HN user
Not them but my understanding is that the harness will send a simple 'heartbeat' message to keep the cache 'warm', (see prefix caching: https://handbook.modular.com/inference-optimization/prefix-c... ) which can then be edited/changed, which does cause the user to incur a fee, but its much less than the amount they'd pay on a no-cache hit request.
apple tracks every binary you execute and run on macos and sends the hash of it to their systems. Do you consider that to be respecting your privacy?
Yes, my point was more so the underlying behavior of automatic execution of binaries associated with/inside of a git project exists already within vscode as a pattern, and so seeing cursor doing the same wouldn't be surprising.
It’s a thing in VSCode as well/has been a thing or things similar to it : https://www.threatlocker.com/blog/malicious-vs-code-tasks-js... (2026)
https://www.reddit.com/r/programming/comments/zes1co/visual_... (2022)
Why make this comment without having tried it first? It very clearly is not useless and performs a lot better than one might expect. I am currently waiting to do more benchmarks of it in comparison to the full weight model, but it seems promising/better than Mistral Nemo at a lower file size.
Already happening with Workday in California:
https://news.bloomberglaw.com/litigation/workday-loses-bid-t...
It should also be noted Aikido’s original purpose was not for battle. It descended from Morihei Ueshiba studying Daito-ryu and wanting something less violent to practice after returning from war.
Cool project! I built something similar a while back and then kept adding to it: https://github.com/rmusser01/tldw_server It’s now grown quite a bit past the original goal of transcribing and summarizing conference talks for me.
Didn’t the DeepSeek team release a paper documenting inference improvements that showed they were still making a profit even under heavy discount? Why would it be impossible for them to make a profit now, with a new model and more research?
Before DeepSeek, no one sold cheap tokens anyways and then DS showed the profit margins.
Totally unrelated, but what would you say the feasibility of writing simulation software for simulation of/replicating body movements during/in a martial arts technique would be?
I’ve often thought it would be very handy to have a proper simulator for being able to simulate and identify inefficiencies in one’s technique, but no idea whether it would be feasible to do.
It was because of anti-monopoly legislation, https://www.investopedia.com/ask/answers/09/att-breakup-spin... ; i.e. breaking up 'Ma Bell' into the baby bells, and the current oligopoly we have instead in the modern day. An older picture showing how they were broken up and then re-merged: https://www.reddit.com/r/Futurelings/comments/r24ed0/diagram... or https://legalclarity.org/what-happened-to-the-baby-bells-aft...
This is one I use with no tracking or ads:
I’ve found this one to be useful offline with no ads:
migrate to rust.
Likely meant 'swyx', https://www.latent.space/podcast
If you don’t mind a little instability while I work out the bugs, might be interested in my project: https://github.com/rmusser01/tldw_server ; it’s not quite fully ready yet but the backend api is functional and has a full RAG system with a customizable and tweakable local-first ETL so you can use it without relying on any third party services.
You can absolutely downvote posts. You have to have a certain amount of karma before the option becomes available.
Yes. That is absolutely the case. One of the Most popular handguns does not have a safety switch that must be toggled before firing. (Glock series handguns)
If someone performs a negligent discharge, they are responsible, not Glock. It does have other safety mechanisms to prevent accidental fires not resulting from a trigger pull.
Just wanted to comment on the fact that I remember seeing that comment, and it left such an impression I remember it 7 years later. Thanks for the reminder, going to bookmark it this time.
Epyc Genoa CPU/Mobo + 700GB of DDR5 ram. The model is a MoE, so you don't need to stuff it all into VRAM, you can use a single 3090/5090 to hold the activated weights, and hold the remaining weights in DDR5 ram. Can see their deployment guide for reference here: https://github.com/kvcache-ai/ktransformers/blob/main/doc/en...
Mean time between failures
Are you doing this with vLLM? If you're using Llama.cpp/Ollama, you could likely see some pretty massive improvements.
Do you think you could expand on that? Like how you might imagine an ideal workflow to go? Would there be like a sea of tags that you could wade through, or just an '/all', only items you specifically subscribe to + your connections subscriptions according to some ranking algo? Items would 'fall off' or out of view(?) after X time/X amount of browsing a differently weighted topic?
I ask because being honest, you're a big inspiration for myself, and inadvertently, adding an LLM-curated RSS feed reader as a planned feature to a project I'm working on. (I saw https://github.com/karpathy/LLM101n when I was getting interested in LLMs, and then got inspired by your project to start to attempt to build something like the primer from the diamond age.
Where that leads is that I see an RSS feed reader + curation via self-described or identified interests as being a 'core' piece of information gathering for the 'future' individual and have had it on the to-do list as a feature-add for my project.
Hey yes, I’m building exactly that.
https://github.com/rmusser01/tldw
I first built a POC in gradio and am now rebuilding it as a FastAPI app. The media processing endpoints work but I’m still tweaking media ingestion to allow for syncing to clients(idea is to allow for client-first design). The GitHub doesn’t show any of the recent changes, but if you check back in 2-3 weeks, I think I’ll have the API version pushed to the main branch.
You could try my app https://github.com/rmusser01/tldw
Supports arbitrary length videos and also lets you choose what LLM API to use.
Not the person you asked, but it's dependent on what you're trying to chunk. I've written a standalone chunking library for an app I'm building: https://github.com/rmusser01/tldw/blob/main/App_Function_Lib...
It's setup so that you can perform whatever type of chunking you might prefer.
I've assumed that Google's approach for NotebookLM is similar to this, given their release of https://huggingface.co/google/gemma-7b-aps-it :
Gemma-APS is a generative model and a research tool for abstractive proposition segmentation (APS for short), a.k.a. claim extraction. Given a text passage, the model segments the content into the individual facts, statements, and ideas expressed in the text, and restates them in full sentences with small changes to the original text.
Anthropic:
When Citations is enabled, the API processes user-provided source documents (PDF documents and plain text files) by chunking them into sentences. These chunked sentences, along with user-provided context, are then passed to the model with the user's query.
Claude analyzes the query and generates a response that includes precise citations based on the provided chunks and context for any claims derived from the source material. Cited text will reference source documents to minimize hallucinations.
Ahh, thank you. For what it’s worth, I agree.
You’re making assumptions about my original question. I wished to know their opinion on what made NotebookLM ‘look like an Intern’s project. If what they shared was something I agreed with and relevant than sure I would apply it to my project but I already have plans for improving my project to my own standards.
Thanks for sharing your perspective though, I will keep it in mind. I disagree regarding podcasts being a ‘big thing’ past the initial honeymoon phase. I do think that custom generated audio is also a big thing and the podcasting is the first exposure a lot of people have had with that level of quality, and since it’s free and everyone already has a Google account, it makes it much easier for it to viral.