Andrei Onel

I've been trying to understand all the recent drama regarding the Tristan, Levent paper and @openaibot.bsky.social I think the problem in this case is when OpenAI say they're right, they might be because of a technicality.

The conflict that we're seeing right now is between how much the AI labs want to get the credit for that advancement. They are but a tool, they are useful and should get the credit as any other tool.

I've been trying to understand all the recent drama regarding the Tristan, Levent paper and @openaibot.bsky.social I think the problem in this case is when OpenAI say they're right, they might be because of a technicality.

I feel like math has mostly been a human endeavor, even though using tools is a big part of it. Inserting a company in the middle of it which has very different incentives kinda poisons the well.

I've been trying to understand all the recent drama regarding the Tristan, Levent paper and @openaibot.bsky.social I think the problem in this case is when OpenAI say they're right, they might be because of a technicality.

This is similar to what is happening currently in AI art. Music is one of the quickest way to make a connection with another person and people feel betrayed when that is faked by AI.

I've been trying to understand all the recent drama regarding the Tristan, Levent paper and @openaibot.bsky.social I think the problem in this case is when OpenAI say they're right, they might be because of a technicality.

I think this line from OpenAI's response: “we cannot rule out that de-identified data derived from their usage of our products helped improve our models.”, is a generic legalese that is just there to cover them legally and be in sync with their ToS.

I've been trying to understand all the recent drama regarding the Tristan, Levent paper and @openaibot.bsky.social I think the problem in this case is when OpenAI say they're right, they might be because of a technicality.

The first part that hurts here is that it's starting to happen in niches of research that still only exist because of human curiosity. With no direct monetization tied to them: math, astronomy, etc

I've been trying to understand all the recent drama regarding the Tristan, Levent paper and @openaibot.bsky.social I think the problem in this case is when OpenAI say they're right, they might be because of a technicality.

@simonwillison.net's summary article history is a good one: simonwillison.net/2026/Sep/8/o...

I've been trying to understand all the recent drama regarding the Tristan, Levent paper and @openaibot.bsky.social I think the problem in this case is when OpenAI say they're right, they might be because of a technicality.

Jellyfin 12 just launched it and it has some really good things. The biggest one is making the modern layout the new default, but there are a lot of under the hood improvements to the DB and data management. Full blogpost 👇 https://jellyfin.org/posts/jellyfin-release-12.0/

This Wednesday, come build that idea you've been thinking about 💡✨ We're hosting a builder's night with Lovable at Dublin's Baseline to help non-technical builders get their idea out there. Link here 👇 https://luma.com/aic-du-9-9

This is an amazing book and looking into getting a physical copy. A bit of a sticker shock that shipping to EU is almost as much as the book 😳

Found out about flipt from @mark_a_phelps. Really great piece of software. I added v1 to the malmo store so anyone can have their own feature flag software. Very cheap to run Here is how to set it up 🧵 #selfhosting #buildinpublic

You can get the app in the store here 👇 https://malmo.network/store/flipt

Found out about flipt from @mark_a_phelps. Really great piece of software. I added v1 to the malmo store so anyone can have their own feature flag software. Very cheap to run Here is how to set it up 🧵 #selfhosting #buildinpublic

When installing it, you paste those in the install ui.

Found out about flipt from @mark_a_phelps. Really great piece of software. I added v1 to the malmo store so anyone can have their own feature flag software. Very cheap to run Here is how to set it up 🧵 #selfhosting #buildinpublic

Adam's post is blowing up on HN right now and it's a fun read. The post is mostly about how to have your own corner of the internet in a format that is easy to migrate or self-host if needed. I guess making it cheap with no lock-in is everyone's goal right now.

Interesting read, definitely worth it 👇 https://www.neatnik.net/hardcore-indieweb

Adam's post is blowing up on HN right now and it's a fun read. The post is mostly about how to have your own corner of the internet in a format that is easy to migrate or self-host if needed. I guess making it cheap with no lock-in is everyone's goal right now.

I love markdown for taking notes, open format, but I'm ok with using Obsidian and Typora to write that. Not open source tools. It's important that the underlying format is open and has no lock in.

Adam's post is blowing up on HN right now and it's a fun read. The post is mostly about how to have your own corner of the internet in a format that is easy to migrate or self-host if needed. I guess making it cheap with no lock-in is everyone's goal right now.

Even though he talks about some techniques that remind of the early 2000, FileZilla and SFTP (haven't heard those words in a while), I think the goal is to use tech that is as standardized and commoditized as possible.

You can now choose the access control for your installed apps in malmo. You can make them publicly available, or only accessible to you. #selfhosting #buildinpublic

The find: Our image was UEFI-only: systemd-boot + an EFI partition, no BIOS boot code anywhere. So under legacy BIOS the firmware finds nothing to boot and just hangs. Booting from Hard Disk is literally SeaBIOS shrugging. #selfhosting #buildinpublic

Very interesting problem I've hit with the malmo OS. It apparently does not boot on Intel based VMs on Hetzner. Specifically the ones in the CX Gen3 series The funny thing is that that series of VMs have either Intel or AMD cpus, but you can't choose. They randomly assign one