Show HN: I built a Plant Identification Tool Powered GPT4 Vision https://ift.tt/tp3FDd2

Show HN: I built a Plant Identification Tool Powered GPT4 Vision https://ift.tt/pZVDmBx April 30, 2024 at 09:53PM

Show HN: I replicated Anthropic's monosemanticity research using just my MacBook https://ift.tt/xs2cRNG

Show HN: I replicated Anthropic's monosemanticity research using just my MacBook Hi everyone, I've been working on an open-source implementation of Anthropic's research on monosemanticity ("Towards Monosemanticity"). The problem Anthropic is trying to solve is that language models are hard to interpret because individual neurons can be responsible for multiple different things. The research finds that training a small autoencoder on neuron activations can result in "features" which are much easier to interpret. When I was reading the original research, I got really excited when I realized that the models they used were really small, and I could probably train them from scratch with just my M3 MBP. My models are somewhat undertrained compared to what Anthropic produced, but I think my results are still very compelling. Let me know what you think! https://ift.tt/CEcGWPS April 30, 2024 at 10:56PM

Community-written abstracts for research papers https://ift.tt/tvcF6nW

Community-written abstracts for research papers https://ift.tt/uLWf9DN April 30, 2024 at 11:04AM

New Parking Enforcement Effort to Focus on Safer Sidewalks and Streets

New Parking Enforcement Effort to Focus on Safer Sidewalks and Streets
By Madhu Unnikrishnan

Keeping cars off our sidewalks and bikeways will help make it safer for everyone to get around. 

This week, we will begin a focused parking enforcement plan to help make sidewalks and streets safer for all San Franciscans. We’ll carry out this work on a rotating basis in each supervisor’s district. Our goal is to limit violations of existing parking regulations. 

Improving safety for people who walk, bike, roll and drive 

The new Neighborhood Operations Plan will prioritize enforcing safety-related violations. These include parking on the sidewalk, in bike lanes and crosswalks. This way, we can help ensure that people who walk, bike and roll on the city’s sidewalks and streets do not have to enter traffic lanes to get where they need to go.  

Our work will improve safety for people using mobility devices like wheelchairs and crutches. It will also help people pushing children in strollers. People who drive will also find it easier to navigate city streets. The Neighborhood Operations Plan will ensure the traffic lanes for cars are safer.  

Our plan will help make travel safer for people who walk, bike, roll or use mobility devices.  

Working with city partners to create safer conditions 

The plan aligns with Mayor London Breed’s commitments for the next phase of Vision Zero, San Francisco’s policy to prevent traffic fatalities and severe injuries. Visit the Vision Zero program webpage for more information. 

As we reach normal staffing levels for parking control officers, we're in a better position to enforce the city's parking laws. 

Moreover, city leaders with this plan are responding to requests from residents. “I am stepping up enforcement of our laws, because that’s what residents deserve and that’s what our city needs,” Mayor Breed said in her State of the City address earlier this year. 

The operation will deploy parking control officers to each supervisor’s district for one week of enhanced enforcement. These officers will also carry out their regular duties.  

We are in ongoing discussions with the Mayor’s office and each member of the Board of Supervisors on how we will implement the plan for each district. 

Our parking control officers work hard to keep streets and sidewalks safe.

Helping people park safely: key guidelines to follow  

So what does that mean for people who park on the city’s streets?  

In short, if you obey existing laws, there will be no difference. For a guide on parking rules in the city, visit our How to Park Legally in San Francisco webpage. 

As a refresher, here are guidelines for parking in the city:  

• Do not block driveways or crosswalks. A driveway begins at the “curb cut.” That’s where the driveway begins to slope downward toward street level. Residents can park parallel along the street in front of their own driveways if the driveway serves one to two units and the vehicle is registered to the address. 

• Do not park on sidewalks, even if the pedestrian path is partly clear or if a vehicle is parked across a driveway. A sidewalk citation can be given even if the pedestrian travel path is partly clear or if the vehicle is parked across a driveway. For more on the city’s parking regulations, visit the San Francisco Planning Department’s Code Enforcement webpage.  To check your sidewalk’s width, please visit the Department of Public Works’ grade map.  

• Do not obstruct bikeways. 

