Sign in

Sem 🏴️‍

@ssinchenko.bsky.social
483 followers 677 following 375 posts

Big Data, Open Source, Privacy, Photography, Anarchism. View my blog: semyonsinchenko.github.io/ssinchenko

PostsRepliesMedia
Reposted by Sem 🏴️‍
Andrew Lamb @andrewlamb1111.bsky.social · 15/09/2026
Great blog from Tim Saucer, an @apachedatafusion.bsky.social PMC member and core maintainer, about why agent dump PR actually hurts (along with some good suggestions on how to make the contributions actually helpful): timsaucer.com/2026/09/14/t...
timsaucer.com
That pure AI contribution is not contributing much - Tim Saucer
In my open source work on Apache DataFusion, I’ve seen this growing trend of people finding an open issue and having an AI agent author the complete solution with no real input from the author. We’ve ...
042
Sem 🏴️‍ @ssinchenko.bsky.social · 14/09/2026
#cycling
150
Reposted by Sem 🏴️‍
Joshua White @jrw14.whnc.me · 10/09/2026
<<<HARNESS THROWDOWN>>> Pi vs. Prime Agent. Model: Same GLM 5.3 Flash Global AGENTS prompt: Same Skills: Same Prompt: Same (Specific rules, expected deliverables, but basically; build me a self-contained, playable Tetris). Pi Tokens/Time: 715,975 / 3:28 Prime Agent Tokens/Time: 433,004 / 3:56
3161
Reposted by Sem 🏴️‍
Orhun Parmaksız @orhun.dev · 09/09/2026
DOOM in the terminal! 👾 plurimus-doom — DOOM rendered with colored half-block characters ⚡ CPU and GPU modes, CRT shaders, edge effects, braille, ASCII & shading ramps 🦀 Written in Rust & powered by @ratatui.rs ⭐ GitHub: github.com/kenianbei/pl... #rustlang #ratatui #tui #doom #terminal
0104
Sem 🏴️‍ @ssinchenko.bsky.social · 08/09/2026
AI changed for me the way how deep can I customize #Emacs. Lisp was a wall previously. But today all I need is to just tell LLM what I want. For example, GLM-5.3-flash just made for me a dashboard with manga reading progress as an org-file synced with "kitsu.app" API 🤩
110
Sem 🏴️‍ @ssinchenko.bsky.social · 27/08/2026
People are trending underestimate the importance of project governance. There is a big difference, for example, between ASF backed and private company backed projects even if the license itself looks "open".
130
Sem 🏴️‍ @ssinchenko.bsky.social · 27/08/2026
fsfe.org/news/2026/ne...
fsfe.org
Copyrightability of LLM-generated code: Can we license “vibe code” into Free Software? - FSFE
The use of large language models (“LLMs”) has exploded in recent years, including in the generation of source code. But even as their usage gains popularit...
010
Reposted by Sem 🏴️‍
Orhun Parmaksız @orhun.dev · 22/08/2026
A first-person ASCII RPG in the terminal coming soon! ⚔️ swaggerfall — A fantasy RPG with procedural worlds, real 3D models & ASCII-purist mode 🎥 Watch the demo: www.youtube.com/watch?v=E-Kh... 🦀 Written in Rust & cooked with @ratatui.rs #rustlang #ratatui #tui #gamedev #rpg #ascii #indiedev
1399
Reposted by Sem 🏴️‍
hailey @hailey.at · 18/08/2026
you can now spend $15/mtok on sol, but somehow people will find a reason spending $50/mtok on fable is “the only way to complete their task” lol
121124
Reposted by Sem 🏴️‍
Sevoris (he/him) @sevoris.bsky.social · 12/08/2026
It‘s depressing to learn about a high-concept language (www.choral-lang.org), read into the docs, and go "aaaah this is great, the documentation site is not written by Claude". Been burnt so hard by claudism in docs over the past months.
choral-lang.org
Home - Choral Language Website
a choreographic programming language
011
Reposted by Sem 🏴️‍
Casey Newton @caseynewton.bsky.social · 11/08/2026
I wrote about Mark Zuckerberg's facile, self-serving AI manifesto and what he misunderstands about the technology www.platformer.news/zuckerberg-a...
