With agentic stuff there's also a large amount of commits which are not code.
For instance with OpenClaw and similar, they often simulate institutional and short term memory with markdown files in folders. Other tooling that runs companies using agents as staff, for example, do the same - but also with files for inputs, outcomes, handovers etc.
All of this means a lot of extra churn as these kinds of files can be changing with every interaction not just every traditional commit point.
A 1-2-1 in this context likely refers to the 1-2-1 sessions some leads/managers have on a regular basis with each of their reports. One on one catch-ups on how they're doing.
- I write in Markdown files using VS Code and a custom syntax highlighter
- A custom C# tool stitches it together (with YAML metadata)
- It generates reports on chapters, dialogue, character presence etc
- It uses custom code to generate an EPUB (v2)
- It uses custom code to generate a DOCX (validated)
- It uses custom code to generate a PDF (print or with interactive links)
Works great and gives perfect results in seconds (beats Vellum, which I quite like). The only drawback is the yak shaving involved in my totally stand-alone solution leads to more time tinkering than writing.
One of my trifecta of 'perfect' DOS software: DataEase 4.53, WordPerfect 4.2, and TopSpeed Modula-2. Given the restrictions of the day, all three were incredible.
I'm a fair bit lower than some others as I only use it outside of work hours on my own small projects, but my Cursor account shows (for a random recent date) 12,184,233 tokens in a day. That day feels pretty representative.
That's with 86 interactions spread intermittently over a couple of hours so if I did a full working day like that I'd be looking at maybe 40 to 50 million.
Some interesting stuff I wasn't familiar with, thanks.
I really like the book Mazes for Programmers by Jamis Buck [1].
Also, my open source Dungeon generator is a (slightly-misnamed) maze generator [2]. It produces 2D maps, and also 3D files (OBJ and MTL) for use in Blender etc. I like to think it does a more 'reasonable' job than many, but I am biased.
> So, a UK-only advice, and it strangely assumes that any other service in UK wouldn’t be bound by the same laws.
I suspect it's because whilst other services would be affected we only know about Apple currently and, thanks to iOS and Mac, a large percentage of the population will be using Apple by default for the services impacted. Only Google (Android) and Microsoft (Windows) really overlap in that regard.
* They offered local councils the chance to request it if they were going through a reorganisation or devolution process.
* 18 councils requested and 9 were accepted as justified.
* And even those are only delayed until May next year (one year after the rest of the UK).
So to be clear the UK government not only didn't postpone the general elections but half the councils who requested the local elections were postponed were denied, with the other half having reasons and still doing it a year later anyway.
And all that is actually covered in the page you link to.
Why are we hearing that "studies" have "uncovered the concept of context rot as the number of tokens in the context window increases"? It's obvious, and we've always known this.
Agents are stateless, hence the need for context. This means that all they know about the ongoing session is what's in that context (generally speaking). As the context grows any particular element within it becomes a smaller and smaller percentage of the whole. The LLM is not 'losing focus'; it's being diluted with more tokens. But then I suppose anthropomorphism comes naturally to a company named Anthropic, and 'losing focus' does make it sound more human.
They didn't need a study and article, but it likely contributes towards the mystique. Hence the use of phrases like "this results in n² pairwise relationships for n tokens" to make it sound more erudite and revelatory.
Not quite what you wanted, but the Lone Wolf ones are (legally) available online [1]. You may be able to read the downloads, or even the online play versions, with Dutch translations.
If you're looking for a GUI editor for your LaTex content, personally I like Lyx [1]. It's featuresome, easy to use, and also includes outline navigation.
I'm a Kobo user so haven't bought Kindle books in many years, but IIRC you can only click to buy a single book at a time, rather than add them into a cart for a bulk buy.
The Kobo book store is smaller than Amazon's, but still huge. The UI may look different, but it is still the same process of search/browse, buy/preview. Either on the web site or from the device.
All the talk of conversions and Calibre etc in the thread is possibly muddying the waters as it is totally unnecessary if you just want to mirror the Kindle flow.
I'm on a Kobo H20 as well after upgrading from a Kindle many years and several devices ago.
* Also works in lots of countries
* Great iOS and Android apps
* Well-priced devices, and no ads to remove
* Great hacking community (even complete replacement software) and DeDRM support
* Insanely robust library that has never let me down
* Ability to easily side-load content without needing a send-to feature (plug it in and drag/drop, or use Calibre)
* No Audible, but (a) it's an ereader not an audio player (it does actually do mp3s though) and (b) Kobo have an audiobook store
So all in all, the Kobo has worse audiobooks but better hacking and side-loading.
Even if you prefer Kindle over Kobo, it is clearly not second-rate. And as for the last comment about being tied into an ecosystem, that's purely an Amazon thing. Kobo don't require that kind of acquiescence.
- For the AI provider it is a massive money sink due to cost/revenue maths
- For the AI consumer it's not a massive money sink if you arrange things so your ongoing costs are lower than your savings
It isn't a simple yes or no overall, only per side of the fence.