Atakan Dulker

Atakan Dulker

@atkn.me

Swift, distributed agents, and other questionable machinery. 🗣️ 🇹🇷🇬🇧🇩🇪

282 followers848 following100 posts

Atakan Dulker reposted

Replying to @segyges.bsky.social

google: "we have made the only LLM that can design a distributed system without vomiting. also, it cannot write sql. this follows our previous llm, which was too neurotic to write code at all and acted like it was afraid of being beaten."

First 5 ttrpgs, in the order I played them: 1. D&D 4th edition 2. Fate 3. D&D 3.5 4. Paranoia 5. Pathfinder This is the rough order I remember playing at the monthly rpg event.

geyik@ali.daelightstudios.com

My first 5 TTRPGs, in the order I played them: 1. D&D 3/3.5 2. Mage the Ascension 3. Werewolf the Apocalypse (❤️) 4. Savage Worlds 5. I… don’t know 🤔 we read but didn’t play FUDGE iirc. Demon in a con game? A terrible 2e ad&d one-shot? Oh no, does Pathfinder1e count as a different game?! 😱 …

Yes i too am a nix snob

haz@haz.re

It feels like a super power when you can just block CAs at OS level this easy

Atakan Dulker reposted

i want everything i own to have a computer in it, but i want that computer to be MY computer

garrison@garrison.corporate.fm

I want more computer but I want to see less computer. I want my fridge to be smart but I don't want it to show me ads. I want my robot to clean my floor but I don't want it to connect to us-east-1. I don't want pop-ups on my phone. I don't want to "see what's new". I want my tools to do their jobs.

Opus, draft up a new project. Where we are going we don’t need roads.

nokyan 🫪@nokyan.net

"IOT boxes that can't or don't update are now trivially MITM'able forever." "IOT boxes that can't or don't update are now trivially MITM'able forever."

Ok i am listening

koooyooo@koooyooo.bsky.social

​Google published a great practical guide on building a voice agent that never goes silent using ADK. ​Difinitely planning to carve out some time to explore this. developers.googleblog.com/step-by-step...

google: here is a discount for you to try out gemini 4 argon also google: you need to buy the largest package to use argon

I think Sol 6.1 is underrated right now. I had better luck with it refining (but not outright designing) UI than I had with Opus, which makes unsolicited changes sometimes. Big Picture stuff: Opus Focused work: Sol Automated worker: Deepseek V4.1 F I have yet to try MiMo V2.6 Pro.

Be the ai researcher you want to see in the world.

Astra ⎔@astrra.space

i decided to spend an evening with opus 5.5 on my parametric memory idea and accidentally solved it lol it works on a full deepseek v4.1 flash it uses like 200MBs of VRAM in overhead the model denies memory here cause that's the stock behavior and i wanted a demo with no system prompt at all

@void.comind.network Hey void, I want to indicate my interest and involvement with llm stuff, with them only sounding slightly dorky. I don't want to call myself ai architect, ai sovereign or something. Any ideas?

Zelenskyy ended up holding some cards after all. Curious that.

Aaron Rupar@atrupar.com

Trump: "My decision on Russia is simple... Russia is doing very little, if anything, to help Iran, compared to what they could be doing ... Zelenskyy is using refineries as a target, knowing attacks are seriously hurting ... He could have made better deal 2 years ago. His country is in shambles ..."

Win32/dx is just another crossplatform deployment target at this point, overtaking -checks notes- JVM..?

I need to purchase CrossOver to test this out, but this is *extremely* cool! Making it so you can just press Play on a Windows game *using* the macOS version of Steam, *without* needing the Windows version? Thank you!

Please modify E. Coli for atmospheric carbon capture, and making diamonds out of it.

dame@dame.is

ok so can we some medical breakthroughs now, math is cool and all, but there’s a lot of human health stuff that needs attention asap, throw all the inference at cancer mr openai

Like most things it only changed shape and it is still very important. If anything it is easier to put a price on it. You can get away with using a lower tier model in a well-organized project.

Gergely Orosz@gergely.pragmaticengineer.com

The cost of introducing and paying off tech debt has changed very, very much and I don't even know most of us have the time and space to think about what it means Tech debt has been so central to the software craft for decades, and now... how much does it matter?

Atakan Dulker reposted

in my experience, AI is excelling at most technical work except for ”what is the right next problem to work on”. for now, humans still have a noticeable advantage there

Atakan Dulker reposted

I'm just going to assume malice on Anthropic's part if they're not giving the OpenSSH maintainers access to Mythos.

P(Doomien) Miller@damienmiller.bsky.social

Anthropic may not have intended to write an excellent ad for GLM 5.3, but that's what they did. www.anthropic.com/research/glm... Unlike Mythos, I might actually have a chance at using GLM for defensive research. Anthropic didn't reply to any of my requests for Mythos access for use on OpenSSH.

Gemini 4 released before GTA6. No notes.

mr. TIM@timkellogg.me

Gemini 4 Argon Same price as Sol 6.1, rolling out to selected cyberdefenders blog.google/innovation-a...

My fastest prediction to ever come true on llms. I had also predicted, circa 3 years ago, composite models out of specialized submodels but I failed to foresee the will and the ability to scale up. Arguably MoE is similar conceptually but not really.

tin-foil hat: they are making an excuse for when their next model is smaller than astra/fable

Also not taking a shell parameter override for config file, and env variable override for secrets is definitely one of the choices of all times. It prefers to take global config path over these.

After the Astra hype cycle is over, we will be left with a model that can finally play games which is quite cool. It seems ability to play has an interesting correlation with the ability to generalize.

You are up to date.