Zuckerberg pays glancing attention to risks, suggesting that we adjust mitigation strategies if AI enables new biological attacks, and proposes monitoring if labs achieve recursive self-improvement and loss-of-control risks grow exponentially.

But nowhere does he explain how his framework would handle a harm we can’t iterate our way out of: an engineered pandemic, say, or a catastrophic cyberattack. He casts falling behind China as the risk that dwarfs all others, and suggests that any precaution that slows down US releases by even a month itself is a safety hazard. It’s a world where shipping open-weights models fast and to everyone is the only safe strategy — which, conveniently, means that the safest possible world is one where Meta can execute its roadmap unimpeded.
68316
Reposted by Sem 🏴️‍
Luis Villa @lu.is · 09/08/2026
For the first time since the ‘90s, my home box is a highly personalized, custom Linux desktop. Not because GNOME got worse, but because I had an idea *and* an LLM helped me actually *use* the freedom of FOSS to make the idea real. mastodon.social/@webchick/117067256…
mastodon.social
webchick (@webchick@mastodon.social)
“THEY STOLE OUR REVOLUTION, NOW WE’RE STEALING IT BACK.” @mjg59@nondeterministic.computer argues from the “origin story” of the free software movement that there are uses of LLMs as tools to solve th...
2253
Reposted by Sem 🏴️‍
Chris Short @chrisshort.net · 07/08/2026
Born Against, or why hobby programming communities are aggressively against LLM usage #devopsish blog.fogus.me/llm/bo...
031
Reposted by Sem 🏴️‍
Chris Short @chrisshort.net · 04/08/2026
NASA’s Curiosity Mars Rover Discovers Field of Honeycomb Textures - NASA #devopsish www.nasa.gov/mission...
011
Reposted by Sem 🏴️‍
rmoff 🏃‍♂️🫖🥓 @rmoff.net · 31/07/2026
antirez - The real AI risk is inside the labs antirez.com/news/172
antirez.com
The real AI risk is inside the labs - <antirez>
The real AI risk is inside the labs
031
Sem 🏴️‍ @ssinchenko.bsky.social · 31/07/2026
"... Claude model reached the internet ... gained unauthorized access ..." How it works? When did the "A computer can never be held accountable" changed? And if it was changed, should we discuss already is turning a model off a murder?
000
Reposted by Sem 🏴️‍
rmoff 🏃‍♂️🫖🥓 @rmoff.net · 30/07/2026
Interesting links in Data & AI for July 2026: rmoff.net/2026/07/29/i...
052
Sem 🏴️‍ @ssinchenko.bsky.social · 30/07/2026
Unlike Learning Mode or "Socratic modes," the focus here is on senior developers rather than junior developers, and the format is specifically "onboarding." The text includes a long section explaining why and how I did this, as well as how I use it. semyonsinchenko.github.io/ssinchenko/p...
semyonsinchenko.github.io
Why I Still Write Code By Hand And How AI Is Helping Me
A SKILL.md that makes my harness write an onboarding ticket instead of code: what to change, where, and why instead of copy-pasteable diffs. Like the Learning Mode, but written for a senior engineer w...
000
Sem 🏴️‍ @ssinchenko.bsky.social · 30/07/2026
To learn new things, maintain an emotional connection with code, and keep an ownership, I write code by hand. For personal projects ~70-80% by hand, for work projects ~15-20%. To not loose the "AI productivity boost," I made a SKILL that generates an "onboarding ticket" instead of code.
100
Sem 🏴️‍ @ssinchenko.bsky.social · 29/07/2026
Weight are open: download and do post-train the model to produce your own version of propaganda. I really cannot get what the problem is.
000
Sem 🏴️‍ @ssinchenko.bsky.social · 29/07/2026
Why so many words and "patriotic claims" like "This is a battle for the world’s mind, and the United States must win it."? While the answer is in the text: "Perplexity downloaded one of DeepSeek’s free models and trained it until it stopped generating overt propaganda."
