Hacker News Reader

Comment by aurareturn | Hacker News Reader

Counter point: Would breaking up Meta, Google, Apple, Microsoft decrease the progress in AI since those major companies are forking over hundreds of billions to build out mega AI infraustructure?

View on HN

Wouldn’t this point have had greater resonance a week or two ago before recent news about entrants in the AI space?

DeepSeek is only thing because they used OpenAI for refinement.

"Apple only exists because they used Motorola IP" type comment

It wasn’t just IP though. DeepSeek actually used OpenAIs API and resources. I’m not saying it was unethical.

Almost all innovation is derivative or incremental in some way.

The difference is that DeepSeek couldn’t actually be created as cheaply as it was without using OpenAIs resources - not just IP.

It did refinement by using OpenAI to create synthetic data.

Ok.

But it now exists, and OpenAI seems to have no way to control Deepseek, or any legitimate claim they can enforce to stop it.

Oops.

Maybe, but why should we care? So far those billions haven't actually given us anything useful, it's all just toys. I'm not against research, but when the research hasn't yet produced a useful outcome then I don't feel like there is a strong societal good in keeping it going.

I don't think that's a fair assessment. I'm not anywhere near as high on LLMs as many, but Google and Meta's moonshot side projects and insane investments into infrastructure have given us Waymo, TensorFlow, PyTorch, Kubernetes, React, MapReduce, http2, the entire Go programming language, V8, all of Facebook's investments into improving PHP and OCaml.

We can quibble over how valuable each of these things individually is, but it's not just toys. It's foundational tooling used all over the place that the larger community pretty just got for free. They've also been some of the biggest contributors to the Linux kernel.

To be fair, I don't know that we needed Google et al for this. We used to get this kind of thing from DARPA and university researchers, but Google and Facebook and friends made so much money from ads that they hired away all the best researchers so they no longer work at universities, at the same time government was hollowing out funding for basic research.

We could get this thing from real public sources if we made the public investments, but I don't know that the political will or taxpayer willingness exists to do that now. We need to get it from somewhere.

Do you really think the government funding this would lead to better outcomes? We see the censorship that China is doing with DeepSeek and we just saw the dictator in chief freezing all spending unilaterally even though that wasn’t in his power

Google Translate (and all the new competing products) alone should be reason enough to demonstrate outstanding societal good.

You can now travel anywhere in the world and communicate in the local language. Using offline models that translate between any two languages. That’s almost literally biblical magic levels of social good.

(Plus we got amazing image recognition tech, LLMs, voice and NLU stuff, AlphaFold for protein analysis, etc)

If Google Translate was to disappear, I wonder if say Duolingo could create translation software with all of the linguistic data they’ve accumulated.

what data have they accumulated?

I'm a happy DuoLingo user, but educational software with an owl-with-a-BBL mascot is not going to be a trove of translation capabilities.

I’m not sure, but I thought their business model involved applying machine learning on user answers, like their translations of excerpts of written text / handwriting samples.

I'm pretty sure they just gamified mediocre language learning techniques to make it palatable for more people to consume.

I don't think they have any significant data collected, nor do they do any handwritten anything.

I think what I was recalling is the crowdsourcing translations mechanic, which is far more low tech:

> But wait – how could a beginner-level student translate advanced sentences? The solution that Duolingo employs uses the power of crowdsourcing, which involves many students offering their attempts at translating individual sentences. As each student submits a sentence, they can rate others’ translations, and the most highly rated translations “rise to the top.”

Over time, entire documents are translated and students gain many skill points for their language practice. It’s easy to see how the data collected from users could be useful to improve the algorithms that underly computer translation[…]

https://lile.duke.edu/blog/2012/11/duolingo-learning-a-new-l...

Duolingo has come a long way from those origins. It’s a gamified language learning app now, but with support of some languages reflecting that earlier crowd-sourced era. This video is a super interesting dive into the history and current state:

https://m.youtube.com/watch?v=SoTT-GGmiXA

FWIW, I like Duolingo and think it is a healthy and productive use of gamification, but that does come at the cost of pure efficiency and comprehensive treatment of grammar. It’s best when paired with other tools.

Most people’s problem with learning a language is not speed, it’s the quit rate.

Reply on news.ycombinator.com