wombat23

UPDATE: I did more experiments - this time with llama-server and pi harness. i'll leave the results here for posteriority:

  ./llama.cpp/build/bin/llama-server \
    -c "$context_size" \
    -ctk q4_0 \
    -ctv q4_0 \
    --no-models-autoload \
    --models-dir ~/bonsai \
    --reasoning-preserve \
    -ngl 99

the max context size i could serve is 64K on GPU only (the -ngl 99 setting). tested with pi harness and it is very fast. the /thinking level always gets reset to off though and it is not very smart like this. haven't figure out a way to fix that.

I don't see any significant speed up. also with defaults according to --help

  -b,    --batch-size N                   logical maximum batch size (default: 2048)
  -ub,   --ubatch-size N                  physical maximum batch size (default: 512)

I also tried double the default. that also means that they can be left to default settings, apparently.

I managed to run it with RTX 3070 (8GB VRAM) following the "Quickstart" on HF model card with minor modifications (modify the architecture 86 for your own hardware):

  git clone https://github.com/PrismML-Eng/llama.cpp && cd llama.cpp
  cmake -B build -DGGML_CUDA=ON -DCMAKE_CUDA_ARCHITECTURES=86 && cmake --build build -j

Then downloaded & verified `Ternary-Bonsai-2-27B-PTQ1_0.gguf` from HF and ran

  ./llama.cpp/build/bin/llama-cli -m Ternary-Bonsai-2-27B-PTQ1_0.gguf -ngl 99 -fa on -c 32768 -b 256 -ub 64 --temp 1.0 --top-p 0.95 --top-k 20 -ctk q8_0 -ctv q8_0 -p "hello world in x86 assembler" -n 256

the parameters were suggested by gpt-5.6-luna to reduce memory footprint, as the defaults ran OOM on my gpu. result looks good:

  [ Prompt: 165.6 t/s | Generation: 40.6 t/s ]

would be nice if they upstreamed their changes so that it runs with the original llama.cpp

Super interesting.

Last week-end I was exploring the current possibilities of automated Ghidra analysis with Codex. My first attempt derailed quickly, but after giving it the pyghidra documentation, it reliably wrote Python scripts that would alter data types etc. exactly how I wanted, but based on fixed rules.

My next goal would be to incorporate LLM decisions into the process, e.g. let the LLM come up with a guess at a meaningful function name to make it easier to read, stuff like that. I made a skill for this functionality and let Codex plough through in agentic mode. I stopped it after a while as I was not sure what it was doing, and I didn't have more time to work on it since. I would need to do some sanity checks on the ones it has already renamed.

Would be curious what workflows others have already devised? Is MCP the way to go?

Is there a place where people discuss these things?

I've been wondering the same. Apparently, where I live (HCOL) people don't care about ever paying off the full amount, as long as the monthly installment is low. Regulations only require that you paid off 33% of the house value after 15 years.

You do need to start at 20% capitalization though, which at current prices means they need to own assets already (market risk), or have bonkers amounts of currency sitting around (inflation risk). I understand the FOMO some people must have.

I wonder if ultimately there is a cascade happening, where increased valuations lead previous owners to either take new mortgages, or sell their old house, and pay more for a new one. Which creates a cycle of ever increasing prices? I've never read anything though whether such an effect exists.

> Even just running correlation analysis on these various assets over the last 20 years showed that crypto was super underpriced

care to elaborate?

When I read stories like this about cool technology from the past that people can still enjoy today, I think it's a pity that the source code is locked away and lost as people and companies move on to new things.

IMO, there should be some kind of archive that conserves and publishes them after some time has passed, so that they could be ported to new hardware and kept accessible. and somehow documented for future historians.

I found them here just before the references:

https://www.sciencedirect.com/science/article/pii/S096098222...

> What neuromorphic hardware can I buy to run your code/ the SNN?

Current neuromorphic hardware is not easily accesible, but you can simulate spiking neural networks. Check out, e.g. https://brian2.readthedocs.io/en/stable/ or Nengo.ai

Looks like a different approach. Intel's chip is based on digital circuits. They try an analog approach.

Well it kinda is.

Empirically, you could have bought a share of the SPX at any point in time, and sold it with profit later. The real problem is what happened in the time between, and whether you were able to hold on.

Nice reminder of some general principles. I wonder how well it applies to young scientists given today's academic career prospects, though.

