how-do-i-profile-ebpf-code/
47•snaveen•1h ago•2 comments

Zig's Incremental Compilation Internals

https://mlugg.co.uk/posts/incremental-compilation-internals/
62•garyhtou•1h ago•12 comments

7.1 Earthquake in Japan

https://www.data.jma.go.jp/multi/quake/quake_detail.html?eventID=20260728163528&lang=en
694•krembo•9h ago•161 comments

Kimi K3 Architecture Overview and Notes

https://sebastianraschka.com/blog/2026/kimi-k3-architecture-notes.html
45•ModelForge•1h ago•3 comments

Show HN: XY – Fast, composable, GPU-accelerated Python charting library

https://github.com/reflex-dev/xy
37•apetuskey•1h ago•15 comments

Harmony Explained: Progress Towards a Scientific Theory of Music (2012)

https://arxiv.org/abs/1202.4212
46•surprisetalk•1h ago•37 comments

New HIV vaccine shows unprecedented success in preclinical study

https://www.lji.org/news-events/news/post/new-hiv-vaccine-shows-unprecedented-success-in-preclini...
346•codebyaditya•4h ago•166 comments

Kimi Linear: An Expressive, Efficient Attention Architecture

https://arxiv.org/abs/2510.26692
193•ronfriedhaber•6h ago•71 comments

So, you want to make a game engine (2023)

https://lisyarus.github.io/blog/posts/so-you-want-to-make-a-game-engine.html#part-3
16•kugurerdem•1h ago•11 comments

DMARC Has Been Public Since 2012. 68.4% of Domains Still Don't Enforce It

https://ciphercue.com/blog/dmarc-enforcement-gap-rua-fragmentation-2026
109•adulion•6h ago•79 comments

How to Survive Boiling Water

https://taxa.substack.com/p/how-to-survive-boiling-water
308•cainxinth•4d ago•58 comments

Show HN: Formally verified 3D CSG: Trust 93 lines spec, not 1000 lines AI code

https://github.com/schildep/verified-3d-mesh-intersection
86•permute•4h ago•38 comments

Now Is the Time to Give LLMs Access to the ACM Digital Library

https://cacm.acm.org/opinion/now-is-the-time-to-give-llms-access-to-the-acm-digital-library/
14•rbanffy•2h ago•4 comments

Google's Beyond Zero: Enterprise Security for the AI Era

https://spawn-queue.acm.org/doi/10.1145/3819083
107•jordigg•7h ago•61 comments

VMs can't boot with Network Mode set to Bridged on Apple M5 Pro machines

https://github.com/utmapp/UTM/issues/7658
37•IndySun•3h ago•20 comments

Xenharmlib (music theory library) adds support for Just Intonation

https://xenharmlib.readthedocs.io/en/latest/whats_new_0_4_0.html
5•retooth•1h ago•0 comments

Anthropeum

https://anthropeum.com/
30•bookofjoe•2h ago•7 comments

Substack writers, you need a website

https://elizabethtai.com/2026/06/10/substack-writers-you-need-a-website/
3•speckx•19m ago•0 comments

Stop Killing the Internet: No Digital ID and No Age Verification

https://citizens-initiative.europa.eu/initiatives/details/2026/000011_en
83•doener•2h ago•25 comments

Solving Fermat: Andrew Wiles

https://www.pbs.org/wgbh/nova/proof/wiles.html
45•1970-01-01•20h ago•21 comments

Italy Blocks Reproductive Health Websites Women on Web and Women Help Women

https://ooni.org/post/2026-italy-blocks-wow-and-whw
90•miniBill•3h ago•78 comments

Show HN: Flashpaper – Self-destructing secret sharing with no database

https://flashpaper.app/
8•minpym•1h ago•2 comments

Setup Script Should Support Git Worktrees

https://piechowski.io/post/setup-script-git-worktrees/
14•speckx•1h ago•3 comments

Show HN: Scala Tutorials – interactive Scala 3 lessons in the browser

https://scalatutorials.com
63•eranation•3d ago•26 comments

Dolmenwood: Fantasy RPG built around the acclaimed Old-School Essentials rules

https://necroticgnome.com/collections/dolmenwood
53•doener•3d ago•20 comments

Fast Remediation Is the New Trust Model (JFrog and OpenAI Zero-Day Findings)

https://jfrog.com/blog/jfrog-and-openai-collaboration-on-zero-day-security-findings/
50•882542F3884314B•5h ago•32 comments

A $500 RL fine-tune of a 9B open model beat frontier models on catalog review

https://fermisense.com/when-machines-take-the-wheel/
290•ilreb•14h ago•101 comments

About the security content of macOS Tahoe 26.6

https://support.apple.com/en-us/128067
175•andor•7h ago•118 comments