this post was submitted on 28 Jan 2025

230 points (96.7% liked)

Technology

61778 readers

5953 users here now

This is a most excellent place for technology news and articles.

Our Rules

Follow the lemmy.world rules.
Only tech related content.
Be excellent to each other!
Mod approved content bots can post up to 10 articles per day.
Threads asking for personal tech support may be deleted.
Politics threads may be removed.
No memes allowed as posts, OK to post as comments.
Only approved bots from the list below, to ask if your bot can be added please contact us.
Check for duplicates before posting, duplicates may be removed
Accounts 7 days and younger will have their posts automatically removed.

Approved Bots

founded 2 years ago

MODERATORS

230

Open-R1: a fully open reproduction of DeepSeek-R1 (huggingface.co)

submitted 1 week ago by [email protected] to c/technology

23 comments fedilink hide all child comments

all 26 comments

sorted by: hot top controversial new old

[–] [email protected] 48 points 1 week ago (9 children)

Can someone explain this? Keyloggers???

[–] Robin 34 points 1 week ago (1 children)

Playing devil's advocate here. Mouse movements and key presses have been commonly used as bot detection method for a decade now. Like that captcha service that is just a checkbox, that's part of how they guessed that you are not a bot.

[–] homesweethomeMrL 25 points 1 week ago

Yeah, no. I mean yes - that's true, and yes it's a way to detect bots, and no I'm not going to allow that wherever possible.

[–] [email protected] 30 points 1 week ago (2 children)

I wouldn’t necessarily call it key logging but all these services are going to store anything you search.

[–] homesweethomeMrL 12 points 1 week ago

"keystroke patterns or rhythms"???

Fuckin' hell.

[–] zerozaku 14 points 1 week ago

Damn it we need Private-R1 now

[–] [email protected] 9 points 1 week ago (1 children)

Whose ToS is this?

[–] [email protected] 14 points 1 week ago (1 children)

DeepSeek.

[–] [email protected] 2 points 1 week ago

So the Open-R1 wouldn't be doing this?

[–] [email protected] 7 points 1 week ago

This is true for the deep seek app, not the published network.

[–] [email protected] 6 points 1 week ago

Fuuuuck that.

[–] ChogChog 1 points 1 week ago (1 children)

Expect to see this in more applications, especially when dealing with AI. Why do you feel like you’ve noticed an uptick in having to complete captchas on every website you visit?

It’s an easy way for them to validate if you’re human or some competitor AI/scraper bot that’s trying to train on their data.

OpenAI is so scared about the possibility of DeepSeek distilling their model, I guarantee they are adding a keystroke/key pattern recognition system into their own front ends to combat it. If it’s not there already which would surprise me.

Expect your privacy to continue to be eroded in the name of ~~profit~~ technological progress.

[–] 5gruel 1 points 1 week ago

Wait but distillers will surely usw the API instead oft the Frontend, right?

[+] [email protected] -6 points 1 week ago (3 children)

Of course – if the AI is supposed to give you an answer, they have to know what you are writing, so yes, logging your keystrokes is quintessential for every online service you interact with. You cannot get an answer without asking.

The wording is strange, though, and I'm not sure whether this ToS allows them to collect and process what you are typing while using their service, or all your typing.

[–] [email protected] 34 points 1 week ago (1 children)

logging your keystrokes is quintessential for every online service you interact with

No, it is not. Services expect the "complete" payload, whether a prompt, a text message, or whatever, it doesn't matter if you typed it, if you copy-pasted it or something else. None of them need to analyze stuff you've typed, deleted and never sent.

[–] lapping6596 5 points 1 week ago (1 children)

Generally yes, but there is one use case where every key stroke is often recorded and analyzed, a search bar. If it's trying to fill out suggestions as you type, every keystroke is recorded as you go.

[–] [email protected] 5 points 1 week ago

They don't need to read the keystrokes, they need to read what's in the input box. In programming terms, you're evaluating the field in real time, you're not waiting for the "send request", nor are you keylogging, otherwise the existence of the field would be irrelevant.

[–] [email protected] 15 points 1 week ago

Keystroke patterns and rhythms is above and beyond, though. That's not remotely necessary and the kind of thing that can only be used to track an individual across multiple platforms and attempts at anonymity. I don't know how effective it is at that, but that is the sole purpose unless maybe they are training a better autocorrect tool and think that would be helpful.

At any rate, that's the point where I noped out. They are completely honest about putting every effort into identifying users and associating them with real identity. Such a system would be quite capable of de-anonymizing marketing profiles, health data, etc. by correlating vast amounts of data.

[–] [email protected] 9 points 1 week ago (1 children)

Quintessential does not mean "really essential", and does not make sense in this context.

You can't really be quintessential "for" something; only quintessential of something.

[–] [email protected] 1 points 1 week ago

Ethymologically you are right, I wasn't really aware of the alchemical background of five rounds of destillation when I wrote my comment.

Nonetheless, "quintessential for" is not unheard (or rather unread?) of:

It will take another generation or two until this usage becomes normalized, so thank you for pointing me to a better style.

[–] [email protected] 14 points 1 week ago (1 children)

Is there a tech focused summary on everything about DeepSeek and the situation with OpenAI?

[–] kiagam 25 points 1 week ago

https://youtu.be/Nl7aCUsWykg

Fireship maybe? It is not that complicated, they just make a good cheap AI and big tech is panicking because they can only make good expensive AI