> If you want to make progress in any area, you need to be willing to give up your best ideas from time to time. [...] Medawar notes that he twice spent two whole years trying to corroborate groundless hypotheses.

Unfortunately, being willing to scrap your idea is only one part of the equation. Securing funding after 2 "failed" post-docs is an entirely different matter.

slightly OT, but does anyone know if there have been any attempts at creating an open source version of the original CS? Kinda like OpenRA or similar?

Are there more sources for technical details about the new infrastructure? The interview linked above left me with more questions than answers.

In this context, the dimensions are the degrees of freedom that a particle can move in. I think it becomes meaningful when you consider the metric (ie. definition of distances) and the laws of nature inside that space, and if additional dimensions help in formulating more phenomena with a less complex system of equations.

-> Classically, there are 3 spatial dimensions (directions of movement) with Euclidean metric, and all the classical laws (Newton's laws, gravity, etc.). Quite successful, but breaking down in extreme cases (high energy, ie. velocity/mass/etc). +there is a whole range of seemingly unrelated laws necessary.

-> with relativity theory (both special and general), time becomes a 4th dimension, but it has a special status in the metric (opposed sign), making spacetime non-Euclidean. One effect of this is suddenly you don't need a law of gravity anymore. Things just follow geodesics in this space. It also explains some effects that could not be derived from classical gravity. Essentially explaining more phenomena with less "overhead", in exchange for more dimensions.

Yes, that's what I meant. Perhaps it's related to the way the printing company produced the books, so this might not be an issue with other companies.

My memory is a bit fuzzy about the details, maybe it was about the inner margins of left-hand side and right-hand side changing. Maybe someone else recognises what I mean? :)

Since Latex is very powerful, I'm sure there is a way to code that behaviour in. But I had spent too many hours already on that project and I was happy with the result anyway, so I stopped investigating.

Having your own book in your hands feels awesome. I've done the same, but not with KDP. Unfortunately, I can't publish it as I had to give up my rights to it (screw you OUP/publishing requirements!).

On the technical side, it wasn't so straightforward to me to get a good result. In my case, I had to iterate a few times and order a new copy with adjustments to margins, etc. until it looked right.

One issue I didn't manage to solve back then was that the inner margin needs to slightly increase towards the end of the book. That's because the binding is eating up visual space since you can't open the book flat without destroying it. Professional publishing tools do this automatically, but Latex doesn't have an option for it. At least that's how I remember the issue. But maybe things have changed?

It's funny, but I feel in the opposite position. To me, computers had become associated with work-only, so much that I needed a change. I decided to assemble a new machine for fun coding and maybe some games (though I probably won't really have much time for it). starting up the new Flight Sim feels great and gives me a different perspective, after staring at a screen of e-mails and calendar and presentation slides for hours.

Thanks for your reply!

I see the corporate aspect of F#, but can you elaborate on what you mean by "special" about Haskell?

What I'd like to understand: When/why does one choose Haskell over other functional languages e.g. F# or OCaml?

It looks to me like they would satisfy the same points that the article makes.

Edit: just did a quick comparison of the last 2 SO developer surveys, and it looks like Haskell "replaced" F# in their popularity ranking last year.

yes, and I agree with that.

But I also feel like the language is mainly used in a type of business environments that are already tied to .NET infrastructure anyway, and who actually appreciate the interlock.

And it could get much more traction outside of that circle, especially in the open source world.

the emphasis is on "deeply".

I think .NET can be quite an advantage, as stated also in the article.

but if I want to, say, set up a dev environment on linux, it comes with a set of quite different dependencies than what other languages use.

> I evaluated and expected not to like F#. I actually quite like it.

It's funny because that's exactly how I felt when I learned F#. I felt pleasantly surprised and never felt like there was anything bad about the language itself.

I just wish it wasn't so deeply embedded in the .NET ecosystem.

It looks like a nice project. I've also felt the urge to do some TUI programming recently. Actually, I also thought a simple music player was a nice way to do that. Must be the home-office effect (??)

Does anyone have some resources to learn how (UNIX) terminals work under the hood and how one can achieve nice graphic effects with it?

Also, is ncurses still the go-to solution, or do people use something like termbox or something else?

it's kinda scary that the go-to personal investment plan is based on a leveraged buy & sell with the implied promise that things perpetually increase in value.