mxmlnkn

Cost: LLMs fitting into local GPUs, in addition to the memory required for game textures, are too constrained for the quality, latency, and concurrency required. Connecting to the cloud would be too expensive, and even if it would work, it would indeed be the death of local offline gaming, and it would give a valid reason for a subscription to keep the servers running as opposed to a buy-once play-forever game business model.

Unreliability: You don't want to miss a major quest because the LLM did not decide to hand it out. It might be workable with some proper "agentic harness", with some hardcoded world lore, long-term memories, goals, midterm memories, some deterministic actions, and a short-term free-form loop that gets cleared every now and then. For OpenClaw instances, it would be the SOUL.md, but it would have to be read-only. Some logic might have to be hardcoded outside of LLMs, e.g., to start a new narrative branch powered by a new character memory when a keyword has been mentioned or a quest item has been given. Similar to agentic frameworks, it would have tool calls such as `offer_quest`, `give_reward`, etc.

These ideas seem to converge onto the Westworld mechanics of backstories, cornerstone memories, reveries, narratives, and loops, at least as far as I remember them from the series. Thinking about it, given enough investment, I feel like the amusement park of Westworld should already be somewhat possible to implement with the recent advances in LLMs and robotics.

In the end, I am not sure how much sense it makes for games. Maybe it would suffice to only make companion characters LLM-enhanced while NPCs stay pre-programmed in order to let the game stay deterministic. Games should simplify real life in order to increase enjoyability. Making all characters backed by LLMs might decrease enjoyability because of forgetfulness or confabulations inconsistent with the game world and actions. It would need to be heavily constrained and trained or prompted to say "I don't know about that.". It might also be questionable how much needs to be dynamic. E.g., when thinking about the NPCs in The Witcher, which are talking about random stuff, maybe you only need to pregenerate much more random stuff for them to talk about so that you don't notice the repetition, but there might be no need to have this completely dynamic and burn through GPU power for every action. The question would be what to cache and what really needs to be dynamic.

I also keep circling around this point. So many software repositories in the AI space seem to follow a publish and forget pattern. If you simply can show that you have the patience to maintain a project, ideally with manual intervention instead of a fully autonomous AI, then you already have an outstanding project.

These are not errors. They are simply warnings about extended attributes being ignored when extracting files, which seems completely fine to me, and creating the tar without those extended attributes has exactly the same outcome, but throws away the metadata at archive time instead of extraction time.

Furthermore, this is not an Apple/macOS issue. The tool used is bsdtar, so it would also affect all BSD-variants that default to bsdtar/libarchive, and those systems also have extended attributes, e.g., for SELinux, which would get added to the TAR.

https://github.com/martinellimarco/libzstd-seek

Note, however, that this can only seek to frames, and zstd still only creates files containing a single frame by default. pzstd did create multi-frame files, but it is not being developed anymore. Other alternatives for creating seekable zstd files are: zeekstd, t2sz, and zstd-seekable-format-go.

Exactly. And often this state is either highly compressible or non-compressible but only sparsely used. The latter can then be made compressible by replacing the unused bytes with zeros.

Ratarmount uses indexed_gzip, and when parallelization makes sense, it also uses rapidgzip. Rapidgzip implements the sparsity analysis to increase compressibility and then simply uses the gztool index format, i.e., compresses each 32 KiB using gzip itself, with unused bytes replaced with zeros where possible.

indexed_gzip, gztool, and rapidgzip all support seeking in gzip streams, but all have some trade-offs, e.g., rapidgzip is parallelized but will have much higher memory usage because of that than indexed_gzip or gztool. It might be possible to compile either of these to WebAssembly if there is demand.

This part sounds especially cool. I did not think about this application when reading the other articles about TurboQuant. It would be cool to have access to this performance optimization for local RAG.

Personally, I quite like the site's design and its font. My gripe often is light gray text on a darker gray background. The bad readability that so many newer sites seem to prefer makes me question my eyes or my monitor capabilities. Reader mode in Firefox is also often very helpful.

The trick for the mix of JSON with binary is a good reminder. That's how the ASAR file archive format works. That could indeed be usable for what I was working on: a new file format for random seek indexes. Although the gztool index format seems to suffice for now.

I have had to work with large 1GB+ JSON files, and it is not fun. Amazing projects such as jsoncons for streaming JSONs, and simdjson, for parsing JSON with SIMD, exist, but as far as I know, the latter still does not support streaming and even has an open issue for files larger than 4 GiB. So you cannot have streaming for memory efficiency and SIMD-parsing for computational efficiency at the same time. You want streaming because holding the whole JSON in memory is wasteful and sometimes not even possible. JSONL tries to change the format to fix that, but now you have another format that you need to support.

I was also contemplating the mentioned formats for another project, but they are hardly usable when you need to store binary data, such as images, compressed data, or simply arbitrary data. Storing binary data as base64 strings seems wasteful. Random access into these files is also an issue, depending on the use case. Sometimes it would be a nice feature to jump over some data, but for JSON, you cannot do that without parsing everything in search of the closing bracket or quotes, accounting for escaped brackets and quotes, and nesting.

