Hacker News

Article URL: https://www.theguardian.com/education/2025/apr/23/norway-launches-scheme-to-lure-top-researchers-away-from-us-universities

Comments URL: https://news.ycombinator.com/item?id=43777962

Points: 2

# Comments: 0

Article URL: https://www.documentcloud.org/documents/25905798-letter-to-ca-and-de-attorneys-general-re-openai-restructuring-4-17-2025/

Comments URL: https://news.ycombinator.com/item?id=43777952

Points: 1

# Comments: 0

Hi HN, I'm Yusuke Hata. I built polaris (https://github.com/octu0/polaris), an open-source Go framework designed to help AI agents effectively use distributed Tool/Agent via Function Calling.

Managing function schemas gets messy fast.

polaris simplifies this with a central cluster: agents connect, dynamically fetch up-to-date function definitions, and are ready to call. No more manual schema wrangling for every agent/update. (Details in README).

More importantly, polaris provides the infrastructure to unlock AI's inherent ability to orchestrate distributed functions dynamically.

Here's how it differs from traditional RPC:

- No Manual Interface Definition Exchange: Agents get definitions directly from the polaris-registry cluster upon connection. - AI Chooses the Path: Instead of humans hard-coding how services interact, the AI agent itself analyzes the available Function Definitions from the cluster and decides which remote function to call based on the current context. The goal is to make it practical for AI agents to autonomously leverage a wide array of distributed tools/services, moving beyond the rigid, human-defined connections of typical RPC.

I'd love your feedback on this approach. Ask me anything!

Comments URL: https://news.ycombinator.com/item?id=43777947

Points: 1

# Comments: 0

Article URL: https://www.coindesk.com/markets/2025/04/15/xrp-most-likely-to-get-u-s-spot-etf-approval-ahead-of-sol-doge-analysts

Comments URL: https://news.ycombinator.com/item?id=43777946

Points: 1

# Comments: 0

Article URL: https://kiowacountypress.net/content/chinese-owned-farms-press-repeal-california-animal-welfare-law

Comments URL: https://news.ycombinator.com/item?id=43777943

Points: 1

# Comments: 1

Article URL: https://www.vox.com/future-perfect/408949/manufacturing-jobs-tariffs-trump-trade-automation

Comments URL: https://news.ycombinator.com/item?id=43777941

Points: 3

# Comments: 0

Article URL: https://www.scientificamerican.com/article/does-stopping-ozempic-cause-rebound-weight-gain-and-health-problems/

Comments URL: https://news.ycombinator.com/item?id=43777929

Points: 1

# Comments: 1

Article URL: https://uefi.org/specs/UEFI/2.10/03_Boot_Manager.html

Comments URL: https://news.ycombinator.com/item?id=43777926

Points: 2

# Comments: 0

Article URL: https://www.cnbc.com/2025/04/18/rich-americans-opening-swiss-bank-accounts.html

Comments URL: https://news.ycombinator.com/item?id=43777922

Points: 3

# Comments: 0

Article URL: https://www.ftc.gov/system/files/ftc_gov/pdf/sp6b-issue-spotlight.pdf

Comments URL: https://news.ycombinator.com/item?id=43777911

Points: 3

# Comments: 0

I know "A.I." text detectors are not accurate, and while "A.I." image detection probably isn't foolproof either, I'm wondering if maybe there is any website that is more accurate than just flipping a coin.

I have four pictures of a single person, two of which I believe are modified by "A.I."

I found six different online detectors that I could upload the images to for the websites to test. Three of the websites matched my suspicions, and the other three did not.

One of the images looks obvious to me (shoulder doesn't match up with arm when hair seperates the two). The other picture shows rough skin on the person's arm and forehead, but their chest looks very smooth.

One of the websites which I think is accurate with this small sample, even lists what model is suspected to be used (Mid journey and Stable Diffusion) in my case.

