doragasu

https://mastodon.sdf.org/@doragasu

I know a little bit about almost everything, and a whole lot about almost nothing.

GitLab: https://gitlab.com/users/doragasu/projects
Blogger: https://kernelhacks.blogspot.com/

Public Key

npub1nwps98njplh0j3ze96v6z5nlr30pwhmjrxs2wp3zsgccajw6q9rs2yt2dy

NIP-05 Address

doragasu_at_mastodon.sdf.org@momostr.pink

Profile Code

nprofile1qqsfhqczneeqlmheg3vjaxdp2fl3chshtaepng98qc3gyvvwe8dqz3cprpmhxue69uhhyetvv9ujumt0d4hhxarj9ecxjmnt56cm84

Publishing to

relay.momostr.pink

Author Public Key

npub1nwps98njplh0j3ze96v6z5nlr30pwhmjrxs2wp3zsgccajw6q9rs2yt2dy

Show more details

Last Notes

2026-03-09 14:42:08 UTC

- reply

Or better this one: https://arxiv.org/abs/2410.05229
https://mastodon.sdf.org/system/media_attachments/files/116/199/738/635/778/944/original/012ce92fcdbc6424.png

2026-03-09 14:37:52 UTC

- reply

Like in here: https://arxiv.org/abs/2305.04388

2026-03-09 14:35:20 UTC

- reply

In those papers trying to confuse LLMs, what was very effective IIRC, was adding data you don't need to use to the statement. The LLM tried to use all data you gave it to solve the problem and fail. Just like when a child is solving maths problems from a text book, all problems look similar so the child internalizes that you have to add two numbers and divide by the third one. Then you change the problem and the child fails because applies the same "formula".

2026-03-09 14:26:28 UTC

- reply

I wonder if you can still design a challenge to be "LLM unfriendly" by changing the wording, just like those papers showing how an LLM aces problems like "river crossing", but if you change wording a bit, they just fail in weird and spectacular ways.

2025-12-30 11:16:23 UTC

- reply

What a great talk. I find difficult though having hope, because in my experience, most people just don't care. They will even defend the silos, no matter how I try convincing them. And if we do not get people to care, what can we do?

2025-12-16 08:50:13 UTC

- reply

I've been using LineageOS for microG for years. Also tried Graphene but I want root, thanks. Sailfish is very interesting to me, but the closed parts draw me back, hope they open them soon.

2025-12-16 08:38:05 UTC

- reply

Sailfish are working on open sourcing that. But to this day you are right, some parts of the OS are closed (and AFAIK there is intention but still no commitment on a date to open source the closed parts).

2025-12-08 08:49:33 UTC

- reply

I don't think there was many people doing word processing on the ZX81 😂

2025-11-14 13:00:18 UTC

The headline / The reality
#AI #AIBubble #Cybersecurity #Anthropic #Claude
https://mastodon.sdf.org/system/media_attachments/files/115/548/169/037/942/211/original/655f9f59671d30be.png
https://mastodon.sdf.org/system/media_attachments/files/115/548/169/370/849/032/original/aae28af3622ad4cb.png

2025-08-28 08:53:25 UTC

- reply

Base science does not need to be useful, it just grows the base knowledge of the specific field. Then eventually someone might find a use for this, but wouldn't be able to do it if this was not known.

2025-07-16 09:05:35 UTC

- reply

I have to try reading one of his books, been on my list for a long time but never get enough time for reading.

2025-06-06 06:49:03 UTC

Having a #RaspberryPi with #LibreElec (a media center #Linux distro) installed, is too subversive for #YouTube 🤷‍♂️. https://www.jeffgeerling.com/blog/2025/self-hosting-your-own-media-considered-harmful

2025-02-27 19:48:07 UTC

- reply

You are right that Nintendo deserves a lot more criticism for those tactics directly learned from mafia. But well, as for the articles in NintendoHQ, as long as it does not degrade the quality of the Deck contents, I will just ignore them.

2025-02-27 18:53:54 UTC

- reply

Could you cover them via emulation on the Deck? Just joking 😂

2025-01-25 09:07:19 UTC

"Every time we let AI solve a problem we could’ve solved ourselves, we’re trading long-term understanding for short-term productivity. We’re optimizing for today’s commit at the cost of tomorrow’s ability."
https://nmn.gl/blog/ai-illiterate-programmers

2025-01-14 17:09:10 UTC

So, for #Amazon to launch the AI-enabled #Alexa, they must first solve this small little tiny problem nobody knows how to fix (because maybe it is just not possible to fix with transformer tech used in #LLMs). https://arstechnica.com/ai/2025/01/amazon-must-solve-hallucination-problem-before-launching-ai-enabled-alexa/?utm_brand=arstechnica&utm_social-type=owned

2024-12-22 08:04:44 UTC

- reply

🤣

2024-12-12 17:27:47 UTC

- reply

1 GB is too low. 2 GB might be manageable if you are patient. You will be able to run distros with lightweight DEs like lxqt. But as soon as you open heavy applications like Firefox, 1 GB won't work, and 2 GB will be the bare minimum if you don't multitask.

2024-12-06 20:53:11 UTC

- reply

About 10 years ago, I worked at one of the ITER sub projects (Interlock), and we had this internal joke: "Fusion is the future of energy. And it will always be"

2024-11-20 10:00:00 UTC

- reply

Thanks for the info. What makes me think that there must be a blocker for federation is that I do not know of any instance other than the official. But could be there is not enough interest yet (or there are but I do not know about anyone).

2024-11-20 09:55:59 UTC

- reply

Interesting, what will you be running, just a PDS? Also a Relay or any other thing? It's my understanding that you cannot completely run an independent but federated instance, but I could be perfectly wrong since I have not dug into it.

2024-09-23 14:08:25 UTC

- reply

Does it use the infrared for pointing or is it based on accel/gyro?

2024-09-06 12:09:36 UTC

- reply

👍 That's totally fine, just do not expect too much success avoiding those comments.

2024-09-06 11:50:50 UTC

- reply

Of course people love to comment everything, specially things they do not like 🤷‍♂️. I am not telling you to stop posting those, just that if you continue, those comments will keep coming, no matter how many times you ask for people to stop. And I am trying to explain why this is like this in here. Average Mastodon user is different than average Twitter/Instagram/etc. user. Many of us do not like those platforms, and some are belligerent against them.

2024-09-06 08:24:05 UTC

- reply

That might be true, we can leave that for other conversation, because that's not my point. My point is that some criticism is to be expected when posting that kind of comments on Mastodon. So if you post something that is controversial in here, you should handle the criticism that is to be expected from controversy.

2024-09-06 07:56:16 UTC

- reply

Most people in here are because of philosophical reasons. They see Mastodon as a shelter from other abusive platforms they fled from. And you are reminding them those abusive platforms. What kind of response were you expecting?

2024-09-04 10:55:54 UTC

- reply

I regularly use Zathura, and can confirm it **does not properly work** with the ARM manual. Just for that I use Evince and at least for me it does the job without a problem.

2024-08-01 17:00:42 UTC

Please #Marketing geniuses, take note. It's tiring reading #AI even on soft drink labels. https://news.wsu.edu/press-release/2024/07/30/using-the-term-artificial-intelligence-in-product-descriptions-reduces-purchase-intentions/