I'd like to add a few failure modes: - LLM removes/disables/weakens tests (disallowing manipulation of tests is not really possible in Python since the language is too dynamic, so the entire execution has to be…
> But on a tangent, why do you believe in mixture of experts? The fact that all big SoTA models use MoE is certainly a strong reason. They are more difficult to train, but the efficiency gains seem to be worth it. >…
Even with inflated RAM prices, you can buy a Strix Halo Mini PC with 128GB unified memory right now for less than 2k. It will run gpt-oss-120b (59 GB) at an acceptable 45+ tokens per second:…
Looks like the pop-ups that cover 70 % of my netbook's screen are still there: https://i.imgur.com/uPbVW2o.png I wonder if getting rid of those pop-ups would have slowed Stack Overflow's decline. The number of new posts…
To put it into your metaphor: I am not advocating against the existence of bandsaws. I would just rather have bandsaws that do not cut off your fingers if you do not read a book about them first and make it difficult to…
If most people are not using a tool properly, it is not their fault; it is the tool's fault. Git is better than what came before, and it might be the best at what it does, but that does not mean that it is good. - The…
> this code is not copyright protected, therefore you are not allowed to apply a MIT LICENSE to this project. Why not? You still can (and probably should) disclaim warranty and whether the code is copyright protected…
> Specifically, we collected new data created after January 2025, including: [...] new fiction on Archive of Our Own (Various, 2025), Not sure how to feel about this. From a researcher's point of view, reproducibility…
You can still be excited! Recently, GLM-OCR was released, which is a relatively small OCR model (2.5 GB unquantized) that can run on CPU with good quality. I've been using it to digitize various hand-written notes and…
Were there any particular challenges when implementing your library? I have implemented my own serialization library [1] (with a focus on not allowing arbitrary code execution), but had skipped dataclasses for now,…
A word of caution: There are SVGs which can freeze a page, so make sure that you do not link to any third party SVGs. This is a known bug, but both the Google Chrome and Mozilla team do not want to fix it. Here is an…
Here is another inference implementation in Python (only dependency is PyTorch). https://github.com/99991/SimpleTinyLlama The new checkpoints did not seem much better and they changed the chat format for some reason, so…
I found CLIP to be _amazing_ for all kinds of image search, like search-by-text or search-by-image. I even ported it to NumPy to understand it better. The whole thing is less than 500 lines of Python (including blank…
If you ever use the middle mouse button to open links in a new tab, you should disable this "feature" because it is a security vulnerability which allows for grabbing the clipboard content. Demo:…
Can't you just unplug the Ethernet cable/turn off the Wi-Fi module? (At least if the website was not lazy-loading resources.)
I also implemented a batched client-side web tool to resize photos to a certain file size (in a few hundred lines of JS). It only supports JPEG, but it resizes photos almost instantly…
There are SVGs which will crash most browsers, so SVGs could still be abused for denial-of-service attacks. For example, you could post one of those SVGs in every issue thread of a GitHub project if you wanted to mess…
I'd like to add a few failure modes: - LLM removes/disables/weakens tests (disallowing manipulation of tests is not really possible in Python since the language is too dynamic, so the entire execution has to be…
> But on a tangent, why do you believe in mixture of experts? The fact that all big SoTA models use MoE is certainly a strong reason. They are more difficult to train, but the efficiency gains seem to be worth it. >…
Even with inflated RAM prices, you can buy a Strix Halo Mini PC with 128GB unified memory right now for less than 2k. It will run gpt-oss-120b (59 GB) at an acceptable 45+ tokens per second:…
Looks like the pop-ups that cover 70 % of my netbook's screen are still there: https://i.imgur.com/uPbVW2o.png I wonder if getting rid of those pop-ups would have slowed Stack Overflow's decline. The number of new posts…
To put it into your metaphor: I am not advocating against the existence of bandsaws. I would just rather have bandsaws that do not cut off your fingers if you do not read a book about them first and make it difficult to…
If most people are not using a tool properly, it is not their fault; it is the tool's fault. Git is better than what came before, and it might be the best at what it does, but that does not mean that it is good. - The…
> this code is not copyright protected, therefore you are not allowed to apply a MIT LICENSE to this project. Why not? You still can (and probably should) disclaim warranty and whether the code is copyright protected…
> Specifically, we collected new data created after January 2025, including: [...] new fiction on Archive of Our Own (Various, 2025), Not sure how to feel about this. From a researcher's point of view, reproducibility…
You can still be excited! Recently, GLM-OCR was released, which is a relatively small OCR model (2.5 GB unquantized) that can run on CPU with good quality. I've been using it to digitize various hand-written notes and…
Were there any particular challenges when implementing your library? I have implemented my own serialization library [1] (with a focus on not allowing arbitrary code execution), but had skipped dataclasses for now,…
A word of caution: There are SVGs which can freeze a page, so make sure that you do not link to any third party SVGs. This is a known bug, but both the Google Chrome and Mozilla team do not want to fix it. Here is an…
Here is another inference implementation in Python (only dependency is PyTorch). https://github.com/99991/SimpleTinyLlama The new checkpoints did not seem much better and they changed the chat format for some reason, so…
I found CLIP to be _amazing_ for all kinds of image search, like search-by-text or search-by-image. I even ported it to NumPy to understand it better. The whole thing is less than 500 lines of Python (including blank…
If you ever use the middle mouse button to open links in a new tab, you should disable this "feature" because it is a security vulnerability which allows for grabbing the clipboard content. Demo:…
Can't you just unplug the Ethernet cable/turn off the Wi-Fi module? (At least if the website was not lazy-loading resources.)
I also implemented a batched client-side web tool to resize photos to a certain file size (in a few hundred lines of JS). It only supports JPEG, but it resizes photos almost instantly…
There are SVGs which will crash most browsers, so SVGs could still be abused for denial-of-service attacks. For example, you could post one of those SVGs in every issue thread of a GitHub project if you wanted to mess…