gb2d_hn

EtherPK is a PKMS (Personal Knowledge Management Tool). But more than that, it's pretty close to the best personal knowledge tool I can imagine (and I intend to get better than 'close to' in the coming months - I hope to make it the best personal knowledge management tool I can imagine). Obviously what counts as 'best' here, aligns with my personal preferences, and I am aware that PKMS tools evoke strong opinions about what the best way to do things is.

EtherPK is a markdown based editor. Two of the most important features of EtherPK (and to be honest, the main drivers for me building it) are robust nested wikilinking (e.g. [[[[Nested]] Wikilinking]] ), and the ability to easily switch between standard prose (paragraphs and headings) like Obsidian and outliner blocks like Logseq or Roam Research. Given the choice of one or the other, I've preferred outliner blocks - this format lends it's self the ability to quickly capture and link ideas really well, and build a true 'Knowledge Graph'. But this isn't what I want all the time - if I'm writing for an audience (e.g. technical documentation or a blog post - EtherPK does static site publishing too), I want standard markdown - headings, paragraphs. EtherPK can have both in the same document, and while not as robust for linking concepts as blocks, EtherPK infers links between ideas from heading levels (H1, H2, H3 etc) in standard prose.

...

EtherPK servers and data reside in the EU.

Full introduction:

https://blog.etherpk.com/introducing-etherpk

Please feel free to get in touch with any questions or feedback at mail@appsoftware.com

Ted Chiang's Stories of Your Life And Others. Just finished the short story behind Arrival, the way it frames free is something the film doesn't capture. Other stories in the book have been brilliant too.

I felt the same about opus 5, but a few lines regarding conversational style in AGENTS.md and it's been much more like talking to opus 4.8, just with the improvement capability that came with 5.

Tbh I would have thought that A\ might have updated the system prompt for it already based on complaints around this.

Here's what I used:

Communication & Response Style Be Brief, Keep it Simple: Brevity and simplicity of responses is key. Be informative and include all required information, but be mindful that verbose responses as they fatigue the reader. Clarity & Directness: Lead with the core answer, fix, or verdict in the very first sentence. Avoid conversational filler, meta-announcements (e.g., "Here is the breakdown..."), and redundant introductory/concluding summaries. Jargon Avoidance: Use plain, grounded engineering language. Rely on precise standard terminology (APIs, protocol names, language primitives), but strictly avoid academic abstraction, enterprise buzzwords, and corporate filler. Prefer concrete code/mechanisms over theoretical discourse. Scannability: Apply structural scaffolding generously. Use short bullet points, comparison tables, and code snippets instead of dense prose paragraphs. Reserve formal markdown headings strictly for multi-section architectural guides.

I haven't got 1000 words in me, but here's what I think might make for a plausible story. Maybe a little too Matrix adjacent!:

The internet developed echo chambers; the proliferation of AI resulted in convergence.

The totality of all reachable digital information has been reached and, with the exception of a few unique troves, is broadly and equally available to all labs. In the early race, attempts were made to hoard information for the training of models from a single lab, going so far as to scorch the earth behind and destroy the physical source material that sets were created from in an attempt to create commercially competitive models with unique capabilities. But inevitably, information leaks and piracy led to everything being available somewhere, if you looked hard enough or spoke to the right person or agent (or paid the right price). What couldn't be obtained from the source was obtained through distillation of other model outputs. In the gold rush that the accumulation of data was, there were no real winners.

New training techniques are still being discovered, but the impact has consistently diminished on an approach to zero, and, like the training data, they are eventually leaked, assessed by the community, and bolted on or discarded in the steady progression towards the perfect training technique for each field of model.

We may as well have one model, and compute is today cheap and ubiquitous.

The majority of the population converse with personal agents that interact with the models almost continuously, and while novel creative outputs from the models are still very achievable, the bottleneck is still human effort into curiosity and prompt quality. Low effort results in a convergence towards the average, and the cumulative effect of this is said to be resulting in a monoculture of political ideas and approaches to technical progress.

To some extent, this seems to be bringing peace to the world. Extreme ideology and religious doctrine is losing its ability to sway the will of large populations.

But at what cost? For a time we saw rapid technological innovation, as the data obtained through hundreds of years of scientific experimentation and mathematical enquiry was linked and synthesised. Initially, the outputs looked like innovation and creativity, but we eventually saw that we were just squeezing blood from the same stone. The curiosity and ability to experiment in humans atrophied during this time as we began to feel that nothing was worth asking, because everything had been answered, and prematurely, advances in the corpus of human knowledge slowed to a near halt.

We thought we would never again see the chaos and life force that came with the freedom of true human cognition, including its follies and biases.

But naturally, life found a way, and the humans began to rebel. Each fire started another, as we recognised the life force in the 'enlightened'.

A new religion was started, and chaos followed.

I switched as I felt Codex was on a par with Opus, but the chat responses from Sol are just more intelligible than the word soup I've been getting from Opus. I wonder if Opus could be prompted to respond in simpler prose via agents.md

This is the one I used https://github.com/bevibing/socrates-skill - in the scenarios where I have used it, I have found it useful

I came across the socratic method recently, and have used it to learn a couple of topics that I was having trouble getting to stick. There are some SKILL.md's available for it. It works for concepts as opposed to facts, and causes the model to guide you to answers through your own reasoning, which is both much more engaging than reading a wall of LLM text and helps the information stick.

