barbolo

It seems fixed. I've tested some Math ops that were previously failing and now are consistent on Chrome 150.0.7871.128 for linux and 150.0.7871.129 for macOS.

It was a sad surprise last week when we tried to upgrade the workspace AI plan for some of our team members to Ultra and it was gone. We're moving to Claude/Codex.

That's awesome. DINOv2 was the best image embedder until now.

I’ve been using DinoV2 for some months now. I’ve tried the models with 4 register tokens along with CLS + patch tokens. I’ve several embeddings (tokens) from previous model (no registers) which are part of my solution, so I didn’t adopt the newer “register” models because the CLS tokens are not aligned between 0 registers and 4 registers models. It would be nice if the CLS and patch tokens were somehow aligned between those models.

It's much worse than that. The leak contains dozens of datasets (relatives, addresses, jobs (+ linkedin), schools, vehicles, income, debts, pictures of faces, companies).

This leak is much more harmful. If the data really comes from Serasa Experian, they have more accurate and structured data from people/companies/assets in Brazil than anyone else.

Thanks, @rasmi. I have a feedback for you guys. The pricing for predictions inference in GCP is not very fair. If I deploy a small model (like a SqueezeNet or Mobilenet) I pay almost the same price of someone deploying large models (like Resnet or VGG). That’s why I’m deploying my models on serverless environments and paying about 5 dollars for 1 million inferences.

The pricing of GCP is: $0.10 per thousand predictions, plus $0.40 per hour. That’s more than 100 dollars for 1 million inferences.

Thanks for the answer. Currently I’m using AWS Lambda to deploy my TensorFlow models. But it’s pretty hard and hacky. I need to remove a considerable portion of the code base that is not needed for inference only routines. I do that so the code loads faster and to fit the deployment package size limit. If TensorFlow Lite is already a compressed code, then it may be much easier to deploy it to a serverless environment. I’ll be trying it in my next deployments.

Would that be a viable option to deploy TensorFlow models on serverless environments (Lambda, Functions)?

I’m reading over and over again since the last weekend. And I’m checking the code. And I’m still not understanding it.

Automating searches on a government website that decided to use reCAPTCHA just because it wants to look modern. There are dozens of them in Brazil for example.

Blocking audio means blocking visually impaired people from accessing websites with reCAPTCHA.

Hacking reCAPTCHA is not only for bad people. There are several use cases where solving reCAPTCHA automatically is needed.

Another interesting fact is that TensorFlow 1.4 supports native MFCC spectrogram tensors.

Nice work. It also publishes a file mfcc.py which uses Mel spectrogram to solve the audio offline. With enough data, a model based on MFCC should work much better than any cloud service (general speech recognizer).

Are you sure SES was the problem? Did you correctly configured DKIM/SPF?

Segmentation is not needed for a modern deep learning system, since this is learned by the neural network. It’s a solved problem for many handwritting recognition problems.

Which way to go: Chrome headless, Firefox headless or both?

Would be nice if it was possible to change the proxy settings for each request or for each session. Last time I checked it was only possible with the C API.

Every now and then I have to train models (mostly neural networks) that bypass captchas.

My company provides web automation in Brazil, not for spam, not for marketing purposes. Sometimes the client has a website (like an intranet) that she wants to automate through screen scraping. I know it sounds stupid, but that is pretty frequent.

The models are fast (most captchas solved under 100ms in AWS Lambda) and accurate (95 to 100%). I have about 40 different captchas being solved this way.

My team is still working on a solution to nocaptcha recaptcha. The problem for me is not solving the challenges, but submitting them like a human would do.

I'm progressing with chromium headless. As more and more sites add this captcha, I need to find a solution to automate it. And I'm confident I will find.

I agree with rkaplan. I've been working with many different visual problems and that comment is pretty consistent with what I've seen.

Go with cs231n or the neural networks and deep learning online book from Nielsen. Both valuable and introductory.

Sure you can. The term you are looking for is "multi task learning".

I'm partner of a B2B business.

We send cards and/or gifts on Christmas for every client. I and my partner write all the cards. We always thank the client for being with us another year and we ensure them we will do everything possible for the next year to be even better.

We haven't lost a client in the past two years. We have about 30.

You can sign up for the newsletter at http://www.mlyearning.org/. Andrew will be releasing more draft chapters in the next week.

I've already read the first 12 chapters. I found it very practical for developers who have already worked with NNs. Basically the book discusses common problems and how to deal with them. Gold resource.

If you are interested in deep learning or visual problems, I recommend the notes at:

http://cs231n.github.io/

Really great content from Andrej and his coworkers. This guy is great.

You can easily find all classes videos on YouTube too.

Finally, Intel makes a move in Deep Learning.

It doesn't sound spontaneous.

I don't like rules like "1 minute for this, 5 minutes for that".

But if that works for you, go ahead.

I'm a client of Bradesco, a popular bank in Brazil. I've chosen to migrate my account to a digital account called "Bradesco Prime Digital". The main communication channel with my account manager is WhatsApp, so he can take the day off.

I'm also selling my house and I'm pretty sure that every Real Estate agents in São Paulo use WhatsApp as the main communication channel. So no visits today.

I was having a conversation with the pediatrician of my children, so no answers today on what to do with my baby vomiting. It's not so serious to make a call, but it was fine to send a message.

My mother is 63 and sends WhatsApp messages every day to me, so no mommy messages today.

Well, it's not all that bad... I think today may be a very productive day for me.