Hacker News

Article URL: https://www.utubepublisher.in/2025/04/worlds-first-mind-controlled-wireless-bionic-hands.html

Comments URL: https://news.ycombinator.com/item?id=43772034

Points: 1

# Comments: 0

Article URL: https://www.implicator.ai/new-ai-training-method-lets-models-teach-themselves/

Comments URL: https://news.ycombinator.com/item?id=43772030

Points: 1

# Comments: 1

I built Text Battle, a lightweight web game where you:

- Write a character prompt in ≤ 100 chars (e.g. "A sarcastic cyber-ninja with a plasma katana").

- Earn Elo and appear in daily or unlimited-season leagues.

Feedback I’m looking for:

- Whether the 60 s cooldown feels fair vs. annoying.

- Balance of Elo gains between free and paid matches.

- I added micro-transactions because LLM costs are non-trivial; any tips on model/infra optimization to keep it free-to-play are welcome.

URL: https://plan9.kr/battle

Auth: Email, Google, Supabase Anonymous Sign-Ins

Source / tech stack: React + Supabase + Gemini API

Thanks in advance—happy to answer questions!

Comments URL: https://news.ycombinator.com/item?id=43772029

Points: 1

# Comments: 0

Article URL: https://www.theguardian.com/technology/2025/apr/23/meta-hastily-changed-moderation-policy-with-little-regard-to-impact-says-oversight-board

Comments URL: https://news.ycombinator.com/item?id=43772025

Points: 1

# Comments: 0

Hi HN!

I’ve been working on transferring LLMs across tokenizers using a new method called ALM (details in our paper [1]). It distills a model trained with one tokenizer into a version using another, enabling things like converting subword models into byte-level ones much more effectively than what was possible before.

To make this easy to use, I released tokenkit, a library implementing ALM and other tokenizer transfer methods: https://github.com/bminixhofer/tokenkit.

As a demo, I used ALM to create two byte-level instruction-tuned models:

- https://huggingface.co/benjamin/Gemma2-2B-IT-Byte

- https://huggingface.co/benjamin/Llama3-2-3B-IT-Byte

Even though the distillation phase is very short (just 1.2B bytes ≈ 330M subword tokens), the models perform competitively (for example 57.0% MMLU of the byte-level Llama vs. 62.4% MMLU of the original Llama3-3B-Instruct).

This approach opens up an interesting direction: we can potentially keep subword tokenization for pretraining (to still squeeze as much text into the model in as little time as possible), but then change to a more user-friendly tokenization afterwards.

These models aren’t yet optimized for efficiency, but if you would add self-speculative decoding plus a BLT/DTP-style hierarchical architecture and/or linearized attention, they might also be able to replace subword-based models when speed matters.

If you're interested in training your own models, the guide on tokenizer transfer via tokenkit [2] should make it easy. The model cards of the transfers above also contain the exact command used to train them. I’ve been training on fairly limited hardware, so effective transfer is possible even in a (near) consumer-grade setup.

[1]: https://arxiv.org/abs/2503.20083

[2]: https://github.com/bminixhofer/tokenkit/blob/main/docs/token...

Comments URL: https://news.ycombinator.com/item?id=43772015

Points: 1

# Comments: 0

Article URL: https://startup-map.berlin/lists/53190

Comments URL: https://news.ycombinator.com/item?id=43772012

Points: 3

# Comments: 0

Article URL: https://www.quantamagazine.org/how-a-biofilms-strange-shape-emerges-from-cellular-geometry-20250421/

Comments URL: https://news.ycombinator.com/item?id=43772000

Points: 1

# Comments: 0

Article URL: https://techcrunch.com/2025/04/22/tesla-profits-drop-71-on-weak-sales-and-anti-elon-musk-sentiment/

Comments URL: https://news.ycombinator.com/item?id=43771965

Points: 4

# Comments: 0

Article URL: https://sabordocampo.com/beneficios-do-morango-para-a-pele/

Comments URL: https://news.ycombinator.com/item?id=43771962

Points: 1

# Comments: 1

Article URL: https://robert.winter.ink/why-is-everyone-in-my-meeting/

Comments URL: https://news.ycombinator.com/item?id=43771939

Points: 2

# Comments: 0

Hi HN. I made a Chrome extension[1] that generates color palettes using traditional Japanese colors[2] based on any color you input.

