Realme 5

Sorry about the delay, some bugs to swat 2 hours later and $2 in tokens, Prince of Persia in Swift 6 for MacOS is ready (MacMini M1 and Tahoe 26.6.2 is my setup) Not fully tested by me (AI tested it fully on every milestone, 340 tests) but for sure tonight I'll do it from beginning to end Let me know what you think, feedback is always welcome And btw, cmd+1..5 can change the resolution to a bigger canvas, I prefer it at cmd+5 almost full screen. Haven't tested retina display as I use a TV as monitor and resolution is not optimal Enjoy! I love DeepSeek All rights belong to their owners Interesting docs to read: readme, architecture and prompt There is an app in build folder, that's the game Don't trust code downloaded from the internet, ask your AI model of choice to review it for malicious code. [...]

[...] I've already bought a USB-Floppy (and other USB-based "old media" bits), but I'll definitely be keeping the LS-120 now as I think IDE is more durable (and IIRC LS-120 drives were more "precise" w.r.t. [...]

Owning an EV doesn't necessitate spending a few grand to have a fast charger installed. They charge fine from a regular 110v US wall outlet -- they just do it slowly. And by "slowly," I mean: Somewhere in the realm of 2-5 miles per hour. [...]

[...] I do agree that typed passwords are still susceptible to the length check, sending noise like you mentioned sounds like a good plan. [...] I've considered replay protection but for now I think the tradeoff for managing senders easily just based on pubkey is worth because of the next point 6. [...]

[...] I didn't really build projects, just scripts. [...]

[...] They demand a reproducible build, but then can't do things like install the right version of yarn (no - npm install -g yarn is not correct, our readme says it in bold like 5 times and provides the exact correct command to install the right version), or follow basic setup steps like "Use this version of node (complete with exact steps to install it and a script to automate that for them)". [...] They're incompetent at best, and I'm fairly convinced they're just milking the google search deal income for as a much as it's worth - I don't think they really want to provide an alternative and secure browser anymore.

The biggest day-to-day merge conflict annoyance I have is situations like this. As far as I know, there’s no solution: commit aaaaaaaa diff --git a/README.md b/README.md --- a/README.md +++ b/README.md @@ -1,4 +1,4 @@ -<p align="center" +<p align="left" <img width="200" src="logo.svg" </p commit bbbbbbbb diff --git a/README.md b/README.md --- a/README.md +++ b/README.md @@ -1,5 +1,5 @@ <p align="center" - <img width="200" src="logo.svg" + <img width="345" src="logo.svg" </p Project Commit aaaaaaaa changes some text on line 1. Commit bbbbbbbb changes some unrelated text on line 2. I don’t want to have to manually resolve this—just merge the lines. If there’s a semantic conflict I’ll let the tests sort it out. When I’ve looked in the past there wasn’t a merge strategy that fixes this.

Because it's impossible to get started as someone who has no background in ML. I tried earlier this year and also failed. 1. Obtain the Llama models. Apparently you have to sign up for access and get a download link? I don't want to do that, found some public download links instead. Ok, now I have a few hundred GBs of model files. 2. Compile llama.cpp. Missing dependencies, took an hour to figure out how to resolve. 3. Quantize the model? What does that even mean? 4. Install pytorch. Run the command in the README. Python exception. 5. Install NVIDIA helper libraries. Doesn't work. Try installing the AMD helpers to run on CPU instead. No instructions for how to do this. Eventually figured it out. 6. Try running pytorch again. Same exception. I gave up after a full day of trying to make this work. The Docker image is for people like me.

I currently use DeepSeek-V4.1-Flash and the V4-Pro and -Flash versions before that, because they are extremely cheap. [...] GLM-5.3, GLM-5.3-Flash and Kimi K3 are also fine, but slower, more expensive, and less good for what I use them for, which is mostly Python and CUDA programming with some JS and HTML inbetween. [...] I haven't tried Astra or Fable yet, because I am not made of money and am happy with my current setup. [...]

All instructions to LLMs are merely suggestions to nudge it in the right behaviour. Unless you have a deterministic guardrail that guards against a single specific action, everything else is a slot machine that's biased strongly in your favour. My global CLAUDE.md explicitly states "When commenting on code and configs, or writing MD files, strictly write within the domain of the content being commented on. DO NOT include information, negatives or ramblings from work sessions. For e.g. if commenting on a proto string field that is replacing an int field, do not comment that 'this is not an int field'". This reduced the idiocy of the agent (Opus 5 included) when writing documents. But I'm still catching it writing README.md talking about the negatives that it removed. Those belong in the memory if it is actually that important (most of the time it's junk), but Claude doesn't seem to understand and never ever learns.