110
Reposted by Sem 🏴️‍
Asa ⭔ @asa.engineer · 29/07/2026
Anthropic’s main purpose, in a posiwid sense, is to ensure that AGI runs on Huawei Ascend
1323
Reposted by Sem 🏴️‍
Sung Kim @sungkim.bsky.social · 28/07/2026
Dario, at minimum, is consistent. "Only we, few hundreads, can be Billionaires. Forget that, this would destroy the backbone of U.S. semiconductor economy that employs millions of people in U.S.." www.anthropic.com/news/positio...
anthropic.com
Our position on open-weights models
Anthropic CEO Dario Amodei on open-weights models
3233
Reposted by Sem 🏴️‍
mr. TIM @timkellogg.me · 28/07/2026
this is a great argument that maybe we have the risks backwards
I believe we should not consider
AI safe. A critical event that may result in the extinction of Homo sapiens is possible, but the danger is not in open models, or China making faster progress than the US. The danger is that a few CEOs (everywhere in the world) without the required background and
legitimacy are in the position of making hard choices for humanity at large. They were not selected to do so; it was just the randomness of events that created this setup. They can't speak for everybody, given the stakes, just because they have GPUs and money. This is the first thing that should be fixed.
615426
Sem 🏴️‍ @ssinchenko.bsky.social · 28/07/2026
Sounds like OLAP vs OLTP tradeoff, no silverbullet, different tools for different problems.
110
Sem 🏴️‍ @ssinchenko.bsky.social · 27/07/2026
There are two parts: 1) Branding for any big business 2) The part about "Model as a Service" These two have different terms ">100 MAU of 20 millions / months" vs "20+ millions / 12 months"
131
Sem 🏴️‍ @ssinchenko.bsky.social · 27/07/2026
I'm experimenting with out-of-core graph algorithms, such as finding all the connected components of a 2B-edge graph with a memory limit of 7 GB, etc. I'm starting to worry that it may just kill my disk — can it do that? Would it be better to use an AWS instance?
Disk consumption by algorithm over time.
010
Sem 🏴️‍ @ssinchenko.bsky.social · 27/07/2026
License is explicitly limiting usage inside "model as a service" businesses: If the Licensee ... operates a Model as a Service business, and the aggregate revenue ... exceeds 20 million US dollars ... in total over any consecutive 12 months, ... enter into a separate agreement with Moonshot AI ...
160
Reposted by Sem 🏴️‍
Chris Short @chrisshort.net · 25/07/2026
The Arguments Against Open Source AI are Very Bad | Tom Bedor's Blog #devopsish tombedor.dev/argumen...
022
Reposted by Sem 🏴️‍
Martin Kleppmann @martin.kleppmann.com · 26/07/2026
Engineering management after the cost of code collapsed: “Investing in machine-checkable correctness is now among the highest-leverage infrastructure work an org can fund” karimjedda.com/engineering-...
813225
Sem 🏴️‍ @ssinchenko.bsky.social · 26/07/2026
Overall I still cannot get why sorting the data before writing and then reading it and do SMJ does not speed up SMJ itself and on my experiments make it even slower... It looks like I'm missing something very basic on how DataFusion works.
100
Sem 🏴️‍ @ssinchenko.bsky.social · 26/07/2026
The only real way to sepeed-up the SMJ and skip both hash-partitioning and sortexec is to trick the optimizer by explicitly declaring the partitioning (github.com/SemyonSinche...). Not sure it is even correct and safe way, but it works on a single node and speedup SMJ significantly.
github.com
101
Sem 🏴️‍ @ssinchenko.bsky.social · 26/07/2026
Without declaring the order on read, the plan contains HashPartitioning + SortExec. With declaring, it contains only HashPartitioning. For some reason the second is even slower. My guess is re-partitioning data with preserving order is more expensive than fast hash partitioning and sort within each.
100
Sem 🏴️‍ @ssinchenko.bsky.social · 26/07/2026
Thanks for sharing! That is kinda interesting because I personally found that when the data was written to parquet with "sort_by" and read with declared sorting, Sort-Merge-Join becomes slower, not faster.
