fp.
newest
Ask HN: Is AI not creative because its creators don't care about creativity?
1
•
amichail
•
1m ago
•
0 comments
Job Interviews Are Broken
https://www.theatlantic.com/technology/2025/10/ai-cheating-job-interviews-fraud/684568/
1
•
cebert
•
3m ago
•
1 comments
Oxlint JavaScript Plugin Support
https://voidzero.dev/posts/announcing-oxlint-js-plugins
1
•
kevinak
•
4m ago
•
0 comments
DocMind, Streamlit Application Leveraging LlamaIndex, LangGraph, and LLM
https://github.com/BjornMelin/docmind-ai-llm
1
•
nashashmi
•
5m ago
•
0 comments
Can Sonnet 4.5 hack a network?
https://www.incalmo.ai/blog/2025/10/01/sonnet_eval/
2
•
bsingerzero
•
6m ago
•
0 comments
Dependency on monopoly cloud providers a security vulnerability?
https://www.euronews.com/next/2025/10/20/huge-internet-outage-hits-mobile-apps-and-websites-such-...
3
•
DaveZale
•
7m ago
•
0 comments
AI Trading in Real Markets
https://nof1.ai/
1
•
hiddencost
•
7m ago
•
1 comments
The AI Cloud
https://rauchg.com/2025/the-ai-cloud#pages-to-agents
2
•
tamasnet
•
8m ago
•
0 comments
The Price of E. Coli
https://press.asimov.com/articles/price-of-ecoli
1
•
mailyk
•
12m ago
•
0 comments
Bedside Manners: Can empathy be taught in medicine?
https://harpers.org/archive/2025/10/bedside-manners-rachel-pearson-empathy-medical-education/
1
•
Vigier
•
14m ago
•
0 comments
Anthropic brings Claude Code to the web
https://techcrunch.com/2025/10/20/anthropic-brings-claude-code-to-the-web/
1
•
fatihkocnet
•
17m ago
•
1 comments
Most Accurate National IQs Possible
https://www.sebjenseb.net/p/most-accurate-national-iqs-possible
2
•
JumpCrisscross
•
17m ago
•
0 comments
Insights and Surprises from the 2025 DORA report
https://substack.com/inbox/post/176671189
1
•
mathattack
•
18m ago
•
0 comments
Ion: GraphQL on the backend (and everywhere else)
https://the-nerve-blog.ghost.io/ion-graphql-on-the-backend-an/
1
•
mprast
•
22m ago
•
0 comments
How to use infer keyword in TypeScript
https://www.geodev.me/blog/how-to-use-the-infer-keyword-in-typescript
1
•
icelandico_
•
26m ago
•
0 comments
To Fight ICE, Portland's Leaders Turn to What They Know Best: Zoning
https://www.nytimes.com/2025/10/20/us/politics/portland-ice.html
3
•
JumpCrisscross
•
26m ago
•
0 comments
Resonant Computing Manifesto
https://resonantcomputing.org/
2
•
herbertl
•
28m ago
•
0 comments
First Self-Propagating Worm Using Invisible Code Hits OpenVSX and VS Code
https://www.koi.ai/blog/glassworm-first-self-propagating-worm-using-invisible-code-hits-openvsx-m...
5
•
dnslavin
•
28m ago
•
0 comments
[Aristotle], On Trolling (2016)
https://philarchive.org/archive/BARAOT-9
1
•
aebtebeten
•
29m ago
•
0 comments
BofA Warns of Forced Stocks Selling If Credit Problems Persist
https://www.bloomberg.com/news/articles/2025-10-20/bofa-warns-of-forced-stocks-selling-if-credit-...
10
•
zerosizedweasle
•
31m ago
•
2 comments
Show HN: NativeBlend – Text to fully editable 3D Models that don't suck
https://native-blend-app.vercel.app/
1
•
addy999
•
37m ago
•
0 comments
What if AI worked like a forest?
https://worldsensorium.com/what-if-ai-worked-like-a-forest/
1
•
dnetesn
•
37m ago
•
0 comments
More Than a Feeling: How awe and wonder transform science and you
https://nautil.us/more-than-a-feeling-1242661/
1
•
dnetesn
•
38m ago
•
0 comments
Graying Hair May Reflect a Natural Defense Against Cancer Risk
https://www.ims.u-tokyo.ac.jp/imsut/en/about/press/page_00079.html
3
•
geox
•
40m ago
•
0 comments
Ask HN: How Do You Hire?
1
•
lazaruslong
•
42m ago
•
2 comments
Anthropic Sandbox Runtime (Srt)
https://github.com/anthropic-experimental/sandbox-runtime
2
•
lawrencechen
•
43m ago
•
0 comments
How YouTube algorithm manufactures consent (2024)
https://archive.org/details/youtube-icw
4
•
lr0
•
43m ago
•
2 comments
Go's experimental Green Tea garbage collector didn't help performance
https://www.dolthub.com/blog/2025-09-26-greentea-gc-with-dolt/
1
•
PaulHoule
•
43m ago
•
0 comments
Nix CI Benchmarks
https://garnix-io.github.io/benchmarks/
2
•
jkarni
•
43m ago
•
0 comments
China has 55% of the high-IQ working-age people
https://sofiechan.com/p/4571
15
•
speckx
•
43m ago
•
18 comments
Load More
Open in hackernews
Optimizing LLM Context for Vulnerability Scanning
https://blog.fraim.dev/optimizing_llm_context_for_vulnerability_scanning/
2
•
excitedrustle
•
2h ago
Comments
prestonprice57
•
2h ago
This is super interesting! It would be cool to see the same experiment with code and the context around it (ie callsites from other files, etc)
prestonprice57•2h ago