Hacker News Reader

atrudeauon Less is more: Recursive reasoning with tiny networks

Very cool to see the effectiveness of recurrence on ARC. For those interested in recurrence, here are other works that leverage a similar approach for other types of problems:

Language modeling:

Scaling up Test-Time Compute with Latent Reasoning: A Recurrent Depth Approach https://arxiv.org/pdf/2502.05171

Puzzle solving:

A Simple Loss Function for Convergent Algorithm Synthesis using RNNs https://openreview.net/pdf?id=WaAJ883AqiY

End-to-end Algorithm Synthesis with Recurrent Networks: Logical Extrapolation Without Overthinking https://arxiv.org/abs/2202.05826

Can You Learn an Algorithm? Generalizing from Easy to Hard Problems with Recurrent Networks https://proceedings.neurips.cc/paper/2021/file/3501672ebc68a...

General:

Think Again Networks and the Delta Loss https://arxiv.org/pdf/1904.11816

Universal Transformers https://arxiv.org/abs/1807.03819

Adaptive Computation Time for Recurrent Neural Networks https://arxiv.org/pdf/1603.08983

atrudeauon Most promoted and blocked domains among Kagi Search users

Similar alternative for Google Search: https://chrome.google.com/webstore/detail/search-filter/eidh...

Even the defaults match this Kagi top list.

atrudeauon Brave Search Goggles: Alter search rankings with rules and filters

Filter lists can be hosted anywhere and imported with the @ syntax:

# Make these domains stand out in results +en.wikipedia.org +stackoverflow.com +github.com +api.rubyonrails.org # SPAM - never show these results experts-exchange.com # Pull filters from external source @https://clobapi.herokuapp.com/default-filters.txt

This default list is the only one I distribute but users have come up with own lists.

It would be nice to have a Github repo with such lists (or meta lists: the @ syntax works recursively, allowing lists to import other lists).

Your suggestion of having a standard for the list syntax is interesting.

atrudeauon Brave Search Goggles: Alter search rankings with rules and filters

I love the initiative, but I agree with parent that some prior art should be acknowledged.

My extension has existed since 2012: https://chrome.google.com/webstore/detail/search-filter/eidh...

In June 2020, I added support for external (collaborative) filter lists which are incredibly similar to Goggles.

Not saying anything was copied, just that it might be nice to cite/reference prior work.

atrudeauon Winning Wordle

A related question is: "What is the smallest fixed set of guesses which always solve Wordle, narrowing down possible hidden words to just one?". So far the answer is 8: MODEL LEVIN TAPPA GRABS DURGY FLYTE CHAWK SPOOR [1].

Bringing this down to 5 would mean that one could always win at Wordle with the same set of 5 guesses. Seems unlikely that such a solution exists, but interesting question nonetheless.

[1]: https://github.com/alexandres/magicwordschallenge

atrudeauon Hindley-Milner Type Inference (2012)

Why is Damas dropped when referring to HM type inference? Not being critical, just genuinely curious. I imagine there's a good reason.

atrudeauon Show HN: Watch what files any Linux process accesses

For doing the opposite - what processes access a given file - I like to use Audit (https://wiki.archlinux.org/index.php/Audit_framework#Audit_f...).

atrudeauon Show HN: A Firefox add-on to strip Google search results of 'blacklisted' URLs

https://addons.mozilla.org/en-US/firefox/addon/g-search-filt...

Similar, but also enables highlighting of certain domains.

Also available on Chrome: https://chrome.google.com/webstore/detail/google-search-filt...

atrudeauon Ask HN: Is there a search engine which excludes the world's biggest websites?

For Chrome: https://chrome.google.com/webstore/detail/google-search-filt...

atrudeauon New Method for Compressing Neural Networks Better Preserves Accuracy

Very cool idea. How does fine tuning the SVD initialization compare to training from random initialization using the same architecture? I couldn't find this in the paper.

atrudeauon Wittgenstein’s theories are the basis of all modern NLP

These older word embedding models (word2vec, GloVe, LexVec, fastText) are being superseded by contextual embeddings ( https://allennlp.org/elmo ) and fine-tuned language models ( https://ai.googleblog.com/2018/11/open-sourcing-bert-state-o... ). These contextual models can infer that "bank" in "I spent two hours at the bank trying to get a loan" is very different from "The ocean bank is where most fish species proliferate."

atrudeauon Getting started with Tmux

One feature that I really like on my Mac was iTerm2's persistence (tabs and outputs preserved across reboots). This is particularly important where I live because we have frequent electric outages.