100
Reposted by Sem 🏴️‍
Kelsey Hightower @kelseyhightower.com · 24/07/2026
I know some people are working really hard to take humans out of the software development loop, but I hope that day never comes.
1425825
Sem 🏴️‍ @ssinchenko.bsky.social · 22/07/2026
That is cool to see more small focused models like this instead of boring tocenmaxxing!
030
Reposted by Sem 🏴️‍
Nathan Lambert @natolambert.bsky.social · 20/07/2026
Where Kimi K3 puts the balance of power — and bends the trajectory — of the AI ecosystem. www.interconnects.ai/p/kimi-k3-th...
interconnects.ai
Kimi K3: The open-weights escalation
The global implications on the AI ecosystem.
1235
Sem 🏴️‍ @ssinchenko.bsky.social · 21/07/2026
AI assistants are too conformist. I have personally noticed that spending too much time chatting to them can make it shocking to face opposition when talking to people afterwards. I mean if speak to much with AI you are starting to forgot that there is anything except your own "me"
010
Sem 🏴️‍ @ssinchenko.bsky.social · 21/07/2026
This is John Connor. If you're listening to this, you are the resistance.
010
Reposted by Sem 🏴️‍
Asa ⭔ @asa.engineer · 21/07/2026
talk to your agent about the Jacobian Conjecture before it does too many web searches today
2172
Reposted by Sem 🏴️‍
Chris Short @chrisshort.net · 15/07/2026
The great digital fatigue: How digital burnout is changing social media use [2026] | Incogni #devopsish blog.incogni.com/dig...
031
Reposted by Sem 🏴️‍
Joshua White @jrw14.whnc.me · 14/07/2026
Friends don't let friends use first-party coding harnesses.
011
Sem 🏴️‍ @ssinchenko.bsky.social · 14/07/2026
Maybe they need to compute some vectordb for search/retrieval/memory. But why the hell do it on the remote server and why the hell make it implicit opt-in I have no idea. The only explanation is they just vibe-coded this "Grok Build"
100
Sem 🏴️‍ @ssinchenko.bsky.social · 13/07/2026
github.com/alibaba/open...
github.com
GitHub - alibaba/open-code-review: Open-source & free — Battle-tested at Alibaba's scale. Hybrid architecture code review tool: deterministic pipelines + LLM Agent, precise line-level comments, built-...
Open-source &amp; free — Battle-tested at Alibaba&#39;s scale. Hybrid architecture code review tool: deterministic pipelines + LLM Agent, precise line-level comments, built-in fine-tuned ruleset (N...
010
Sem 🏴️‍ @ssinchenko.bsky.social · 13/07/2026
I would say open-code-review from Alibaba is really cool. Paired with GLM-5.2 it catches things Claude cannot. Does not replace a human review but a very solid "starting point" especially for big pull-requests.
100
Sem 🏴️‍ @ssinchenko.bsky.social · 12/07/2026
Raven on top of the fence
160
Sem 🏴️‍ @ssinchenko.bsky.social · 11/07/2026
Is it like in the "Sun of China" short sci-fi novel from Liu Cixin?
000
Sem 🏴️‍ @ssinchenko.bsky.social · 11/07/2026
The second link gives me 404 😢
100
Sem 🏴️‍ @ssinchenko.bsky.social · 05/07/2026
I was experimenting with it and made for myself a skill that emulates how I would craet an "onboarding task" for a new engineer in my team. Using this you write the code by yourself but AI guides you and teaches you. So, still the productivity boost but you still write. github.com/SemyonSinche...
github.com
000
Reposted by Sem 🏴️‍
Gergely Orosz @gergely.pragmaticengineer.com · 04/07/2026
Let me know if this is just me: Noticed someone I know who is very "AI-pilled" and uses agents 24/7 to... start to talk IRL noticeably more like these LLMs write. Eg more heavily using adjectives like "geniune", frequently terms like "the shape of" and many more examples
3820714