It’s not obvious how to get good results when trying to use LLMs for technical task. After many long discussions online I’ve come to realize that there’s a lack of proper introduction guides. Here’s mine.
Is the LLM smart or are you not?
A look at how we might overestimate our own complexity and how psychological hacks like the Anchoring Effect mirror the way we interact with LLMs.
Nginx Rate Limiting: Taming the Meta-Crawler
The bot scrapers use many different source IPs to get around rate limits. If you use Nginx as your reverse proxy, it’s easy to counter them by assigning their whole IP-ranges to single rate-limited buckets.
Reverse engineering proprietary algorithms can be such a chore. So why not outsource that tedious work to the machines? Here’s how “I” reverse engineered an ECC algorithm for Winbond NANDs.
Swedish Speech-To-Text in OpenWebUI
It’s far from obvious how to use OpenWebUI with another Whisper model than the defaults from Systran. I wanted the Speech-To-Text to use the models from the Swedish Royal Library - and found a way.
When you get enough complexity, predictability goes down. LLMs are showing emergent behavior difficult to gauge from dismissing them as being autocompletes on steroids.
How to use Mistral Vibe as a development companion that remembers the context of the different projects you work on
Vibe coding is the name given to just prompting an LLM to implement according to your directions. In this post I detail what my experience is after having done just that, viewed from the point of a senior Software Engineer and Cybersecurity expert.
Corporate mandates like "return to the office" and "AI integration" are driven not by productivity or efficiency, but by financial interests tied to real estate and IT investments
The Coming Cognitive Disbelief
Both humans and large language models (LLMs) are fundamentally statistical pattern-matching systems with no inherent consciousness or "magic"