A classic mystery submission from Australia: a man asked an OpenClaw agent to book him into a gym class. It booked classes weeks further ahead than the gym policy allowed, then canceled a stranger’s reservation to move him up the waitlist. Every outlet ran it as an AI going rogue. Almost none noted this has always been doable with wetware. All it ever needed was intent.
Tagged LLM
App to Organize Coding Agent Output
As much as I rely on Claude Code while managing home lab and writing code for DIY projects, coding agents never meet the media’s perception of AI: “slaves to logic” as I, Robot has put it. Agents hallucinate, make mistakes and take them back, and more often than not, their output is a wall of text that is completely out of order. For Claude, the recent models print more than their predecessors.
Gemini 3.6 Flash Released
Over the last few weeks, my main takeaway with the Fable 5 and GPT-5.6 releases was their respective costs and cost performances. Neither company particularly hides the fact that the real models they offer are paywalled. It creates a distinct entry barrier for those who simply are not the target audience of the marketing materials surrounding the AI hype: vibe coding, content creation, and so on.
Fable 5 Now Included in Max Subscription
Fable 5 saga is finally coming to an end with Anthropic deciding to include the flagship model in Max. Since I am on Max tier for this month temporarily, the announcement is disappointing in two ways: a. Fable won’t be available on a similar price tier as GPT 5.6 Sol, and b. Fable is simply too inefficient when it comes to token counts compared to Sol. Whichever way you cut it, you can get a better deal from Codex.
Fable 5 Trial Extended
Fable 5’s trial period got an extension. If you had a chance to try it on an existing subscription, you already know whether it worked out for your actual project. There’s no stronger evidence than that. For me, Fable 5’s capability was amazing in its own right — but not at usage-based pricing.
Fable 5 Is Back, Also Sonnet 5 Is Out
Anthropic announced Sonnet 5 this week, described to be as capable as Opus 4.8 but cheaper. Rough timing — Fable 5’s return ate all the attention. I haven’t had the chance to do any coding with it in-depth yet. But from the limited exchange I had over chat, Sonnet 5 still struggled with technical specs. For any technical discussion, I had better luck with Gemini Flash 3.5. In terms of output, Sonnet became incredibly verbose. There is no trace of the laconicism I had noticed in Sonnet 4.
What Remains of Fable 5 Ban
Apple said Apple Intelligence would bring AI to its brand. It was a broken promise, unless Genmoji is to be considered a success. Apple is now saying the new Siri AI will deliver the promises they made back in 2024. Neither iOS 27 nor macOS Golden Gate has been officially released (available only in beta), but this definitely marks the beginning of the full transition toward AI-centrism in consumer OS. The dependency is real now — local or cloud, there’s no opting out.
App to Notify the Return of Fable 5
I thought I’d share some light-hearted news before I can drop the pessimistic read on the tech sector as a whole. Anthropic hasn’t yet shared any update on when the US government’s suspension of Fable 5 will be lifted, but there are communities voicing support for the repeal of the ban, while sharing memes around its status.
“Ethical Concern” is AI Dunce Cap
Storage is at an all-time premium nowadays, with even the price of SSDs and HDDs being affected by AI craze. As much as I would love to say ‘rip the discs raw’, there is bound to be a real world limit on how much storage I can secure for a self-hosted media server. My search for encoding these raws led me to looking for configurations used by others, and for what it’s worth, their activity could easily translate into online piracy. Claude shut down the conversation immediately, citing my query as equivalent to looking for a black-market arms dealer.
Are LLMs Terrible at Gauging Efficiency?
A theory that I had for sometime, originally not for LLMs, but for those who believe LLMs can write perfect codes. LLMs care little about the environment on which the codes will run on. I know the statement alone makes no sense — how can it be considered coding, if it doesn’t work on the environment? It’s not that it won’t run, it’s just that it will run extremely poorly.
LLM, Subscriptions, and Tokens
As I have previously said, I don’t see Claude replacing a proper programmer’s job anytime soon. I am subscribing to meager Pro tier, and that’s nowhere near enough to replace my hobbyist’s job. Granted, there is the joy of not needing to fiddle with technicalities. Being able to implement new algorithm and upgrade existing core algorithm to a new one like blocks falling into places feels like a vindication. But this is from a hobbyist’s perspective.
Claude Spiel: A Week with Claude Pro
It has been roughly a week, first week, on the paid subscription LLM service. I did hear good things about Claude, especially its versatility when it comes to answering general questions and coding. For the life of me, I was not satisfied with ChatGPT and Gemini when it came to getting helps on coding, — so much so vibe coding on those platforms would be risky — whereas Claude printed results like a bureaucrat.
How AI Hallucination Affects You
The hallucination in AI, or LLM hallucination, is real. I have read some op-eds and online discussions whether or not hallucinations affect real world use cases — average users making mundane queries. I was on the side of “garbage in, garbage out” camp; hallucination is an expected side effect of the algorithm, and it is up to the user to double check the answers. Now I am wondering if it merits double checking at all.
Coding with Gemini was Fun Short-Lived
During my migration from Pi to NAS via all things Docker, one of the problems I immediately faced was how I was going to handle my aging databases. All these eBooks, magazines, articles, and etc. I have accumulated since high school had to go somewhere. I know there would be some people shouting I should simply throw them out like the 90s memorabilia in the attic. But I took this on as my personal DIY project.
How to Change Ollama Model Path
I’m not particularly a big fan of having a big system storage. Most of my machines usually have 1 TB or less, and when needed, I use external drives or NAS. If you are interested in playing with Ollama, especially with the new Mac Studio, one of the missing key features of Ollama is to change path of where the LLM models will be saved or loaded from. As far as I could see, there were no verbs to do it on CLI or GUI.