SergeyHack's comments

On: ChatGPT Search

[SergeyHack]

For me it feels like Google is adding "buy" keyword to every search.

Part of the answer is here - https://www.wheresyoured.at/the-men-who-killed-google/

On: GitHub cuts AI deals with Google, Anthropic

[SergeyHack]

The same here. It helps my slow autistic context switching a lot.

It's easier for me to recognize the ideas, concepts, keywords it has written than to recall them from memory when my "memory cache" is empty.

On: Stable Diffusion 2.0

[SergeyHack]

> You absolutely do feel the same feelings consuming porn as you do actually engaging in sex

I can't believe someone says this. It's so not true in my experience. These feelings have a lot in common, but they are definitely not the same.

On: Unless democracies defend themselves, the forces of autocracy will destroy them

[SergeyHack]

+1. Here in Ukraine we see the same reasoning in Russian posts written by Kremlin trolls while they wage their information war against us.

On: NYT obtains internal TikTok document describing recommendation algorithm

[SergeyHack]

The usual companies may not have ethics as you write, but politically motivated company can easily embed "ethics" in a recommendation engine like that.

On: Cultivate the Skill of Undivided Attention, or “Deep Work” (2019)

[SergeyHack]

Vipassana meditation to reveal fears and other mental processes that are hidden. Also to repair focus disseminated by the modern culture of constant distractions. Even 15 minute session is enough to feel the potential and become more aware.

Sometimes this awareness is enough to change bad mental habits. But there are other things that you continue to do even if you understand that you don't want it. These are "imprints" NLP terminology.

To change these imprints my friend does really long Vipassana sessions for more than 10 hours. She claims to change at least one strong imprint that was "panic when staying at home alone". Which is related to an event happened when she was less than a year old.

I have not tried so long Vipassana sessions and to change my imprints I do Connirae Andreas "Core Transformation" technique. It's easier to have other person do it to you, however with enough determination you can train to do it by yourselves. Unfortunately it's still a lot of efforts to learn it. Anyway may be that helps.

On: For tech-weary Midwest farmers, 40-year-old tractors now a hot commodity

[SergeyHack]

Interesting, could you give more details on that?

On: Shipping a Neural Network on iOS with CoreML, PyTorch, and React Native

[SergeyHack]

Really great end-to-end write up.

Just a few things: in general case it's better not to use MSE after sigmoid due to slow convergence.

And "logits" variable is not logits actually, it's probabilities. Logits is what you have before applying sigmoid activation.

On: FastText – Library for fast text representation and classification

[SergeyHack]

Sounds interesting. Can these tools work on character n-grams as FastText does?

[tensor → SergeyHack]

In principle if you just put a space between each character it would, though it would also make ngrams between words which you might not want. edit: for vw, maybe the other lib has special support for character ngrams with word boundaries

On: LexVec, a word embedding model written in Go that outperforms word2vec

[SergeyHack]

Not quite about creating synonyms, but in the same area there is Semantic Vectors https://github.com/semanticvectors/semanticvectors.

They process Lucene index and create embedded representation of it. Then you can search over that representation for "semantic" matches.

Last time I checked it about a year ago the embedded collection of documents was kept in the memory and the search was implemented by a linear scan. So I suspect it can be slow on very large collection of documents.