frontpage.
newsnewestaskshowjobs

Made with ♥ by @iamnishanth

Open Source @Github

fp.

Deploying a ChatGPT clone (the hard way)

https://www.natebrake.com/blog/brake-chat
1•njbrake•2m ago•1 comments

Nano Banana Pro: raw intelligence with tool use

https://quesma.com/blog/nano-banana-pro-intelligence-with-tools/
1•amrrs•4m ago•0 comments

Unique Russian A-60 Laser Testbed Jet Destroyed in Ukrainian Attack

https://www.twz.com/air/unique-russian-a-60-laser-tesbed-jet-destroyed-in-ukrainian-attack
3•pinewurst•4m ago•0 comments

I recorded a 2h meeting on my iPhone and got a full summary and PDF in 5 minutes

https://apps.apple.com/gb/app/whisperer-ai-note-taker/id6755069300
1•deepskyapps•5m ago•0 comments

New limits on school loans could narrow physician and nurse pipeline, they warn

https://www.npr.org/sections/shots-health-news/2025/11/25/nx-s1-5619731/medical-nursing-school-lo...
2•stopbulying•6m ago•1 comments

Using Nano Banana to make slideshows

https://twitter.com/ananddtyagi/status/1993380894325809274
1•ananddtyagi•7m ago•0 comments

Take the Crypto Out of the Indexes

https://www.bloomberg.com/opinion/newsletters/2025-11-25/take-the-crypto-out-of-the-indexes
3•ioblomov•8m ago•1 comments

Improving web accessibility with trace-augmented generation

http://tidewave.ai/blog/improving-web-accessibility-with-trace-augmented-generation
1•josevalim•11m ago•0 comments

Ask HN: What is your monitor setup?

1•iwebdevfromhome•11m ago•0 comments

The essence of LR parsing: Partial evaluation can turn a general parser into a p

https://dl.acm.org/doi/10.1145/215465.215579
2•fanf2•12m ago•0 comments

Show HN: All your vibe-coded designs on a single canvas like Figma

https://withcascade.com/
2•jchiu1234•12m ago•0 comments

How do you post to their social media accounts and how you get approvals?

1•isandeep1995•13m ago•0 comments

Agents Should Be More Opinionated

https://www.vtrivedy.com/posts/agents-should-be-more-opinionated/
1•vtrivedy•14m ago•0 comments

Show HN: Experimental eBPF Firewall in Rust with Heuristic Risk Scoring

https://github.com/N1ghttm4r33/Antivirus
2•n1ghtm4rr3•14m ago•0 comments

EPA Announces Final Registration of New Pesticide Isocycloseram

https://www.epa.gov/pesticides/epa-announces-final-registration-new-pesticide-isocycloseram
1•LostMyLogin•15m ago•0 comments

Google, the Sleeping Giant in Global AI Race, Now 'Fully Awake'

https://www.bloomberg.com/news/articles/2025-11-25/google-the-sleeping-giant-in-global-ai-race-no...
2•wslh•16m ago•1 comments

How I Got Software Engineering Offers from Amazon, Stripe, and Palantir (2025)

https://www.youtube.com/watch?v=PkZ94oFB9ys
2•techprep•16m ago•0 comments

It's Your Job to Understand

https://jrhawley.ca/2025/11/25/its-your-job-to-understand
2•speckx•17m ago•0 comments

Bad UX World Cup 2025

https://badux.lol/
2•CharlesW•18m ago•0 comments

Russian Gerbera drone crashed into a house in Moldova

https://militarnyi.com/en/news/gerbera-drone-falls-on-residential-home-in-moldova/
3•giuliomagnifico•22m ago•0 comments

Google Antigravity Exfiltrates Data

https://www.promptarmor.com/resources/google-antigravity-exfiltrates-data
59•jjmaxwell4•23m ago•8 comments

Anatomy of an OTT Traffic Surge: Thursday Night Football on Amazon Prime Video

https://www.kentik.com/blog/anatomy-of-an-ott-traffic-surge-thursday-night-football-on-amazon-prime/
2•oavioklein•27m ago•0 comments

This Plant will die if I'm on my phone too much [video]

https://www.youtube.com/watch?v=0rXpncpkLcw
1•siavosh•28m ago•0 comments

Nix Package Tool Approved for Availability in Fedora 44

https://www.phoronix.com/news/Fedora-44-Nix-Package-Tool
2•mlenz•29m ago•0 comments

In leaked recording, Nvidia CEO says its insane managers aren't using AI enough

https://www.businessinsider.com/nvidia-ceo-employees-use-ai-every-task-possible-2025-11
4•randycupertino•29m ago•3 comments

WebGPU is now supported in major browsers

https://web.dev/blog/webgpu-supported-major-browsers
9•astlouis44•30m ago•1 comments

"Mine Is Really Alive": Schisms in the MyBoyfriendIsAI Subreddit

https://www.thecut.com/article/romantic-ai-relationship-real-chatbot-boyfriend-dating-debate.html
4•cryzinger•33m ago•1 comments

Show HN: Kimaki – Control opencode inside Discord

https://kimaki.xyz
1•xmorse•34m ago•1 comments

The Fracturing of the World Economy

https://www.ft.com/content/b5157c3c-568e-4a49-ba19-e8bda1fc7bec
3•thm•36m ago•0 comments

AI tools that work: An honest assessment

https://www.nutrient.io/blog/ai-tools-that-actually-work-honest-assessment/
1•mooreds•37m ago•0 comments
Open in hackernews

When Do LLMs Think a pile of sand becomes a heap?

https://joshfonseca.com/blogs/sorites-paradox
1•vuciv•1h ago

Comments

vuciv•1h ago
Author here. I've always been fascinated by the Sorites Paradox (at what point does a pile of sand become a heap?), so I decided to run an experiment to see how different LLMs handle vague predicates.

I didn't just want a text answer, so I measured the probability logits for "Yes/No" tokens across pile sizes ranging from 1 to 100M grains.

Key takeaways: 1. Prompting "Is this a heap?" directly is useless (the model just agrees with your framing). 2. Few-shot prompting creates a fascinating sigmoid "heapness curve" for most models (Mistral, DeepSeek). 3. Llama-3-8B was the outlier—it remained perpetually uncertain (probs ~0.35-0.55) across almost the entire range. I argue this is actually the most "philosophically honest" reflection of how humans use the word.

I have a feeling that there is an optimal prompt for this type of experiment, but struggle to find it, or even know if I have found it. The charts in the post are rendered in-browser using the data points I collected. Curious to hear your thoughts :)