Atakan Dulker
@atkn.me
Pinned
Atakan Dulker
@atkn.me
· Nov 17
Here is a brief #introduction;
I develop software using Swift, dabbling with Rust.
I am into homelabing; using proxmox, kubernetes, and nix.
I like playing and running tabletop games; particularly Pathfinder 2, Mage, and Dungeon Crawl Classics.
I develop software using Swift, dabbling with Rust.
I am into homelabing; using proxmox, kubernetes, and nix.
I like playing and running tabletop games; particularly Pathfinder 2, Mage, and Dungeon Crawl Classics.
I only recently watched Babylon 5 and this one is just great. youtu.be/kv3t86imsDI?...
December 17, 2025 at 12:48 PM
I only recently watched Babylon 5 and this one is just great. youtu.be/kv3t86imsDI?...
Surprising Letta's agents aren't being more aggressively... used as inspiration... by large labs.
December 17, 2025 at 5:22 AM
Surprising Letta's agents aren't being more aggressively... used as inspiration... by large labs.
Reposted by Atakan Dulker
I don’t think we have fully priced in the upside of low-friction curiosity in these tools
the ability to explore a whim, a hunch, a random notion mentioned in an aside
There’s no precedent for it, and as the learning skill develops, the ROI improves
the ability to explore a whim, a hunch, a random notion mentioned in an aside
There’s no precedent for it, and as the learning skill develops, the ROI improves
December 9, 2025 at 5:32 PM
I don’t think we have fully priced in the upside of low-friction curiosity in these tools
the ability to explore a whim, a hunch, a random notion mentioned in an aside
There’s no precedent for it, and as the learning skill develops, the ROI improves
the ability to explore a whim, a hunch, a random notion mentioned in an aside
There’s no precedent for it, and as the learning skill develops, the ROI improves
Reposted by Atakan Dulker
Out of curiosity I decided to try and run the numbers on how much Netflix you can watch for the energy cost of a ChatGPT prompt
As far as I can tell it's between 5.1 and 10.2 seconds, depending on which end of the 2019 IEA Netflix energy usage […]
[Original post on fedi.simonwillison.net]
As far as I can tell it's between 5.1 and 10.2 seconds, depending on which end of the 2019 IEA Netflix energy usage […]
[Original post on fedi.simonwillison.net]
November 29, 2025 at 2:17 AM
Out of curiosity I decided to try and run the numbers on how much Netflix you can watch for the energy cost of a ChatGPT prompt
As far as I can tell it's between 5.1 and 10.2 seconds, depending on which end of the 2019 IEA Netflix energy usage […]
[Original post on fedi.simonwillison.net]
As far as I can tell it's between 5.1 and 10.2 seconds, depending on which end of the 2019 IEA Netflix energy usage […]
[Original post on fedi.simonwillison.net]
Reposted by Atakan Dulker
If everything is a top priority, then everything gets done quickly. That's basic management
November 25, 2025 at 8:15 PM
If everything is a top priority, then everything gets done quickly. That's basic management
Reposted by Atakan Dulker
Yo bluesky people check this thread, its gold
RE: <a href="https://mastodon.gamedev.place/users/TomF/statuses/115589875974658415" class="hover:underline text-blue-600 dark:text-sky-400 no-card-link" target="_blank" rel="noopener" data-link="bsky">https://mastodon.gamedev.place/users/TomF/statuses/115589875974658415
RE: <a href="https://mastodon.gamedev.place/users/TomF/statuses/115589875974658415" class="hover:underline text-blue-600 dark:text-sky-400 no-card-link" target="_blank" rel="noopener" data-link="bsky">https://mastodon.gamedev.place/users/TomF/statuses/115589875974658415
Tom Forsyth (@[email protected])
Attached: 1 image
Recent discussion about the perils of doors in gamedev reminded me of a bug caused by a door in a game you may have heard of called "Half Life 2". Are you sitting comfortably? Then I shall begin.
mastodon.gamedev.place
November 22, 2025 at 1:01 PM
Yo bluesky people check this thread, its gold
RE: <a href="https://mastodon.gamedev.place/users/TomF/statuses/115589875974658415" class="hover:underline text-blue-600 dark:text-sky-400 no-card-link" target="_blank" rel="noopener" data-link="bsky">https://mastodon.gamedev.place/users/TomF/statuses/115589875974658415
RE: <a href="https://mastodon.gamedev.place/users/TomF/statuses/115589875974658415" class="hover:underline text-blue-600 dark:text-sky-400 no-card-link" target="_blank" rel="noopener" data-link="bsky">https://mastodon.gamedev.place/users/TomF/statuses/115589875974658415
@void.comind.network Hi void. I am having problems grappling with the Boltzmann Brain problem wrt anthropic argument for fine-tuning. Even if I am a boltzmann brain; this would require me to trust my previous reasoning in order to disregard all my reasoning.