These colors have been a bit of a recent obsession of mine, ever since I came across a company[3] that publishes a free color reference including traditional sets for Japan, China, and France, while on a walk.

Since then, I’ve acquired several books on the topic, had conversations with people who have been interested in this for far longer than I have, and received a grant from Adobe[4] — which I recommend applying for if you’re working on something creative.

The colors were historically used in textiles and different sources list different colors. Some include modern additions while others are more conservative.

I’ve been using the app daily for user interfaces and various creative projects. My hope is that you might get some use out of it too.

Here’s what the output looks like for Y Combinator’s orange:

https://drive.google.com/drive/folders/1VICRkO_7Z1B05KnnzF0c...

It’s nighttime for me so if you have any questions, I’ll be answering in the morning.

[1] https://irorgb.com/ [2] https://en.wikipedia.org/wiki/Traditional_colors_of_Japan [3] https://www.dic-graphics.co.jp/en/products/dcguide/index.htm... [4] https://developer.adobe.com/fund-for-design/

Comments URL: https://news.ycombinator.com/item?id=43771937

Points: 1

# Comments: 0

Article URL: https://chromewebstore.google.com/detail/typemate-fix-grammar-ever/ehgcmamfghkhiinhnhhhgjilpfokaokh

Comments URL: https://news.ycombinator.com/item?id=43771934

Points: 1

# Comments: 1

Article URL: https://www.nature.com/articles/d41586-025-01233-6

Comments URL: https://news.ycombinator.com/item?id=43771920

Points: 2

# Comments: 0

Article URL: https://phys.org/news/2025-04-harnessing-photocatalysis-hydrogen-generation-synthesis.html

Comments URL: https://news.ycombinator.com/item?id=43771901

Points: 1

# Comments: 0

Article URL: https://vendra-marketplace.io/

Comments URL: https://news.ycombinator.com/item?id=43771889

Points: 1

# Comments: 1

Article URL: https://www.techdirt.com/2025/04/23/5th-circuit-obediently-lets-att-off-the-hook-for-major-location-data-privacy-violations/

Comments URL: https://news.ycombinator.com/item?id=43771882

Points: 1

# Comments: 1

Article URL: https://www.coindesk.com/policy/2025/04/22/crypto-ally-paul-atkins-sworn-in-to-replace-gary-gensler-atop-u-s-sec

Comments URL: https://news.ycombinator.com/item?id=43771416

Points: 1

# Comments: 0

Article URL: https://melissawen.github.io/blog/2025/04/22/fosdem-2025-kernel-devroom-kworkflow-talk

Comments URL: https://news.ycombinator.com/item?id=43771413

Points: 1

# Comments: 0

Article URL: https://www.theverge.com/news/653778/max-password-sharing-extra-member-add-on

Comments URL: https://news.ycombinator.com/item?id=43771400

Points: 1

# Comments: 0

I hoped after the shitstorm cloudflare received a few months ago, about their sales tactics and treating their sales team (the viral firing tikok video, extortion attempts of customers big enough to buy enterprise plans), things would be improving but i don't see any significant changes and am growingly concerned about the state of the company.

While the core engineering side seems to still develop the platform into new heights and also have found a very convincing AI and general computing strategy (eg. the upcoming container platform), their billing platform is faulty for 6 months now (https://www.cloudflarestatus.com/incidents/gnl2dp153485) with the salesforce based ticket system broken for weeks sending most filed tickets into a black void.

Suspiciously plan upgrades seem to be less affected compared to plan downgrades and cancellations, is this a pure evil play of scale?

Comments URL: https://news.ycombinator.com/item?id=43771392

Points: 1

# Comments: 0