I'd appreciate an explanation of what "open weight model" means. Is it a "weight model" that is open, or a model with open weights (so should be "open-weight model"), or is it weights that can be applied to a model?
Are weights separable from a model? And if not, what is the point of saying "open-weight model" instead of just "open model?"
To the newcomer, it's hard to determine what the components of an AI system are from the throwing-around of these terms.
As another commenter said a "model" is a file (or group of files, there's multiple formats available; GGUF format is all in one file for example). You download it to the hardware of your choice (ie your own desktop with NVIDIA GPU). You run the inference engine (llama-cpp, ollama,lm studio etc) and tell it where the downloaded model is and it runs inference (so you can start chatting with it, or run agents).
"Open weights model" means the developer made the model available for everyone for free. You can download it from huggingface.co for example and do whatever you want with it.
Why "open weights" and not "open source"? Because the "source code" for LLM would include things like training data, training methodologies and tools, so that you can do the training and produce the model (files) yourself. That would be like compiling from source code. Which is not done with these models, it's company's know-how, they only share the end result.
It's more analogous to "freeware" which is what we traditionally call freely distributed binary executable files. But people started calling them "open weights" instead and the term stuck.
The trained weights are open, the training software is open, and the data that goes into training the model is open.
Not many models are fully open.
An open weights model is one that has freely available trained weights, and maybe fine-tuning tools, but it lacks the original training data (and usually lacks the training software). These are the most commonly used local models, like Google's Gemma series, Meta's Llama, or Alibaba's Qwen.
So you can apply different weights to those "non-open" models?
Also, I've read a bunch of descriptions of AI components, but none of them has said what the weights are applied to in the model. I guess that every model contains a dictionary of words and phrases, and the weights map relationships between them?
All the descriptions simply talk about weights being applied to "input," but neglect to say what that input is compared to. If a user submits a query, are the words in the query weighed against the words in the model?
The weights are just numbers. I don't what technical background you have in other areas of computing, but I think that this is a good, short introduction that doesn't assume too much:
To quote part of it, Training a model can be thought of as tuning the dials on a really big machine. The way that a language model behaves is entirely determined by these many different continuous values, usually called parameters or weights.
Longer and slightly more technical, "Intro to Large Language Models" by Andrej Karpathy:
An "open weights" model is one where you can download all the data and the code that you need to run inference with that model on your own hardware (typically from Huggingface.co).
That data includes not only the "weights" but also various files with required information, e.g. the tokenizer, the chat template, files that describe the structure of the "weights", e.g. number of layers, the number of "experts", routing information, etc. All this information may be distributed in many files (e.g. *.safetensors files with weights, *.json files etc.) or it may be aggregated in a single container file (with the .gguf extension).
You can see an example of the files included in a very simple open weights LLM here:
Bigger LLMs have much more files, especially much more *.safetensors files, which contain the "weights". The "weights", i.e. matrices of numbers that are used in the computational algorithm that generates the output tokens, constitute the bulk of the data needed to run a model, i.e. from a few gigabytes to a couple of terabytes, which is why the term "open-weights" is used, but in fact by this term it is understood that all data needed for running inference is open.
For an open weights LLM, you do not have access to the data set used for training the model or to the algorithms that have been used during the training of that model.
You can still do some fine-tuning of the model, using your own training methods and your own additional training data. To facilitate this, several open weights models offer not only a model version that can be used for inference to implement a chat application or an agentic workflow, but also a "base" or "raw" version that is not suitable for being used directly for inference but which is suitable for you to do a post-training/fine-tuning, to create a model more appropriate for your particular needs.
An "open weights" model is sufficient for most of the potential LLM users, because training a model is something that requires expertise, expensive hardware and a lot of time, so few would be able to do it even when given access to the necessary data.
Because I wanted native TypeScript support which Node only gained relatively recently and only partially, plus the issues I showed are regressions - they worked fine in earlier Deno versions.
> Node compatibility was a distraction for the product
I've gone back and forth on this point over the years.
Yeah, Node compat has probably affected the rest of the product. I imagine at some point they (or their investors) freaked out because adoption wasn't happening. And the reality is it really doesn't matter if your product is better when nobody is using it.
