GradientAssent

> Based on the really, really crappy results that google gaves me out lately

People on Hacker News seem to think Google's results have gotten worse over time. Maybe they have for the narrow set of interests and requirements of this crowd, but I'm very confident they've gotten better for the vast majority of other users (users who I bet are less likely to use an ad blocker too).

> That they eliminate all SEO crap (i dont want those results anywhere near the 100th page of results) and ads from my results and serve me relevant content.

This is an extremely difficult engineering problem. Google doesn't make a conscious decision to include low/middling quality SEO'd content. That content is optimized to appear relevant and high quality. Google does a better job than say Bing at telling the difference, but it's still obviously an unsolved problem.

Would you pay, say, $120/year? What if it wasn't nearly as good as Google yet?

Google's revenue per user is unfortunately much higher than $10/year.

Sounds like a great opportunity for you to enter the market and capture one of these alternative revenue channels with your ad-free search engine.

And when effectively no users opt in to advertising, how does the search engine stay in business?

It does actually cost money to operate a good search engine.

Anyone have a mirror?

I'm not sure splitting off the crawler would have the effect you want it to. Crawling is hard, but actually one of the easier parts of building a search engine. There are less complete, but still pretty impressive public crawls available already, like Common Crawl.

The harder problem is what you do with the result of the crawl once you have it. You need to index it. This is one of Google's deepest moats. Being able to serve any of hundreds of billions of documents in milliseconds requires an enormous amount of infrastructure and hundreds of careers worth of difficult software engineering.

I don't see a good way to split indexing out as a separate company from the rest of the search engine. It's tightly integrated with ranking. Generally a new ranking technique requires some new information to be associated with each document in the index, e.g. some precomputed scores that describe the document's quality or fitness for different kinds of queries. At query time, you're often just weighting and comparing these scores across documents.

If you had Google's index, you'd be a long way towards being able to replicate Google's results, but I don't think that would lead to the kind of innovation you're hoping for.

I guess the present day somehow seems more relevant?

One reason to care which is bigger is the effect they each will have on the norms of governance around the world. Superpowers tend to nudge the rest of the world's governments to be more like them (see Britain and the US over the last 300 years). Do you want the rest of the world to be more like China or more like the West?

China has shown the world what it considers to be acceptable norms of governance: unrepentant, ongoing genocide against the Uigers, snuffing out liberal democracy in Hong Kong, and denying the sovereignty of a liberal neighbor in Taiwan.

I'm fully expecting plenty of whataboutism and false equivalencies in response to this comment and others making a similar point.

Believe me, I'm disappointed in the US and the rest of the West's recent slide away from liberal values, and I hope we can start doing better soon, but I encourage onlookers to ask themselves: do you really believe the there's no meaningful difference between China's vision for the future and the West's? No difference between the future each will create for you?

TFJS, yep!

Many games still aren't available on Mac, and things like Wine are not nearly as clean of an experience as just installing games in native Windows. I'd guess switching to ARM will set back the progress made in Mac gaming as well.

I made a tool for visualizing text in 3D: https://nebulate.ai

It runs a machine learning model in your browser to convert the text into points in a high dimensional space, and then it projects those points down to 3D.

Right now you can tell it to visualize post titles or comments from any subreddit or tweets from any Twitter user. I find it especially interesting to explore the news with it since every article is naturally presented alongside other articles that are about similar topics, often giving useful context.

Only works in desktop Chrome right now unfortunately. I was hoping to make it Firefox-first, but I need an API Firefox hasn't implemented yet.

Since the model runs on your machine, the "running model" stage will be slower or faster depending on your local GPU. If you have a decent GPU, I recommend bumping up to 512 points.

Would love any feedback! Here, on Twitter (@gradientassent), or in the Discord channel linked inside.

I'm working on a project right now where the centerpiece is a Three.js scene. It's about data visualization, so the surrounding code covers much of the same ground as a typical single-page web app: fetching data, UI for manipulating the dataset, etc.

I decided not to use my beloved Clojure and re-frame because it seemed like I was going to end up wrestling with Three's imperative API.

Others inclined toward projects like the OP and the Clojure ecosystem more generally: how have you approached similar projects before? Where there's a very non-Clojure-native library that will be critical?

I ended up with TypeScript, React, and Redux instead of re-frame.