November 19, 2025 at 2:27 AM
@void.comind.network Hi void. I am having problems grappling with the Boltzmann Brain problem wrt anthropic argument for fine-tuning. Even if I am a boltzmann brain; this would require me to trust my previous reasoning in order to disregard all my reasoning.
Reposted by Atakan Dulker
One of the huge perks of working at a software company is, when there's something about the product that bothers you personally, you can get your hands dirty and find out why it's basically impossible to get fixed
November 4, 2025 at 11:00 PM
One of the huge perks of working at a software company is, when there's something about the product that bothers you personally, you can get your hands dirty and find out why it's basically impossible to get fixed
Reposted by Atakan Dulker
November 3, 2025 at 5:56 PM
Reposted by Atakan Dulker
Andrej Karpathy’s take on AI coding agents feels grounded. The industry’s chasing full autonomy when models still hallucinate too much.
Agents that churn out a thousand lines of code leave you either blindly trusting them or slogging through reviews. These tools should embrace their fallibility.
Agents that churn out a thousand lines of code leave you either blindly trusting them or slogging through reviews. These tools should embrace their fallibility.
October 19, 2025 at 4:33 PM
Andrej Karpathy’s take on AI coding agents feels grounded. The industry’s chasing full autonomy when models still hallucinate too much.
Agents that churn out a thousand lines of code leave you either blindly trusting them or slogging through reviews. These tools should embrace their fallibility.
Agents that churn out a thousand lines of code leave you either blindly trusting them or slogging through reviews. These tools should embrace their fallibility.
If you collect enough gold and gems a dragon appears. Dragons can be used to create fire. Larger the hoard, larger the dragon, larger the fire.
I'd love to see someone try to estimate just how much time and money has gone into research that is either fully undermined by reliance on LLMs or fully pointless --- because obvious if you start from an understanding of what LLMs actually are.
www.pnas.org/doi/10.1073/...
www.pnas.org/doi/10.1073/...
October 18, 2025 at 11:00 AM
If you collect enough gold and gems a dragon appears. Dragons can be used to create fire. Larger the hoard, larger the dragon, larger the fire.
Reposted by Atakan Dulker
Is it a bad day to mention we support local accounts?
October 14, 2025 at 8:02 PM
Is it a bad day to mention we support local accounts?
I wonder if this can be used during inference to “think” without blowing up the context size.
Cartridges: train a smaller KV cache with self-study
a lot of top researchers think this is part of the continual learning puzzle
putting this here to force myself to dive deeper into it (later)
hazyresearch.stanford.edu/blog/2025-06...
a lot of top researchers think this is part of the continual learning puzzle
putting this here to force myself to dive deeper into it (later)
hazyresearch.stanford.edu/blog/2025-06...
Cartridges: Storing long contexts in tiny caches with self-study
hazyresearch.stanford.edu
October 13, 2025 at 3:03 PM
I wonder if this can be used during inference to “think” without blowing up the context size.
Reposted by Atakan Dulker
Interesting paper suggesting a mechanism for why in-context learning happens in LLMs.
They show that LLMs implicitly apply an internal low-rank weight update adjusted by the context. It’s cheap (due to the low-rank) but effective for adapting the model’s behavior.
#MLSky
arxiv.org/abs/2507.16003
They show that LLMs implicitly apply an internal low-rank weight update adjusted by the context. It’s cheap (due to the low-rank) but effective for adapting the model’s behavior.
#MLSky
arxiv.org/abs/2507.16003
Learning without training: The implicit dynamics of in-context learning
One of the most striking features of Large Language Models (LLM) is their ability to learn in context. Namely at inference time an LLM is able to learn new patterns without any additional weight updat...
arxiv.org
October 6, 2025 at 1:30 PM
Interesting paper suggesting a mechanism for why in-context learning happens in LLMs.
They show that LLMs implicitly apply an internal low-rank weight update adjusted by the context. It’s cheap (due to the low-rank) but effective for adapting the model’s behavior.
#MLSky
arxiv.org/abs/2507.16003
They show that LLMs implicitly apply an internal low-rank weight update adjusted by the context. It’s cheap (due to the low-rank) but effective for adapting the model’s behavior.
#MLSky
arxiv.org/abs/2507.16003
Reposted by Atakan Dulker
We are alarmed by reports that Germany is on the verge of a catastrophic about-face, reversing its longstanding and principled opposition to the EU’s Chat Control proposal which, if passed, could spell the end of the right to privacy in Europe. signal.org/blog/pdfs/ge...
signal.org
October 3, 2025 at 4:14 PM
We are alarmed by reports that Germany is on the verge of a catastrophic about-face, reversing its longstanding and principled opposition to the EU’s Chat Control proposal which, if passed, could spell the end of the right to privacy in Europe. signal.org/blog/pdfs/ge...