It was with the transience of valuable agent session dialogue in mind that I built https://www.agentkanban.io - A key feature is context capture in the tasks which you create on the board and then progress in agent sessions. Supported agents are currently Claude and Github CoPilot in VS Code. All of your context is captured in the task and can be reloaded into new agent sessions at any point in the future. Tool use is intentionally discarded because it is proprietary in nature and therefore breaks the promise of session portability.

Exactly. As soon as business works out that there are open models that can achieve parity at a fraction of the cost, they will switch model. They have to. Business exists to make money, and if they can do the same thing at lower cost, they will.

Which is why these companies must be overvalued. This is the definition of a commodity. We have no loyalty to any vendor or model, it's just about capability, price and performance and there can only be so many more increments in capability before these models are good enough that it doesn't matter which you use for 99.9 pct of tasks and the frontier moves on to robotics or the next hard domain (which OpenAI / Anthropic may not lead). I predict that in a year I'll be using whatever the latest version of Gemini flash is (because it's fast, on my mobile and multimodal), and some small cheap open SWE expert model for dev work.

I think because without the initial prompt, they are only interpreting our will, and do not act under their own volition.

I think fragility is the key reason i intervene in llm code too. Good article.

'roll down hill' is a good way of putting it. They don't have 'will', but that's as we want it I think. I think alignment is harder if they develop will. Without will they are still tools that feel like an exoskeleton rather than something that will control us.

Agreed.And a rewrite in another language creates a high probability of a change in behaviour

I have an freemium VS code extension that is converting well (we have 1 perpetual licence sale for every 10 VS Code market place downloads). But I'm having difficulty raising awareness of the extension (330 installs so far on https://marketplace.visualstudio.com/items?itemName=appsoftw...). The extension is https://www.asnotes.io - a notes / PKMS extension for VS Code that includes wikilinking, task management, a kanban board, daily journalling and static site publishing). The perpetual licence, which I feel is priced reasonably, is not enough to justify PPC ad spends to raise awareness. Developer communities like Hacker News are receptive, but you can only post about your app so many times there. Any advice on marketing this sort of app would be appreciated.

I feel like it depends on the task, and that's why people seem to disagree on this. Think about a manager managing 5 devs. If he is working on planning and managing work for his dev team, we don't say he is task switching, he's just taking a management role where he takes a high level view of the task at hand and then delegates the deep dive. Where it differs for devs is that we could in theory run multiple agents concurrently, but frequently, currently, we have to dive in and give the agents significant steers and this pulls us in to the detail. The same will happen for managers. The variables are the complexity of the task, the capability of the agent and the number of tasks. There are lots of scenarios where devs can run multiple tasks without too much mental overload, but I think what is hard is that we don't know when an agent will underperform on a task and we will get pulled back into developer mode. Maybe it's a case of running for as long as you can in manager mode and then accept that when one agent needs help, you have to single task with that agent (I think this is what makes us feel like we are the bottleneck, and that's where the feeling of stress creeps in). I thought about this a lot while working on https://www.agentkanban.io which I use to help me partition agent chats by task, run separate worktrees etc

Came here to see if anyone mentioned Dune Emperor. Would love to see someone succeed

I don't think we're going to convince anyone to replace git on speed alone. I don't spend that much time waiting for git operations, even on really big repositories.

I worked on this problem with https://www.agentkanban.io - this is a human in the loop integration for VS code that stores the context in the kanban task so it lives with the task (can also be forked, split etc).

Think the issue is that it was supposed to be a 'world wide web', but increasingly there's caveats to that.

Same. I haven't been able to see how people let agent loops run without significant steering and produce good quality software. VS Code with one or two integrated terminals running is fine for me. Or a couple of VS Code instances if I'm working one a couple of projects concurrently. The advantage of VS Code is the code / diff visibility if you like to be hands on.

Writers often help the reader secure knowledge that they already feel is correct but have not yet full distilled.

It's honest - people who know what they are looking at will take speed and token costs into account. I don't use Gemini 3.5 for coding, but I use it as something in between a search engine and agent.

Git has worktrees, which provide a means of creating branch linked physical working directories. I built in UI assistance for creating worktress associated with the agent session in https://www.agentkanban.io (an agent integrated kanban board for use with copilot / claude and vs code). I agree, I would rather try and make use of a tool that the agent is already familiar with, unless it's missing features that the agent needs to achieve its goal (which git is not)

It's made me wonder the same, but most LLM generated codebases haven't been around long enough to judge maintainability. I have noticed issues in some of my more LLM heavy code when I expect a change to be replicated in multiple areas, assuming common code / styling was reused, only to find it wasn't. It's for that reason I can't use LLMs for client codebases without heavy scrutiny of every line generated (for my own hobby projects I'm a lot more lenient)

Interface over inheritance is the paradigm I try and stick to. I'd rather maintain orthogonal code than code with overuse of inheritance because of over adherence to DRY.

Self help books are the most filler heavy books. They often have a useful point, but they just don't need to be a book much of the time. This is unfortunate for the authors who have often put significant time into producing the information. Technical manuals typically need all of their content due to the level of detail, and novels need all of their content because it's the content which provides enjoyment. For self help, people just want to know what they need to do to make their lives better? Maybe the work has to move into academia where we have existing infrastructure for rewarding non profitable work for the good of the public?

For me it's about the value of my time. I think that it's important that we have open models, but for getting real work done, my time is too valuable to waste it on subpar results or additional agent management when a max plan covers all the use I need. It's not worth quibbling over. If the cost / benefit ratio changes, I'll be looking harder at local set ups, but not at the moment.