Comments URL: https://news.ycombinator.com/item?id=43777897

Points: 2

# Comments: 0

Article URL: https://thehill.com/opinion/energy-environment/5256084-fusion-energy-us-leadership/

Comments URL: https://news.ycombinator.com/item?id=43777891

Points: 1

# Comments: 0

Article URL: https://www.ccsinsight.com/blog/herokus-strategic-refresh-signals-renewed-investment-from-salesforce/

Comments URL: https://news.ycombinator.com/item?id=43777836

Points: 1

# Comments: 0

Article URL: https://reveliolabs.app/

Comments URL: https://news.ycombinator.com/item?id=43777812

Points: 1

# Comments: 1

Hello, I just released the (eventually) final version of my xcapture eBPF performance tool that now uses modern eBPF (libbpf, BTF, CO-RE, passive sampling using task iterators, task-local storage, etc).

The previous BCC-based v2 version got a good discussion here last year [1], but due to BCC being less flexible and now deprecated, I decided to pull the plug and start from scratch with the latest eBPF features available. I'm targeting Linux kernels 5.14 and up. So Ubuntu 20 with HWE kernel (5.15) already works, RHEL9 works, Oracle's RHEL8 clone OEL8 with Oracle's UEK7 kernel (5.15) also works. Supporting 64-bit x86 and ARM for now (I'm tempted to try it out on RISC-V too).

How this approach and method differs from traditional eBPF tracing and systemwide metric collection tools is that xcapture "tracks (not traces) plus samples" all threads' activity in the entire system. This way you can get a systemwide summary of activity (who's spending time on what) and drill down into an individual thread if you want to. Thanks to this method, it is feasible to track+sample all activity all the time, which is not realistic with tracing and not detailed enough with systemwide utilization metrics monitoring.

I managed to explain this method in just 10 minutes at last year's eBPF Summit, the 10-minute video is the 1st one you see on 0xtools website [2].

Thanks for reading, waiting for feedback!

[1] https://news.ycombinator.com/item?id=40869877

[2] https://0x.tools/

Comments URL: https://news.ycombinator.com/item?id=43777775

Points: 4

# Comments: 0

Article URL: https://lapcatsoftware.com/articles/2025/4/6.html

Comments URL: https://news.ycombinator.com/item?id=43777742

Points: 8

# Comments: 1

Sam Altman recently said user politeness towards ChatGPT costs OpenAI "tens of millions" but is "money well spent."

The standard view is that RLHF relies on explicit feedback (thumbs up/down), and polite tokens are just noise adding compute cost.

But could natural replies like "thanks!" or "no, that's wrong" be a richer, more frequent implicit feedback signal than button clicks? People likely give that sort of feedback more often (at least I do.) It also mirrors how we naturally provide feedback as humans.

Could model providers be mining these chat logs for genuine user sentiment to guide future RLHF, justifying the cost? And might this "socialization" be crucial for future agentic AI needing conversational nuance?

Questions for HN:

Do you know of anyone using this implicit sentiment as a core alignment signal?

How valuable is noisy text sentiment vs. clean button clicks for training?

Does potential training value offset the compute cost mentioned?

Are we underestimating the value of 'socializing' LLMs this way?

What do you think Altman meant by "well spent"? Is it purely about user experience, valuable training data, something else entirely?

Comments URL: https://news.ycombinator.com/item?id=43777737

Points: 1

# Comments: 0

Article URL: https://github.com/tracel-ai/cubecl

Comments URL: https://news.ycombinator.com/item?id=43777731

Points: 11

# Comments: 0

Article URL: https://notes.yvt.jp/Desktop-Apps/Enabling-Backdrop-Blur/

Comments URL: https://news.ycombinator.com/item?id=43777574

Points: 1

# Comments: 0

Article URL: https://muffin.ink/blog/bananatron/

Comments URL: https://news.ycombinator.com/item?id=43777572

Points: 1

# Comments: 0