Taras Glek
taras.glek.net
Taras Glek
@taras.glek.net
LLMs, system programming

https://taras.glek.net
I love the idea of my JBL Link display cos it has awesome sound and the voice interaction features from google home...but I absolutely hate how the device gets worse over time. As a result been thinking through and prototyping a simpler solution for past few years... Hate-driven development
January 12, 2026 at 7:19 AM
next time i have free time gonna try this mouse thing github.com/moverest/wl-...
GitHub - moverest/wl-kbptr: Control the mouse pointer with the keyboard on Wayland.
Control the mouse pointer with the keyboard on Wayland. - moverest/wl-kbptr
github.com
January 6, 2026 at 10:21 AM
Reposted by Taras Glek
wifitui v0.10.0 is out (fast friendly featureful wifi tui for linux):

- added support for displaying multiple access points
- improved experimental macOS support (thanks @jimst.bsky.social)
- better key handling (hjkl), and more!

github.com/shazow/wifit...
January 5, 2026 at 2:37 PM
Reposted by Taras Glek
new year, new bloggies: on pre-tenuring in v8 wingolog.org/archives/202...
pre-tenuring in v8 — wingolog
wingolog: article: pre-tenuring in v8
wingolog.org
January 5, 2026 at 3:39 PM
This is the crazyest storage vps i have seen. zfs.rent/pricing.html
BYD. Seems like great value
zfs.rent // pricing
$10/month/drive; Specs: 1 vCPU + 2 GB RAM + 25 GB NVMe. Resources scale to number of drives hosted.
zfs.rent
January 5, 2026 at 6:54 PM
always felt that i didnt know touch typing. Turns out I didnt. github.com/aradzie/keyb... seems very helpful
github.com
January 3, 2026 at 6:40 PM
Reposted by Taras Glek
Wanted a Dabao but couldn't get one at #39C3? Sign up here to be notified when they are ready for sale: www.crowdsupply.com/baochip/dabao
Dabao Evaluation Board for Baochip-1x
A powerful new RISC-V microcontroller with mostly open RTL
www.crowdsupply.com
December 30, 2025 at 7:22 PM
i love that these guys went against "no exotic hw" values in duckdb.
The cost comparison is bit invalid, they shouldve costed nvidia instances on aws,
December 31, 2025 at 9:41 PM
Reposted by Taras Glek
another robot highlight for 2025: man wearing humanoid mocap suit kicks himself in the balls
December 27, 2025 at 5:27 PM
Reposted by Taras Glek
Low power waveguides! What a wonderful idea.
December 28, 2025 at 2:18 AM
Reposted by Taras Glek
At the gpg.fail talk and omg #39c3

You can just put a \0 in the Hash: header and then newlines and inject text in a cleartext message.

Won’t even blame PGP here. C is unsafe at any speed.

gpg has not fixed it yet.
December 27, 2025 at 4:31 PM
Reposted by Taras Glek
joining the "cybertruck owners only" group on fb is one of the best decisions i've ever made. every so often when i'm scrolling through fb getting pissed off, i get to see a little treat like this
December 26, 2025 at 4:11 PM
There is a lot of talk of LLM productivity enhancements but very few examples of useful software that would not have existed if it wasn't for LLMs. I have a small list of software that was only embarked upon due to exploration and productivity boosts that LLMs enable
December 24, 2025 at 9:35 AM
Great overview of passkeys and an interesting hosted service enzom.dev/b/passkeys/
Things I learnt about passkeys when building passkeybot
Interesting things about passkeys in webauthn
enzom.dev
December 23, 2025 at 6:59 AM
this is the best post on what llms are good for and how most of the marketing is wrong chrisvoncsefalvay.com/posts/dorkes...
Dorkestration – Chris von Csefalvay
Why everybody missed the point about tool use in agentic AI, and how a handful of primitives can orchestrate your entire ML workflow.
chrisvoncsefalvay.com
December 21, 2025 at 7:34 PM
Reposted by Taras Glek
Resurrecting some of my old Raspberry Pi projects that had fallen into disrepair, annoyed at having to look up pinouts and count pins all the time when putting Humpty back together again,

Me: "yo, ChatGPT, make me some OpenSCAD of this thingy covering all but groud, SPIO clock+data."

Chat: ....

🤘
December 19, 2025 at 9:08 PM
Reposted by Taras Glek
Do you use LLMs to generate regular expressions? We do, too! Do you *review* your regexes? Is that frustrating? How can we put humans in the loop better, doing relatively few, meaningful tasks? Please try out our new tool PICK:regex, available for VSCode!
blog.brownplt.org/2025/12/11/p...
LLMs ⭢ Regular Expressions, Responsibly!
blog.brownplt.org
December 17, 2025 at 5:38 PM
This time is different! Right?
December 16, 2025 at 8:51 PM
Not endorsing TOON but interesting that llm comprehension of a data format is now a benchmark to consider toonformat.dev/guide/benchm...
TOON
Token-Oriented Object Notation (TOON) – Compact, human-readable, schema-aware encoding of JSON for LLM prompts
toonformat.dev
December 15, 2025 at 8:22 AM
Reposted by Taras Glek
So Amazon buying iRobot is a competition problem, according to Lina Khan. A Chinese company "Shenzhen PICEA" acquiring it, however, is apparently no issue at all.

This reflects a deeply misguided approach to antitrust enforcement, under Lina Khan.

www.bloomberg.com/news/article...
Robot Vacuum Roomba Maker Files for Bankruptcy After 35 Years
iRobot Corp., the company that revolutionized robot vacuum cleaners in the early 2000s with its Roomba model, filed for bankruptcy and proposed handing over control to its main Chinese supplier.
www.bloomberg.com
December 15, 2025 at 5:23 AM
This seems like a neat attempt to teach higher level programming concepts rather than any specific language implementation blog.brownplt.org/2024/04/12/b...
Finding and Fixing Standard Misconceptions About Program Behavior
blog.brownplt.org
December 14, 2025 at 9:16 AM
@anil.recoil.org i am enjoying your aoah posts. Curious what your workflow is for checkpointing and undoing claude code work when it goes wrong
I find it infuriating that coding agents do not use git to differentiate between human and llm code. Perhaps I am just using em wrong?
December 14, 2025 at 9:07 AM
Never really enjoyed working with minio. Looks like nvidia has a higher performing alternative (also in go). Minio never supported erasure codes in open source or had good performance or kept objects accessible as files. Nvidia aistore seems to do all that better github.com/NVIDIA/aistore
GitHub - NVIDIA/aistore: AIStore: scalable storage for AI applications
AIStore: scalable storage for AI applications. Contribute to NVIDIA/aistore development by creating an account on GitHub.
github.com
December 14, 2025 at 8:43 AM
This seems like it could become a nice duckdb component for use in various data analysis webpages. cc @humphd.bsky.social
December 14, 2025 at 8:32 AM