I'm also skeptical partially because I don't like the huge essays generated by any llm. CLAUDE.md/AGENTS.md/README.md that are 5+ pages long are all equally bad imo. I prefer following the idea that if something is too verbose for me to want to get anything useful out of it, then the llm should behave similarly. Even if it's not true, why waste 2 paragraphs explaining something that could be explained in one short sentence? My CLAUDE.md or AGENTS.md is usually just a bulleted list of reminders with high level information. If the agent needs more steering, I add more reminders. I try not to give it too broad of a task without prior planning or it'll just go off the rails. Something I haven't really experimented with is having claude generate ADRs [1] like your RFC/ idea. I'll probably try that and see how it goes. [1]:

[...] Again, if it is an one-in-a-while job, the time taken to build jigs is not important. [...] Having done so much work, I just can't see dropping that kind of money for something that isn't likely to give me any real advantages. [...] Best of all, if you need five people working on a project, making extra jibs and fixtures is easy and cheap.

I threw Claude Code at an existing codebase a few months back and quickly quit— untangling its output was slower than writing from scratch. [...] • 95 %+ - built Backlog.md, a CLI that turns a high-level spec into those task files automatically (yes, using Claude/Codex to build the tool). [...] For simple features I can even run this from my phone: ChatGPT app (Codex) → GitHub app → ChatGPT app → GitHub merge. [...]

To say that 5-MeO is a short duration psychedelic doesn't convey what's actually happening -- basically, you go from your normal waking state to likely the most profound experience of your life, and then, if you felt so inclined, back to work...all within 10-20 minutes. That's precisely what I didn't like about it, the rapid shift back to everyday consciousness. At least with other psychedelics you can spend an entire day in the woods, feeling the entire ebb and flow of the experience, and perhaps learn something in that process. With 5-MeO it's just instant God realm, and then, hmmm, what's for breakfast? Strange substance :) Finally, it's not for the feint of heart; until you try it you have no idea what you're getting into -- there is no preparation for the rocketship that is 5-MeO.

From the README: Per move, a model gets 5 illegal moves before forced resignation of the round. Most of gpt-4's losses were due to illegal moves, so it may be possible to come up with a prompt to have gpt-4 correct illegal moves and improve its score. My experience exactly with gpt4. I first solved by keeping a list of illegal moves and subsequently prompting to move again and explicitly exclude those moves. Idea was to constrain gpt as little as possible, but that became inefficient. So I began to prompt for each move with the complete set of legal moves to choose from. In any case, at a certain point it becomes somewhat random when the goal is just to produce a legal move. But, it's interesting that producing legal moves is a challenge for gpt4, especially given that it's such a discrete problem.

[...] I disclose that the last 5 slides are lifetime stats in my readme: "Note: The last 5 slides are not necessarily specific to 2025 because of leetcode's graphql api only allows querying up to 20 of the latest submissions from an unauthenticated user. [...]

It is compiled in mere milliseconds or sometimes in microseconds. It is really hard to create an efficient ANTLR grammar. Took me 2 years. So please, do not take 5 seconds to look at the readme and sign it off.

I use vterm which I believe prides itself for being the fastest terminal emulator inside Emacs. The eat README says 1.5 times faster than eat. This matters a lot to me because the proprietary build/test tool at work likes to dump the stack trace for every failing unit test.

[...] A dance instructor friend mentioned they could make two left feet better (not perfect) by teaching them to practice the micro movements, one at a time, sometimes for hundreds of reps, and then putting them together slowly.[3][4] Anecdotal again, but when I tried learning a move the same way, it wasn't as hard learning it a bit at a time vs just throwing myself into it. We know muscle memory is a thing for using tools like Vim and keyboard shortcuts, so I don't think it's out of the realm of possibility.[5][6] References: [1] Development Period of Prefrontal Cortex "Prefrontal cortex matures around age 25." [2] Development of the Prefrontal Cortex "Prefrontal cortex development extends into adulthood." [3] Neuroplasticity Subserving Motor Skill Learning "Motor skill learning involves neuroplasticity." [4] Motor Learning and Plasticity "Practice leads to changes in motor cortex organization." [5] Experience-Dependent Structural Plasticity in the Adult Human Brain "Adult brain changes structurally with new experiences." [6] Micro-learning in the Age of Neuroscience "Micro-learning enhances retention and mastery."

[...] I definitely haven't used it much for that type of use case but I'll keep that in mind. I don't want to give up too much info about my project because it would probably dox me, but it essentially involves generating consistent geometric shapes in a 2d array. Like the following: 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 0 0 0 1 1 1 0 0 0 1 1 1 0 Both current flavors of GPT are unable to generate spatially consistent shapes according to my specifications. Who knows, maybe they will improve in this realm in GPT-5 with the inclusion of their rumored Q algorithm.