The goal of the Neighborhood Operations Plan is not to punish. It’s in response to residents’ and city leaders’ desire to enhance the safety of our streets and sidewalks.  

By preventing parking on sidewalks and bikeways, we make it safer for everyone to get where they need to go. People who walk, bike, roll and use mobility devices can stay in the part of the streetscape that’s safest for their use. People who drive can access clear, safe traffic lanes. 

Published April 30, 2024 at 02:59AM
https://ift.tt/bgqOVNu

Show HN: Attorch – PyTorch's nn module written in Python using OpenAI's Triton https://ift.tt/JzFRyf3

Show HN: Attorch – PyTorch's nn module written in Python using OpenAI's Triton attorch is a subset of PyTorch's nn module, written purely in Python using OpenAI's Triton. Its goal is to be an easily hackable, self-contained, and readable collection of neural network modules whilst maintaining or improving upon the efficiency of PyTorch. In other words, it intends to be a forkable project endowed with a simple, intuitive design that can serve as an accessible starting point for those who are seeking to develop custom deep learning operations but are not satisfied with the speed of a pure PyTorch implementation and do not have the technical expertise or resources to write CUDA kernels. There already exist a number of wonderful PyTorch-like frameworks powered by Triton, but most concentrate solely on Transformers and NLP applications, whereas attorch aims to be more inclusive by also presenting a variety of layers pertaining to areas besides NLP such as computer vision. Moreover, attorch is not an inference-only package and fully supports both forward and backward passes, meaning it can be used during training as well as inference, though its performance for the latter is generally not on par with dedicated inference engines. Questions and feedback are welcome in the comments sections. https://ift.tt/vsRWMr2 April 30, 2024 at 01:07AM

Show HN: Kaytu – Optimizing cloud costs using actual usage data https://ift.tt/GljxafX

Show HN: Kaytu – Optimizing cloud costs using actual usage data Reduce your cloud costs - SREs/DevOps/Cloud Engineers Hi community! We are Kaytu (“Kay-two,” named after the K2 mountain), and we've developed an open-source tool for engineering, DevOps, and SRE teams to reduce cloud costs. Cloud inflation (“cloud-flation”) is real—AWS EC2 costs are up 23% (4-5x global inflation average [1]), and 30% of the capacity that is paid for is simply wasted ([2]). The best way to improve cloud utilization is by simplifying the process so engineers can spot inefficiencies and suggest changes. We built a simple open-source CLI tool that recommends a cost-optimal workload based on actual usage data from observability tools. Check it out at https://ift.tt/iRhWqdV Currently, we support AWS EC2 On-Demand Servers & EBS Storage using observability data from CloudWatch to determine utilization. You can optimize EC2 Servers based on CPU, Network, Memory, and Storage. We're expanding support to include OS License, GPU metrics, RDS, and Prometheus integration, and we plan to add more AWS services like EKS and OpenSearch, as well as Azure. This is more than just a utility—we want to provide a no-nonsense platform that makes it ridiculously easy for engineers to build cost-effective apps on the cloud by optimizing workload configurations and customizing to scenarios. Open Core: Inspired by Sid Sijbrandij and GitLab, we've open-sourced our CLI and are actively working on the server side. Our tooling will always remain straightforward and support open-source tools for free. We made it as simple as possible to try out - it’s one command, no sign-up needed, no SaaS platform to share your credentials. We would love you to try it out and give us your feedback! If there are bugs, we would greatly appreciate it if you reported them on GitHub. Cheers, The Kaytu Team (Anil, Arta, Mahan, and Saleh) References: [1]Tangoe IT Trends Savings Recommendations and Liftr Insights data Cloud Pricing [2] Flexera State of Cloud Report - Multiple reports spanning 2017-2023 https://ift.tt/iRhWqdV April 29, 2024 at 09:27PM

Show HN: Bard PDF – Chat with Pdf in Google Bard or Gemini https://ift.tt/o8q7Fuh

Show HN: Bard PDF – Chat with Pdf in Google Bard or Gemini Chat with pdf in Google Bard or Gemini for free. Several ways to have conversations with pdfs in Google Bard or Gemini. https://bardpdf.dev April 29, 2024 at 08:31AM