IEatPrompts

IEatPrompts on Hacker News

Awesome List on AI for Security

Posted by IEatPrompts on Wednesday April 30, 2025 @05:29AM
from the github.com dept.

https://github.com/AmanPriyanshu/Awesome-AI-For-Security

( Read More... | no comments | 12 points | on HN )

Browser-Based Federated Learning Game with MNIST

Posted by IEatPrompts on Sunday November 24, 2024 @12:33AM
from the amanpriyanshu.github.io dept.

https://amanpriyanshu.github.io/FL-Interactive-Game/

( Read More... | no comments | 4 points | on HN )

New Dataset: RedPajama Dynamic Topic Modeling, 100K Docs W Topic Heirarchies

Posted by IEatPrompts on Monday November 11, 2024 @12:16PM
from the huggingface.co dept.

https://huggingface.co/datasets/AmanPriyanshu/Dynamic-Topic-RedPajama-Data-1T-100k-SubSample-max-1k-tokens

( Read More... | no comments | on HN )

Benchmarking Linear Complexity Multiplication for Time over Cosine Similarity

Posted by IEatPrompts on Monday October 21, 2024 @01:15PM
from the amanpriyanshu.github.io dept.

https://amanpriyanshu.github.io/blogs/posts/2024/linear-cosine/

( Read More... | no comments | 2 points | on HN )

Re:Ask HN: Tech Stack Behind the Claude App? on Saturday October 12, 2024 @01:49AM (#41816757)

That's interesting. Was thinking about how simple it feels as if the app is simply visiting the website

Re:Ask HN: Tech Stack Behind the Claude App? on Saturday October 12, 2024 @01:48AM (#41816747)

Agreed, seems really simplistic and refreshes when one switched apps

Ask HN: Tech Stack Behind the Claude App?

Posted by IEatPrompts on Thursday October 10, 2024 @01:07AM
from the ask-hn dept.

What's the tech stack behind the claude app and why does it feel like a chrome/browser running the website?

( Read More... | 9 comments | 13 points | on HN )

AdaptKeyBERT: Reflections on Two Years of Adaptive Keyword Extraction

Posted by IEatPrompts on Sunday September 22, 2024 @08:13PM
from the amanpriyanshu.github.io dept.

https://amanpriyanshu.github.io/blogs/posts/2024/adaptkeybert/

( Read More... | no comments | 2 points | on HN )

Re:API-LLM-Hub: In-Browser LLM Integration for Static Websites on Thursday September 19, 2024 @02:02PM (#41594524)

Thanks a lot!! Really appreciate the comment

Re:API-LLM-Hub: In-Browser LLM Integration for Static Websites on Thursday September 19, 2024 @02:01PM (#41594520)

Yup, it's another alternative to using the regular API. The issue was that a lot of the providers required you to build their libs and caused a lot of friction running on standalone vanilla JS. This aims to compile all those solutions and package it.

Re:API-LLM-Hub: In-Browser LLM Integration for Static Websites on Thursday September 19, 2024 @01:59PM (#41594501)

Ya, had a lot of trouble figuring out Anthropic as well. They don't have examples for overcoming their CORS issue.

API-LLM-Hub: In-Browser LLM Integration for Static Websites

Posted by IEatPrompts on Wednesday September 18, 2024 @06:49PM
from the amanpriyanshu.github.io dept.

https://amanpriyanshu.github.io/API-LLM-Hub/

( Read More... | 10 comments | 3 points | on HN )

Re:Show HN: Linguistic Forests Mapping YC Pitch Patterns with Grammar ... on Monday September 16, 2024 @05:56PM (#41561443)

Hi breck, I think that's focused on primarily the recent batches W24 & S24. Glad you found it interesting.

Show HN: Linguistic Forests Mapping YC Pitch Patterns with Grammar Trees

Posted by IEatPrompts on Friday September 13, 2024 @08:29AM
from the show-hn dept.

https://amanpriyanshu.github.io/YC-Dendrolinguistics/

( Read More... | 2 comments | 4 points | on HN )

Paper on turning single-shot jailbreaks to multi-shot jailbreaks

Posted by IEatPrompts on Tuesday September 03, 2024 @06:55PM
from the arxiv.org dept.

http://www.arxiv.org/abs/2408.16163

( Read More... | no comments | 2 points | on HN )

Meta's prompt guard defeated using spaces

Posted by IEatPrompts on Wednesday July 31, 2024 @05:24AM
from the theregister.com dept.

https://www.theregister.com/2024/07/29/meta_ai_safety/

( Read More... | 1 comment | 4 points | on HN )

Re:Meta's prompt guard defeated using spaces on Wednesday July 31, 2024 @05:24AM (#41117591)

Meta's new prompt-guard-86M normally flags almost everything as a jailbreak, but apparently spacing out letters makes it see prompts as harmless. Pretty weird way they found this - instead of hammering it with jailbreaks, they just compared embedding weights with the non fine-tuned model.