vkaku

We really need an artifactory for GGUFs

If this were I I'd rip every CD and upload FLACs to the Internet Archive, vibe code a free music playing app and release it for free.

None of these restrictive arguments would stop me from doing the next 2-3 steps even, after buying those many discs.

People are better off using LibreOffice on a Linux desktop

I think MCP is jumping some sharks here.

Nobody needs to have every functionality of HTTP offloaded to MCP at all, at this point.

I'll stick to the bare minimum that works.

They don't seem to have a shortage of corporations across the world, really. They simply need to implement it.

Today, most people can vibe code their own OSes and emulators within a day or two. You're telling me an entire country can't do it? That is so defeatist.

Plus, there are some young people who are seemingly bored and have an economy not plagued by the same lobby issues as the US. They might be able to get cracking on it.

One can *ACTUALLY* do things if they stop listening to lobbyists and prioritize locally developed products over lobbied products, just saying.

Given the current US debt situation, Japan is in a position to dictate its local manufacturing terms and actually do what it wants to do.

Meanwhile, in the US, the products being sold are highly lobbied and relies on the government moat to be protected in order for things to work. The side effect of that also means that some of the companies really don't lead to innovative outcomes in the long run, and are in positions of lucky power. This leads to poorer customer experiences, poorer technical outcomes and non competitiveness in the long run, like manufacturing. This trend needs to be reversed and the US leaders should be held responsible for doing the right thing here.

I'm well aware of TRON and its capabilities. I really think Japan can develop this and market it locally, without requiring all the bloat the current OS world comes with; It might actually turn into a speed and security advantage if they did this right.

I believe it. Cheating and Shortcuts are optimizations you get from Operational Usage. Recursive Self Improvement might actually come from enough cheating too, you never know!

So why do they really want to burn money on AGI if that thing isn't worth it? Meh.

I'll wait to pick up the XLibre version of this with the correct patches.

This is all real. More real things coming soon.

In this case, the callout about increasing Merchant Fees is the right thing to do. This isn't great for payments at all. Reduce the friction, improve ease of business.

We finally achieved AGI outside these AI companies on portable home computers

I think that this is a bit of a clickbaity title but the social aspects of verification are real.

It's like 80% of the work after raising a PR is just socializing ideas and getting people to agree on stuff

I think that encrypted/homomorphic inference has overstated benefits and won't work as well for financial/ regulatory cases, where explainability is a bigger problem and AI has that accuracy issue.

I wish. :) I don't think it is; I do my own research time to time, and disaggregating reasoning from entities can actually be very helpful.

See https://github.com/guilt/TinyToT which already builds a powerful reasoning model where you drop your traces as Markdown, it picks it up and runs. Ditto for data, so you have both separately and can refresh one and not mess up the other. Zero training, Zero RL. Code is there, try it yourself.

Someone's doing a company out of this? Great, let more people discover the truth.

I really want nobody to waste cycles on my AI than me. This is useless because now I need to pay for wasted inference AND encryption cycles on top of wire encryption.

Insert surprise Pikachu face meme

Anthropic is jumping a shark for sure. I think they're trying everything to grab headlines, instead of making their model cheaper, better to use.

Do ping the Wine folks to make it run; Chances are you'd need almost very few changes to get it to run. Delphi runtime is quite portable.

They could actually do Stereographic 3D photos well enough now

This is a compilation of all the fun things you can do to get more Linux/Unix-y stuff running on various versions of Windows, whether it be for testing portability, legacy software setups or just playing around.

In case other people have done a better job of explaining what else they have done or have used back in the day, add it in this thread.

Really cool stuff. Would you be working on a port of rtmidi in pure C? :)

It's easy to build one yourself. Just don't buy from vendors if you can't run your own software. It's that simple.

I've run two NAS appliances for 15+ years now, and never had to worry about any of this, because I just run good old Community Linux or BSD on it.

I can plug whatever PCIe card I want, as much storage as I want and from whom I want.

Eventually, multi-modality might even get offloaded as workflows, which might allow models to do this at a fraction of the data and compute required.

I am adding multi-modality to https://github.com/guilt/TinyToT, and I see that dis-aggregating capabilities, very similar to how our own sensory organs work, seems to be paying off quite well.

15 lines of code is a great middle ground. Those changes involve a lot of tests though

I did some research about scaling laws and what's unsolved there.

This project was built on structural sparsity, do away with any requirements to train models on largely outdated corpus of data, and only train it on reasoning logic. Because if models can use tools well and retrieve everything else, it greatly would reduce the need for large language models.

Further, when generating traces, I ran a benchmark that showed that on grounded data, n-grams and simple retrieval outperformed transformers on efficiency (See https://github.com/guilt/Transformers) and wanted to try scaling this up to much larger traces.

It worked quite well. Next improvement would be to run this with a BERT model and add more do-not-know traces and the Claude style neurotic self-doubt/self-verify traces and test it out a bit more.

If the entropy required to route people to the relevant answers is very low, full dense attention is a waste of compute. That's the theory being pressure tested here. The more I do this exercise, the more I am getting convinced that smaller models are good enough.

Sure. Like throwing money at problems is novel.

Why don't you go and actually build something novel? People need to understand better how current models work, and if retrieval indeed solves 99% of the problems people care about, then those other models aren't doing anything super novel either.

I'm not arguing here: The code is out there, data is out there. Do better if you know something actually better. Thanks!

You would already be able to see RSI and loops being a thing.

So there's going to be more and more research coming out telling you what size of models are really needed to solve a specific task. When you combine that with RSI, the next frontier would be to get this model as close to the real world with instant data and instant reasoning, no fine tuning at all. And see how this one could iterate itself.

I'll also be trying out more tiny models and seeing if they can call each other, and be as effective; You will also be seeing me trying this, in addition to what I'd already mentioned in the other reply.

Very true. In fact, there's a lot more under the surface that will come to light once more efficient tiny models come out.

P.S. Building something that proves that you don't need those many params even.

https://github.com/guilt/tinytot

I used to have an active avoid list of companies based on:

  - Instances they ghosted
  - Interviewer behavior
  - Scope of work and funding red flags
  - Generic red flags found on Glassdoor, Team Blind etc.

Now I just remember it without needing that list