I now have something on Tmux with https://github.com/tmux-plugins/tmux-continuum .

That's one of the best things about Tmux: there's generally a way to get it to do what you need it to do. Or something close enough.

atrudeauon Fast word vectors with little memory usage in Python

Very nice work! :) Are there any benchmarks available? I'm curious how this compares to caching frequent word vectors (Zipf's law helps here) and disk-seeking the rest.

atrudeauon What is the Kullback-Leibler divergence?

Shannon's dissertation is a great introduction (:p) to entropy. https://dspace.mit.edu/handle/1721.1/11173

atrudeauon Multi-Armed Bandits, Conjugate Models and Bayesian Reinforcement Learning

Though mentioned it the article, I'll add it here for posterity: https://web.stanford.edu/~bvr/pubs/TS_Tutorial.pdf

Great tutorial on Thompson sampling.

atrudeauon Multi-Armed Bandits, Conjugate Models and Bayesian Reinforcement Learning

I was about to send you an email! :)

atrudeauon Multi-Armed Bandits, Conjugate Models and Bayesian Reinforcement Learning

The math isn't rendering for me...

atrudeauon Relevant in these crypto days – $5 wrench attack

Given your attacker doesn't know your balance, using Trezor you can create different passphrases for different wallets ( https://doc.satoshilabs.com/trezor-user/advanced_settings.ht... ). So you can unlock a fake wallet with a small balance to circumvent the attack.

atrudeauon Relevant in these crypto days – $5 wrench attack

Your bank imposes withdrawal limits. I imagine it is for this very reason. This would be a interesting idea for Bitcoin. On Ethereum this might already be possible with a smart contract.

atrudeauon Show HN: GiveHongBao – Raise Ethereum for any cause, anywhere in the world

Please make the mechanics and fees clearer on the website. Linking to and explaining the contract would be great for transparency. With that done, I think your service can be useful.

atrudeauon Show HN: GiveHongBao – Raise Ethereum for any cause, anywhere in the world

"We don't charge a percentage of what people give. Instead, we charge a flat amount for any amount donated. Keep more of what people give."

I don't think is the Ethereum gas (transaction fee). I think they take a cut as well. Perhaps someone can check the transaction being generated and see the recipients/gas? This whole thing looks very shady and probably shouldn't be getting HN-like exposure.

atrudeauon Show HN: GiveHongBao – Raise Ethereum for any cause, anywhere in the world

This should be free and GiveHongBao should have their own donation page within the system. Charging a fee (even if flat) for donations doesn't feel right. They aren't offering enough value (they basically give you an HTML page) on top of Ethereum to justify a fee.

atrudeauon The Q Language

Or the query execution plan: Plan Q

atrudeauon Word vectors are awesome but you don’t need a neural network to find them

And manifold learning: https://people.csail.mit.edu/tommi/papers/HasAlvJaa-TACL16.p...

atrudeauon Deploying a global, private CDN

Yep, CloudFront is great. The disadvantage is missing Let's Encrypt support, which is trivial with the two options above.

atrudeauon Deploying a global, private CDN

For those hosting assets on S3, you can use something like http://idiallo.com/blog/creating-your-own-cdn-with-nginx or https://github.com/alexandres/poormanscdn with Geo routed DNS on Route 53. Seems a lot simpler than this (but probably not as feature-rich).

atrudeauon Neural Networks in DyNet (Goldberg talk, June 2017) [pdf]

Does anyone have a video of the talk?

atrudeauon We’ve lost the internet, it’s all about damage control now

IPFS?

atrudeauon The Future of Go Summit – Ke Jie vs. AlphaGo

UPDATE: Commentator thinks the Ke Jie has basically no chance at this point.

atrudeauon The Future of Go Summit – Ke Jie vs. AlphaGo

UPDATE: Redmond just did an early count and AlphaGo is slightly ahead.