Okay don't judge me on the title. It's actually what'll help you save up to 75% tokens while your using any Cloud LLM model. So as you can see in the screenshot below, there's this dev that made claude speak in "caveman" terms. Turns out that saves tons of credits for you. Tbh the more AI evolves, the more dumb and easier ways to save tokens arise, maybe soon enough the em dashes and 50 line answers would be eradicated by default? Anyways, here's the link to the post if you'd like to snoop around there: https://www.reddit.com/r/ClaudeAI/comments/1sble09/taught_claude_to_talk_like_a_caveman_to_use_75/

So there's a lot going on with kimi right now, from what I've looked at, for being completely open source it's quite amazing. The benchmarks look quite promising taking in the fact that it's completely open source. Specialized more towards the longer coding tasks and back end coding languages. A really interesting aspect is that (from their article): Kimi K2.6 successfully downloaded and deployed the Qwen3.5-0.8B model locally on a Mac. By implementing and optimizing model inference in Zig—a highly niche programming language—it demonstrated exceptional out-of-distribution generalization. Across 4,000+ tool calls, over 12 hours of continuous execution, and 14 iterations, Kimi K2.6 dramatically improved throughput from ~15 to ~193 tokens/sec, ultimately achieving speeds ~20% faster than LM Studio. The website design isn't that bad at all, but it feels like they were trained more on aesthetic PPT's than code. Anyways, tons more to explore, the link's here: https://www.kimi.com/blog/kimi-k2-6

You do need at least 1 claude subscription to access claude code in itself, regardless of which one. So how claude's usage works is that it saves all your transcripts and messages locally, that file was never a server side block. So you're basically resetting your usage manually from the log file that claude stores. Don't know how people figured it out. So you need to find your claude local folder and then delete everything except the first line, that's a header. So before you go ahead and do this, you will 100% get banned if you do this. Enjoy! 🤣 (This was for educational purposes only, don't come attacking me for the ragebait!)

So, if you’re looking to set up a memory system but with all the new obsidian, graph, this and that sort of AI news, you just don’t know which one to, here’s what I recommend. I don't know your exact use case so I recommend copy pasting my message, putting it into claude and talking to it about your use case. So you have the obsidian stack, where the AI first looks for a topic, it doesn't exactly ask (query) any sort of external system. How it works is that the AI has a sort of index page, you know how our school text books have those main pages with chapters and page numbers? Something similar, but in this case, every "Chapter" leads to another chapter which is related to the main topic. So let's say you have a topic such as "How do I implement the apple branding guidelines", it would find that first, then it would see that there's a reference of another page, such as "How to implement a luxurious design with react", both relevant so far, now inside of the how to doc, it would refer to "How to save tokens while designing a website", now Opus could possibly think of saving tokens but by default with just the most simple prompt and topic you gave it, it would not go to that doc, it pulls the first 2 relevant documents and then understands that. If you're wondering how you even get those docs, you basically dump your notes, research papers, whatever you'd like to and then it would use AI to sort it out, 1 research paper can become 50-500 different documents, this is the Andrej Karpathy memory system. https://gist.github.com/karpathy/442a6bf555914893e9891c11519de94f The 2nd method would be through Vector, it's not always needed though. Going into Vector, it query's your database but with an unclear task. SQL would be: "find users where plan = paid" You know exactly what you need and need definitive results. Vector's more like: "find docs similar to this concept"

So, rivaling SaaS's like lovable, claude very recently launched claude design, a dedicated platform for designing anything you'd require with 3 main options. Available at https://claude.ai/design with any of the claude plans. So you've got Prototype's, Slide Decks and From a template. Templates work from past projects you've already made. You set up a design system and with the power of Opus 4.7, it makes you a wireframe or a high fidelity prototype. Lots of exciting new things coming out recently even with the new OpenAI Codex improvements, how far will the AI race go? Who knows. There also have been a lot of theory's of claude opus 4.7 being prime opus 4.6 (quick context: when opus 4.6 first released, everyone loved it but as time went on, maybe due to mythos, the performance degraded over time while the costs remained the same, whereas now opus 4.7 is much more expensive and is slightly better than prime 4.6)