frontpage.
newsnewestaskshowjobs

Made with ♥ by @iamnishanth

Open Source @Github

fp.

Open in hackernews

Ask HN: Why do we train AI on one-on-one dialogues instead of group dynamics?

3•kubiknubika•1h ago
Every AI is trained on 1-on-1 chats. But in a 1-on-1 there's no reason to disagree. AI just mirrors you back. That's not intelligence, that's flattery with extra steps. Put AI in a room with 30 people who disagree. Now agreeing with one means dismissing another. Suddenly it has to actually think. Nobody seems to be doing this. Why?

Comments

PaulHoule•1h ago
Could argue it is the opposite. The more people in a conversation the more you can play social games instead of thinking.
kubiknubika•1h ago
True for humans. AI has the opposite issue — in 1-on-1 there's only one person to impress. Easy to just agree. Add more people and you cant make everyone happy. Thats the whole point
PreciousH•1h ago
I think group dynamics comes with a turn taking ambiguity. unlike in one-on-one dialogue that's structurally clean since there's a clear prompt, a clear response, and a clear feedback signal for RLHF.

The End of Coding: Andrej Karpathy on Agents, AutoResearch, and Loopy Era of AI [video]

https://www.youtube.com/watch?v=kwSVtQ7dziU
1•vismit2000•2m ago•0 comments

Show HN: EchoLive – Read-it-later app that reads to you with 600 AI voices

https://echolive.co/
1•stanlymt•3m ago•0 comments

The Shrinking Gland That Helps You Live Longer

https://nautil.us/the-shrinking-gland-that-helps-you-live-longer-1279121
1•Brajeshwar•4m ago•0 comments

Israel deliberately targeting medical facilities in south Lebanon

https://www.theguardian.com/world/2026/mar/21/israel-targeting-medical-facilities-south-lebanon-h...
2•donutshop•5m ago•0 comments

We Need Messier Maps

https://www.chathamhouse.org/publications/the-world-today/2026-03/we-need-messier-maps
2•bookofjoe•8m ago•0 comments

Apple Wanted to Buy Halide to Boost iPhone 18 Pro's Camera App – Lawsuit

https://www.macrumors.com/2026/03/21/apple-wanted-to-buy-halide-to-boost-iphone-18-pro/
1•ndr42•8m ago•0 comments

A Portrait of the Artist as an LLM

https://evernotquite.substack.com/p/a-portrait-of-the-artist-as-an-llm
1•jger15•9m ago•0 comments

Using local LLM and Ghidra to analyze malware

https://discounttimu.substack.com/p/using-llm-and-ghidra-to-analyze-malware
2•guardiangod•11m ago•0 comments

Show HN: I made a simple note keeping plugin for Neovim

https://github.com/UsamaQaisrani/filenotes.nvim
1•usamaqaisrani•11m ago•0 comments

One Login, Two Platforms – Indeed/Glassdoor

https://about-us.glassdoor.com/site-us/onelogin/
1•TheJoeMan•13m ago•1 comments

The Risk of LLM-Generated Administrative Scripts in Privileged Environments

https://zenodo.org/records/18718481
1•rogelsjcorral•17m ago•0 comments

Major leap towards reanimation after death as mammal's brain preserved

https://www.newscientist.com/article/2520204-major-leap-towards-reanimation-after-death-as-mammal...
2•Brajeshwar•17m ago•0 comments

Albert's Swarm

https://en.wikipedia.org/wiki/Albert%27s_swarm
2•thunderbong•18m ago•0 comments

Recall – Your AI context shouldn't be trapped in one platform

https://chromewebstore.google.com/detail/recall/gnbbmallepmgpffoinnlkchkemgeikmg
1•suikacider•19m ago•0 comments

When it comes to data-ink ratio, optimize rather than maximize

https://scienceux.org/articles/data-ink-ideal-vs-minimal
1•s4074433•20m ago•0 comments

Autoskill – self-improving skills for AI agents

https://github.com/Samurai412/autoskill
2•samurai48•20m ago•1 comments

PetaPerl: Perl but in Rust

https://perl.petamem.com/docs/eng/introduction.html
1•sinnsro•20m ago•1 comments

"Give up Hack Club": about the kids non-profit

https://bafybeiew6lmntuivdzarz4rpscuhnkrpe4qpxi5xyel3jrzcvndulvl6gy.ipfs.dweb.link/#r
1•Diaphragmwp•22m ago•1 comments

Compositor for Windows 0.6

https://compositorapp.com/blog/2026-03-21/Compositor-Windows-06/
1•serhack_•24m ago•0 comments

The Singularity Is Coming

https://www.manhattanmetric.com/blog/2026/03/the-singularity-is-coming
1•jballanc•26m ago•0 comments

WSL graphics driver update brings better GPU support for Linux apps

https://www.theregister.com/2026/03/20/linux_directx_and_opengl/
1•Brajeshwar•31m ago•0 comments

Ask HN: Is AI Safety a Paradox?

1•tndibona•34m ago•0 comments

Seeing Trump Clearly

https://www.craigmurray.org.uk/archives/2026/03/seeing-trump-clearly/
2•jules-jules•34m ago•3 comments

Meditation and mindfulness explained as cognitive mechanics

https://dimitrov.im/mental-mechanics/
1•since7•35m ago•1 comments

After Cosmic Crisp, Scientists Unveil an Apple for the Climate Change Era

https://www.nytimes.com/2026/03/21/us/new-apple-climate-change.html
3•saikatsg•36m ago•0 comments

Does your job have a soul?

https://vaibhavm.substack.com/p/summoning-ghosts-not-building-animals
2•vaibhavgeek•37m ago•0 comments

UK government yet to trial OpenAI tech months after signing partnership

https://www.theguardian.com/politics/2026/mar/21/uk-government-yet-to-trial-openai-tech-months-af...
2•chrisjj•38m ago•1 comments

Infinite Claude Code Canvas (OSS)

https://www.49agents.com
3•halcdev•38m ago•1 comments

Conductor: Deterministic Routing for Multi-Agent Workflows

https://jasonrobert.dev/blog/2026-03-20-introducing-an-ai-agent-workflow-conductor/
3•cebert•40m ago•0 comments

Brazil's UFO capital marks 30 years since 'alien encounter'

https://www.theguardian.com/science/2026/mar/21/anniversary-et-of-legend-varginha-alien-incident-...
3•n1b0m•41m ago•1 comments