@wergieluk.bsky.social
There is this beautiful infinite basement called Github history, plenty of room for all your unused code, successful and failed experiments, deprecated features, and all the other stuff that has no good reason to exist in the main branch of the repo.
October 8, 2025 at 10:00 AM
There is this beautiful infinite basement called Github history, plenty of room for all your unused code, successful and failed experiments, deprecated features, and all the other stuff that has no good reason to exist in the main branch of the repo.
Each time somebody complains that model X doesn't work for Y, I show them my model selection list.
There is no need to figure out which model is good for what.
You can just ask the same question multiple times, compare answers, summarize, etc.
There is no need to figure out which model is good for what.
You can just ask the same question multiple times, compare answers, summarize, etc.
September 22, 2025 at 10:01 AM
Each time somebody complains that model X doesn't work for Y, I show them my model selection list.
There is no need to figure out which model is good for what.
You can just ask the same question multiple times, compare answers, summarize, etc.
There is no need to figure out which model is good for what.
You can just ask the same question multiple times, compare answers, summarize, etc.
If predicting the future is too difficult:
simulate scenarios
Then optimize your KPIs over all those scenarios.
simulate scenarios
Then optimize your KPIs over all those scenarios.
August 4, 2025 at 10:00 AM
If predicting the future is too difficult:
simulate scenarios
Then optimize your KPIs over all those scenarios.
simulate scenarios
Then optimize your KPIs over all those scenarios.
Both in logging and in propaganda, a reliable way to overwhelm your audience and obscure information is to flood the channel with irrelevant messages.
July 30, 2025 at 8:01 PM
Both in logging and in propaganda, a reliable way to overwhelm your audience and obscure information is to flood the channel with irrelevant messages.
LLMs are not designed for time-series forecasting, and they do not excel at it.
Instead of forcing a round peg into a square hole, use a hybrid AI agent approach.
Instead of forcing a round peg into a square hole, use a hybrid AI agent approach.
July 30, 2025 at 3:04 PM
LLMs are not designed for time-series forecasting, and they do not excel at it.
Instead of forcing a round peg into a square hole, use a hybrid AI agent approach.
Instead of forcing a round peg into a square hole, use a hybrid AI agent approach.
The Hugging Face Daily Papers team is brewing up something incredibly useful! #huggingface
March 26, 2025 at 11:04 AM
The Hugging Face Daily Papers team is brewing up something incredibly useful! #huggingface
The X11 selection clipboard rocks, but not all apps are on board.
March 25, 2025 at 5:39 PM
The X11 selection clipboard rocks, but not all apps are on board.
I keep coming back to this classic sci-fi book by Stanisław Lem. In terms of our current model capabilities, we are not quite there yet.
March 19, 2025 at 11:03 AM
I keep coming back to this classic sci-fi book by Stanisław Lem. In terms of our current model capabilities, we are not quite there yet.
Productivity restored
March 18, 2025 at 4:01 PM
Productivity restored
Every time a part of my dev setup/infra breaks and I can't fix it immediately, I try to push myself to try out an alternative.
Today it's `aichat` -- a surprisingly usable LLM CLI (for Linux). Added benefit: reduced context switches between dev env and browser.
Today it's `aichat` -- a surprisingly usable LLM CLI (for Linux). Added benefit: reduced context switches between dev env and browser.
March 17, 2025 at 4:03 PM
Every time a part of my dev setup/infra breaks and I can't fix it immediately, I try to push myself to try out an alternative.
Today it's `aichat` -- a surprisingly usable LLM CLI (for Linux). Added benefit: reduced context switches between dev env and browser.
Today it's `aichat` -- a surprisingly usable LLM CLI (for Linux). Added benefit: reduced context switches between dev env and browser.
pacman is an infinite source of useful software
March 17, 2025 at 1:52 PM
pacman is an infinite source of useful software
Here we go.. 😱
March 10, 2025 at 2:24 PM
Here we go.. 😱
DigitalOcean makes it super easy to deploy Streamlit apps. Basically, you just give them your GitHub repo URL and your credit card number. 😅
February 25, 2025 at 5:33 PM
DigitalOcean makes it super easy to deploy Streamlit apps. Basically, you just give them your GitHub repo URL and your credit card number. 😅
2 vCPU 4GiB RAM machine:
DigitalOcean App Plat.: $50/m
Hetzner: $3.85/m
That's a hefty convenience premium
DigitalOcean App Plat.: $50/m
Hetzner: $3.85/m
That's a hefty convenience premium
February 18, 2025 at 3:51 PM
2 vCPU 4GiB RAM machine:
DigitalOcean App Plat.: $50/m
Hetzner: $3.85/m
That's a hefty convenience premium
DigitalOcean App Plat.: $50/m
Hetzner: $3.85/m
That's a hefty convenience premium
There are around 100k Github repositories with more than 500 stars.
February 17, 2025 at 10:08 PM
There are around 100k Github repositories with more than 500 stars.
Difference betw LN Career and Business: $30/mth, 10 inmails, and a button. Sweet.
January 27, 2025 at 11:00 AM
Difference betw LN Career and Business: $30/mth, 10 inmails, and a button. Sweet.