rocauc

Ronnie, I feel like you're just here for the zipline

I'd love a template for: 1. Select internal document sources you want to ingest (Notion/Slab, specific Slack channels, named Google Docs in a folder) 2. Enable Q&A + augmented retrieval in response 3. Serve to end users in a Slack channel at our company

Cloudflare’s blog is among my favorites. https://blog.cloudflare.com/

One of the places edge deployment (building on the post's definition of running on a distinct device closer to the user of the service) is booming is computer vision. There are many things to be seen in the world, and not all of them are near internet connection or able to wait for a roundtrip API call.

This breaks down edge vs cloud deployments (in the context of a computer vision machine learning model), and the tradeoffs are generally applicable: https://blog.roboflow.com/deploy-computer-vision-models/

a new, popular library for basic functionality (converting between annotation formats, evaluating models, doing object tracking) has been supervision: https://github.com/roboflow/supervision

This is really powerful. Can't wait till there's the ability to make custom detection blocks!

The OpenCV Competition produces famously neat projects like open source automated weed sprayers, sign language translators, and rocket projectile trackers [1].

[1] https://opencv.org/wp-content/uploads/2023/01/OpenCV-AI-Comp...

Roboflow (https://roboflow.com/careers) | Software Engineers, ML Research Engineer, Former Founders | Full-time | SF, NYC, Remote

We make tools that enable developers to make the world programmable. Over 250k developers, including those from half the Fortune 100, use our computer vision tools to improve datasets, models, and deployments. For example, Roboflow is Snap's partner for building custom vision models into AR lenses [1].

We have SF/NYC/distributed roles. Every team member also has an annual travel stipend [2] to spend coworking with others, anywhere.

[1] https://twitter.com/SnapAR/status/1671985144524165120/photo/... [2] https://blog.roboflow.com/remote-not-distant/

Good call out. Edit - FWIW, energy sector stocks are also down YTD (certainly correlated with lower prices of an inelastic product).

I also realized the linked sourced from Coatue hasn’t been on HN and is a good source for independent discussion like this. Submitted: https://news.ycombinator.com/item?id=36565261

While I welcome optimism, it’s premature to declare strong top line S&P 500 performance as the end of economic uncertainty. Tech is the only sector with significant YTD gains: Brent Crude Oil, Regional Banks, Transportation, and Retail are all down YTD.

In the S&P 500, the “Magnificent 7” stocks (NVIDIA, Apple, Google, Microsoft, Meta, Tesla, and Amazon) are responsible for *85%* of YTD gains.

Coatue’s East Meets West macro view from June 30 is the source for the above, and it contains a number of great insights: https://www.coatue.com/blog/company-update/coatues-2023-emw-...

Original release from Meta (April 17): https://ai.facebook.com/blog/dino-v2-computer-vision-self-su...

Demos: https://dinov2.metademolab.com/

YOLO-NAS outperforms prior SOTA YOLO models on an NVIDIA T4. It's notable that the Neural Architecture Search (NAS) makes these results hardware-specific for the mAP vs speed tradeoff. The license is more permissive license than some prior YOLO models: you can use it commercially if you train from scratch.

- Florida link - https://www.economist.com/the-economist-reads/2023/03/24/wha...

(noticed you accidentally linked to the Poker list twice and went looking)

Really fun project. Whisper's ability to generalize to novel sounds - like picking up the "Woo-hoos" in Weezer's Buddy Holly - continues to impress.

Asks for mechanics for these trucks are beginning to pop up in rural Iowa.

There's good work happening in this area, e.g. Sourcegraph is working on "Cody" to understand and search your code base https://twitter.com/beyang/status/1614895568949764096

@paulg has been restored

Really cool project. I've lived in DC for a bit and came across Helicopters of DC[1], whom have gamified spotting copters across the city.

The model they use is publicly available, too, to call via API for free: https://universe.roboflow.com/helicoptersofdc/helicopters-of...

Have you thought about collaborating with them on data/models?

[1] https://twitter.com/HelicoptersofDC

nice list. awesome that contributions can be submitted on github too: https://github.com/bleedingedgeai/bleedingedge.ai

"NRR Doesn't Matter [If Your New Customer Base Doubles Every Year / If Your ACV Doubles Every Year]"

NRR is exceptionally important for early stage startups. Retention is proof you're solving a real problem for your customers; NRR is like monetized high customer satisfaction.

Joseph Redmon was the original author of the YOLO family of models, up through YOLOv3. A maintainer of Darknet, Alexey Bochkovskiy (the framework for the original three YOLO models), published YOLOv4. Glenn Jocher used “YOLOv5” and showed the ML community that you can, but not without controversy[1], force a name into existence with adoption.

It appears the authors of YOLOv6 are aiming to employ a similar clever naming strategy.

I’m looking forward to more benchmarks before getting too excited.

[1] https://blog.roboflow.com/yolov4-versus-yolov5/

fantastic build. it’d be neat to complement this with collecting in-game statistics on which resources were collected by which player to further evaluate performance. I have a hunch the best players not only get access to the most resources but have a high ratio of points generated per resource collected

Wonderful - and welcomed sequel to the Chuck Norris facts API: https://api.chucknorris.io/

you know, Jared Polis would likely appreciate this.

Do you comprehend how the tool reduces labeling time in (4) and (5) (compared to labeling with eg CVAT) as the post claims?

Ah, yeah, that one liner resonates with teams establishing their vision infra. Teams build production vision models in the span of an afternoon with better tools, e.g. one dev built this Mountain Dew bottle model during Super Bowl half time.

Some ML engineers find value in things like automated annotation, testing model architectures (models.roboflow.com), and having one-click deploy for custom object detection APIs. Think of it like replacing all the one-off scripts so you can focus on your domain-specific problems instead of reinventing the wheel on vision infrastructure.

I was character limited - thank you

We make developer tools for computer vision so engineers can create models without being machine learning experts: collect data, organize images, annotate, train, deploy, and improve.

Makes a lot of sense, thanks. I'll need to dig deeper into Bayesian active learning techniques.

Nice read.

Can you shed some light on what you think are the most valuable methods for identifying high entropy examples for the model to learn faster? I'm familiar with Pool-Based Sampling, Stream-Based Selective Sampling, Membership Query Synthesis[1], but less certain which techniques are most useful in NLP.

[1] https://blog.roboflow.com/what-is-active-learning/