In retrospect I think it would have been a better decision to target Node compat from the start like Bun did. An impossible option at the time given Deno started trying to make something different from Node.
Or maybe the problem was simply they couldn't afford low adoption after having investors on board.
I never thought that, and I know very little about the space. In attempting to learn about the available options as fast as possible (along with TS and JS), I learned that the same guy launched a total re-do to address the shortcomings he realized through years of experience with his original product. I didn't imagine he could do that without breaking compatibility.
Distributed hash table - ButTorrent extension for discovering torrent's seeders by advertising its hash across known peer pool, think of it as a distributed tracker. Contrary to traditional way of asking a known tracker for peers of that torrent.
Its algorithm is very elegant, using binary search on peers' and torrents' hashes, narrowing down to peers that are more likely to be seeders (or at least know some).
I'm not sure why you're saying that in the HN comments. As far as I know the authors aren't there, and the paper is from 2010. It's not like it will get updated now.
In sans-serif text typefaces capital ‘I’ is traditionally a simple vertical rectangle without serifs.
But we were estimating some folks would miss those serifs so if you want to have a ‘properly-formed’ capital ‘I’ just activate stylistic set 02 (ss02, = ‘Alt I J i j r’). To see what this looks like, take a look here: https://www.monolisa.dev/tester/text
Cloning features and UI in your own product is not dishonorable. Outcompeting someone who didn’t bake in a moat isn’t doing anything wrong, or Burger King and Wendy’s shouldn’t exist.
Well at least they didn't have their product managers reach out to "start a conversation" like Google and Microsoft's who then blatantly rip off the product later.
The content of https://news.ycombinator.com/item?id=48628303 should have been posted as the item's text. Or, that text could've been put in to an Issue, and the link pointed to that.
"Before it, using a computer was synonymous with staring at the black MS-DOS screen. Want to play a game? You had to type annoying commands like cd C:\games\doom and pray the system wouldn't throw a conventional memory error."
Wrong, obviously, since Windows already existed. But then comes this odd statement:
"Windows 3.1, which came before, was just a shell on top of DOS. It was a messy pile of floating windows that easily got lost behind one another."
Also wrong. There wasn't anything fundamentally different about window management in 95. And Program Manager was a much better way to organize and access applications than the Start menu, which violated Microsoft's own guidelines for nested-menu depth.
The Start menu created a problem that has only gotten worse. Installers would create entire sub-menus by vendor name; so instead of looking for "TurboCAD," for example, you'd somehow be expected to look under "Imsi." Obnoxious.
For a while you could still put your applications into groups in the Start menu, if you knew where to go in the filesystem. So, get this, you could put your audio apps into one group, your dev tools in another, your graphics applications in another... INCREDIBLE, right? And some installers did the right thing and asked you where in the Start menu you wanted to app shortcut to be.
But today, organizing your apps is essentially impossible. Yeah, you can "pin" a limited number of them here and there, but still not in groups. The sad thing is that Apple has regressed on this too, with the deletion of Launchpad. Now you're wading through hundreds of applications.
I agree that Microsoft advanced the GUI more than anyone else in the '90s, but it has undone all of that and more by now. What a disaster the Windows UI is today.
Thanks. You used to be able to do it from the menu itself, though. I don't even remember the method...
And now the file system is just an irritating mess. Why are there mirrors of your home-directory structure, most of them "forbidden," littering the left pane? I waste so much time every hour of every day hunting down my most-used directories. Yes, I pinned some shortcuts on my desktop, but that means herding windows out of the way to get to them.
Finder blows, but somehow I navigate better with it now than I can with Explorer.
Neither Windows 3.1 nor Windows 95 were mere shells atop DOS. As early Windows 2.0 in 386 Enhanced Mode (Win/386), it was a virtual machine manager that took complete control of the machine and ran the Windows GUI and DOS in separate VMs.
Breakdown by segment (FY 2025):
Mac: $33.71 B — 8.10% of total revenue
iPad: $28.02 B — 6.73% of total revenue
reply