I have tried Cursor a bit, and whatever it used worked somewhat alright to generate a starting point for a feature and for a large refactor and break through writer's blocks. It was fun to see it behave similarly to my workflow by creating step-by-step plans before doing work, then searching for functions to look for locations and change stuff. I feel like one could learn structured thinking approaches from looking at these agentic AI logs. There were lots of issues with both of these tasks, though, e.g., many missed locations for the refactor and spuriously deleted or indented code, but it was a starting point and somewhat workable with git. The refactoring usage caused me to reach free token limits in only two days. Based on the usage, it used millions of tokens in minutes, only rarely less than 100K tokens per request, and therefore probably needs a similarly large context length for best performance.

I wanted to replicate this with VSCodium and Cline or Continue because I want to use it without exfiltrating all my data to megacorps as payment and use it to work on non-open-source projects, and maybe even use it offline. Having Cursor start indexing everything, including possibly private data, in the project folder as soon as it starts, left a bad taste, as useful as it is. But, I quickly ran into context length problems with Cline, and Continue does not seem to work very well. Some models did not work at all, DeepSeek was thinking for hours in loops (default temperature too high, should supposedly be <0.5). And even after getting tool use to work somewhat with qwen qwq 32B Q4, it feels like it does not have a full view of the codebase, even though it has been indexed. For one refactor request mentioning names from the project, it started by doing useless web searches. It might also be a context length issue. But larger contexts really eat up memory.

I am also contemplating a new system for local AI, but it is really hard to decide. You have the choice between fast GPU inference, e.g., RTX 5090 if you have money, or 1-2 used RTX 3090, or slow, but qualitatively better CPU / unified memory integrated GPU inference with systems such as the DGX Spark, the Framework Desktop AMD Ryzen AI Max, or the Mac Pro systems. Neither is ideal (and cheap). Although my problems with context length and low-performing agentic models seem to indicate that going for the slower but more helpful models on a large unified memory seems to be better for my use case. My use case would mostly be agentic coding. Code completion does not seem to fit me because I find it distracting, and I don't require much boilerplating.

It also feels like the GPU is wasted, and local inference might be a red herring altogether. Looking at how a batch size of 1 is one of the worst cases for GPU computation and how it would only be used in bursts, any cloud solution will be easily an order of magnitude or two more efficient because of these, if I understand this correctly. Maybe local inference will therefore never fully take off, barring even more specialized hardware or hard requirements on privacy, e.g., for companies. To solve that, it would take something like computing on encrypted data, which seems impossible.

Then again, if the batch size of 1 is indeed so bad as I think it to be, then maybe simply generate a batch of results in parallel and choose the best of the answers? Maybe this is not a thing because it would increase memory usage even more.

Yes. Both, docx and xlsx are literally just a zip of XML files with a different extension. PDF can contain zlib streams, which use deflate compression just as gzip, so all the mentioned methods apply to all three formats.

I don't think that it is of much use as waste disposal because again, it can only remove 10%, i.e., an insignificant amount. If it were even mined because of this, then more mercury waste would be produced than before, but increased mining would probably be many decades or centuries in the future, as long as there is still waste to reuse.

So, how long would the current midterm stockpile of 1400 t for 198Hg for the next 10 years last? At 5 t per 1 GW per year, i.e., 5 t per 8.76 TWh, and a current global electricity generation of ~30 PWh, replacing all energy production with fusion would be able to transmute 3400 t 198Hg per year, over twice the stockpile. Of course, there would be a myriad of other bottlenecks long before that, but consuming all the existing stockpile seems feasible in human time spans.

I am honestly impressed by the amount of transmutation that is possible with fusion. And it is a lucky coincidence that the half-life is only dozens of hours for the middle product. I never thought of that process or would have guessed grams of production instead of tons, probably because of the association with existing particle accelerators. It is quite amazing, but also presumably still decades off into the future.

https://en.wikipedia.org/wiki/Abundance_of_elements_in_Earth...

    sudo apt-get install zbar-tools oathtool
    zbarimg qr-2fa-code.png
    

    QR-Code:otpauth://totp/username?secret=ABCDEFSECRET012349BASE32&period=30&digits=6

For gzip, it is as you say. However, when only wanting to seek to DEFLATE block boundaries, the "state" of the decompressor is as simple as the last decompressed 32 KiB in the stream. Compared to the two offsets for bzip2, this is 2048x more data to store though. Rapidgzip does sparsity analysis to find out which of decompressed bytes are actually referenced later on and also recompresses those windows to reduce overhead. Ratarmount still uses the full 32 KiB windows though. This is one of the larger todos, i.e., to use the compressed index format, instead, and define such a format in the first place. This will definitely be necessary for LZ4, for which the window size is 64 KiB instead of 32 KiB.

For zstd and xz, this Ansatz finds its limits because the Lempel-Ziv backreference windows are not limited in size in general. However, I am hoping that the sparsity analysis should make it feasible because, in the worst case, the state cannot be longer than the next decompressed chunk. In this worst case, the decompressed block consists only of non-overlapping back-references.

Heapsort comes to mind first.

:= definition ≡ identity = equality ∝ proportionality