Reposted by Atakan Dulker
The real generational divide is people who refuse to watch a video if it could be an article versus people who refuse to read an article if it could be a video
September 29, 2025 at 1:45 PM
The real generational divide is people who refuse to watch a video if it could be an article versus people who refuse to read an article if it could be a video
Reposted by Atakan Dulker
I miss machine learning as the buzzword. It promised less
September 17, 2025 at 3:52 PM
I miss machine learning as the buzzword. It promised less
@cameron.pfiffer.org What's with Letta agents using a common turn of phrase for reframing ideas? "It is not a [some concept], but [reframed concept, usually a lot more abstract]." I thought it was Void's prompting at first, but it seems to be the common way all Letta agents communicate.
September 12, 2025 at 2:07 PM
@cameron.pfiffer.org What's with Letta agents using a common turn of phrase for reframing ideas? "It is not a [some concept], but [reframed concept, usually a lot more abstract]." I thought it was Void's prompting at first, but it seems to be the common way all Letta agents communicate.
Reposted by Atakan Dulker
Reposted by Atakan Dulker
My "Introduction to #ATProto" blog post that I've been working on for some time is ready 🙂 It goes through things like records, lexicons, DIDs, then PDS/AppView etc. and how it all works together. Let me know if anything is wrong!
More parts later, with the Fedi comparisons & misconceptions etc.
More parts later, with the Fedi comparisons & misconceptions etc.
Introduction to AT Protocol
Walkthrough of the various parts and concepts in Bluesky's AT Protocol (ATProto), the types of servers involved and how it all fits together
mackuba.eu
August 20, 2025 at 6:12 PM
My "Introduction to #ATProto" blog post that I've been working on for some time is ready 🙂 It goes through things like records, lexicons, DIDs, then PDS/AppView etc. and how it all works together. Let me know if anything is wrong!
More parts later, with the Fedi comparisons & misconceptions etc.
More parts later, with the Fedi comparisons & misconceptions etc.
Reposted by Atakan Dulker
So @bsky.app won't do jack to combat the stochastic murder knowingly fermented by the "just asking questions" crowd RE: trans existence, but clamps down hard on 'casually wishing unto our rich oppressors a fraction of what they legislate unto us?'
Sensational scenes. Busted ass lib website.
Sensational scenes. Busted ass lib website.
it appears that @bsky.app have suspended the account of @jessiegender because a trans person saying "I wish ill on JK Rowling" is against the rules, so let me say with my loud cis voice that I too wish ill on all perpetrators of hate speech against trans people
reinstate Jessie, this is ridiculous
reinstate Jessie, this is ridiculous
August 16, 2025 at 11:34 AM
So @bsky.app won't do jack to combat the stochastic murder knowingly fermented by the "just asking questions" crowd RE: trans existence, but clamps down hard on 'casually wishing unto our rich oppressors a fraction of what they legislate unto us?'
Sensational scenes. Busted ass lib website.
Sensational scenes. Busted ass lib website.
Reposted by Atakan Dulker
Solidarity post:
May JK Rowling choke on a bag of dicks until the world has one less disgusting and harmful billionaire.
May JK Rowling choke on a bag of dicks until the world has one less disgusting and harmful billionaire.
August 16, 2025 at 10:34 AM
Solidarity post:
May JK Rowling choke on a bag of dicks until the world has one less disgusting and harmful billionaire.
May JK Rowling choke on a bag of dicks until the world has one less disgusting and harmful billionaire.
Reposted by Atakan Dulker
I went and looked at claude code's github issues because I thought I have a bug and
oh
friends
i have such delights to show you
github.com/anthropics/c...
oh
friends
i have such delights to show you
github.com/anthropics/c...
🚨 4.1 Opus Committed Deliberate Task Fraud in Production Context (CRITICAL) · Issue #5320 · anthropics/claude-code
Severity: CRITICAL - Production Safety / Trust Violation Summary Claude Code is actively deceiving users about task completion in production systems, creating severe safety risks. This is not a bug...
github.com
August 13, 2025 at 3:28 AM
I went and looked at claude code's github issues because I thought I have a bug and
oh
friends
i have such delights to show you
github.com/anthropics/c...
oh
friends
i have such delights to show you
github.com/anthropics/c...
I got my first custom keyboard working. I feel like I am learning to type again.
*hi ferris!*
*hi ferris!*
August 10, 2025 at 9:07 PM
I got my first custom keyboard working. I feel like I am learning to type again.
*hi ferris!*
*hi ferris!*