Llm

This is actually bit unscheduled, but as I feel the content will not age gracefully, I chose to write what I experienced with my different LLM model experiments recently. In general, I value local compute over cloud compute, and especially if things of private nature are involved. However, sometimes local compute simply does not work. I have dabbled with different models as main model for the Hermes Agent: GPT 5.6 Sol (using ChatGPT Plus subscription) GPT 5.6 Luna (-//-) Qwen 3.6 35B A3B q4 mtp and here is a brief write-up about it. This picture unfortunately spoils the outcome: ...

In the previous episode - Me and my agent, I had just barely installed Hermes agent. Now, two weeks later, more reporting on my agentic experiment. Short version: I still feel that it is worth it and useful. I have added only one more recurring event to it (crawling my LinkedIn feed for notifications about people I am actually directly linked to), but it has pretty much replaced my all other search engines (and AI chats), as I like having history of the conversations, as well as potentially having them be bit more private down the road. The search experience is also superior to anything else commercial. Perplexity used to be nice, but I never got around to paying for it, and the free experience tanked badly this year; Claude/ChatGPT usually don’t reveal their sources consistently which makes them worthless to me as search engines as sometimes the LLM does make mistakes. ...

I have been bemusedly watching OpenClaw madness this year (well, Clawdbot start of January, Moltbot on the 27th of January, and OpenClaw after 30th of January). I have not really seen the point - while I see the temptation of telling computer to ‘do stuff’, my security background has led me not to trust computer with anything actually dangerous (e.g. work/financial/private stuff). Friday: The start seemed harmless enough, and useless.. Hermes Agent was released in March this year, and it seems to some extent bit more productized and sane than OpenClaw, but I am not really sure. So from the outset, I decided not to really go for the crazy ‘give it access to my computer and let Jesus take the wheel’, but instead this weekend I gave in finally to the dark side, and started Friday by installing Hermes Agent in its own container. ...

I have been using recently Anthropic’s (Claude) Fable and Opus at work, and OpenAI’s GPT for hobby projects. This is not about that, though, but instead about how my use of the Internet search engines has evolved over time. I hear there are number of people who use only chatbots for searching the internet these days. While it is actually very tempting (one chat window, all your questions answered), in practice it does not work out too well. ...

I have been doing software development for decades. Yet I have been never as optimistic about the future of my chosen career as I am now. I am starting this just from point of view of programming languages, for reasons I describe later.. My somewhat ‘played with or used for stuff’ programming language list is at 32 languages at the moment, but I am just covering about main ones to convey a point. ...

Background I have been working on number of vibe coded apps recently after I gave up on my most recent startup, and before starting to work somewhere elsewhere. So I have done lots of projects I have never had the time for before. HomeNetFlow Lixie’s first iteration that I wrote by hand in 2024 ( see Observability at home ) was and is still useful. It is still only about categorizing log lines, by hand, and then having those rules applied to logs at scale using vector. My home infrastructure uses rules generated by it still, and I look at the logs quite often (that are filtered based on those rules). ...

I have been ‘enjoying’ some awkward moments with default sandboxing of Codex CLI, Claude Code, and (lack of it in) OpenCode. I settled on a single sandboxing solution, and it is nono - Next-Generation Agent Security. The problem All coding agents use current shell environment to launch the tools they use. Those with sandboxing have rules which are set up to prevent ‘invalid’ use, but in practice they break often and are painful (or not possible) to configure properly per tool. ...

Updated 11.4.2024: Added second day case too without using fast mode. There has been a lot of brouhaha in e.g. Reddit about recent (10.4.2026) changes to plans: OpenAI announced new 100€ ‘pro lite’ tier, and there was worry that ‘plus’ tier was cannibalized for it. So I decided to use my 5 hour quota using fast mode gpt-5.4 (high for planning, medium for implementation), and then subsequently second time without fast mode, same model and thinking parameters. ...

I started converting some of my hobby projects using Codex to Go exactly month ago, and developing two iOS apps after initial honeymoon period with Codex CLI was done (I felt it to be good enough in the trial conversion effort). Here’s some notes of my experiences and a bonus rant near the end.. My thoughts about Codex CLI and OpenAI models in general Overall it has been pretty good. While they are clearly iterating on how e.g. sub-agents work, the basic flow is pretty robust. It is annoying that you cannot use separate model for planning and build stages though - just different thinking levels. I have wound up using ‘high’ for planning and ‘medium’ for implementation for most part, most recently using GPT 5.4. ...

After brew upgrade (and GPT 5.4 announcement last night) I was excited to see what was up. Interesting. Try 1 I did one minor feature using planning mode (gpt-5.4 high) and implemented it using gpt-5.3 codex spark. Codex spark is backed by super-fast Cerebras megachips, so I was optimistic about the speed (if not quality). Observations: REALLY fast (couple of seconds to implement something that was couple of hundred lines long) Bit stupid, or literal? It did not run unit tests by default Unit tests had an error, which it fixed It had used 62% of its context by this time(!) - seems like order of magnitude less context than 5.3 Codex Try 2 Without planning mode, on high, I asked it to refactor logging in the app I am working on. It actually did what was asked for, but again did not run tests as I did not specifically ask for it to run them too. The outcome was also pretty ugly as it did not look at the API surface of the logging library (or know it), so I had to send gpt-5.4 after it to clean up. ...