Technically speaking, this is a 3rd party app. Evidence is the "sign in with Google" saying the app is owned by Qwiklabs.
It took a bit of dig but they were acquired by Google. https://www.crunchbase.com/organization/qwiklabs. Strange that they haven't fully assimilated their stack / org into Google / GCP proper.
Edit: seems like they were acquired in 2016. On the one hand, it seems unacceptable that this team is going rogue and autonomous in 2023. Disclaimer: I worked at Google acquiree and worked on projects to integrate into Google proper. On the other hand, they must be frustrated with all the redtape of GCP. Disclaimer: I worked on GCP and also worked with the docs org there. The GCP docs org was the most toxic, incompetent and difficult people to work with during my time at Google. It would not surprise me if they were gatekeeping and blocking on this initiative. They ignored user feedback, and also shutdown any feedback from GCP developers and people who actually used and understood the product.
After trying to reach out to several of their key people on LinkedIn, I tried to reach out to them via a form on their site. This was to GIVE THEM SOME BUSINESS. Never heard anything back. This was for Telco related product that we have so not sure if it is just the Telco people in GCP who are this bad or the rot is wider than that.
Google Workspace acquisitions are already a huge pain for regular orgs. I imagine it takes even longer for such orgs to be assimilated into Google.com, especially since they might decide to utilize dev time to make changes that’ll ease the migration or otherwise enable more data to be directly assimilated.
When we were having a short in-person google cloud training session, we used Qwicklabs. From my understanding, we were trained by a Google contractor. I don't remember if the trainer described themselves as being from Qwiklab or Google.
In the same session, we were also introduced to AWS, and for that training we were trained by someone on the GovCloud team. The target for the training was in the field of proteomics and bioinformatics with machine learning. As a University there were on campus clusters we could use, but this was for using with smaller projects. Most people I knew ended up preferring AWS and sagemaker.
It was an open secret, and discussed off the record, that AWS will always give you better customer service than GCP, regardless of being a University scale client.
I don't think the team is going rogue, it's just another weird way in which Google refuses to provide direct customer support.
Google must comply with all sorts of regulations and compliance.
You get a lot out of the box if you used standard tooling. For example, privacy, GDPR and data retention compliance if you used standardized libraries. You also have dedicated teams who can streamline issues. It's the equivalent of registering your company in Delaware.
Yes, and in order to operate as a b2b SaaS, the "bare minimum" is relatively high, meaning when acquired the value of integration may not be very much.
The owners may have also just demanded to remain independent as part of the acquisition. Or the wind blew a little from the left on a Tuesday in July; there are a million reasons why Qwiklabs might not have fully rebranded or re-integrated such as you would like them to.
I agree in the general case for more complex apps. Firebase for example.
In this case, it seems the content is the real value rather than the app itself. Meaning, the app doesn't have too many roots that can't be uplifted. The main pieces I see are:
- registering as 1st party app ( a couple months with redtape )
- used a standard database ( shouldn't use more than a year )
- used standard auth ( should take less than a year )
- getting an approved ___domain name and real estate ( technically it's not a big of a change. it could be political, which brings me back to my original commentary)
In the worst case, it would take 2-3 years even at Google pace to rewrite the app. It's been 7 years.
> The app gains nothing by being in google proper.
Or Google gets sued and pays fines that exceed the value of the company. As evidenced in this thread, Google is increasingly losing goodwill and trust from tech people. This product that has good content but seemingly haphazardly rolled out isn't helping.
Maybe the solution is for Google to stop acquiring companies unless they have a streamlined pathway to integration and an environment which those companies thrive.
On the other hand they could take that 2–3 years of work and spend it doing anything else. I doubt someone’s highly incentivized professionally from migrating this over.
Here's my compilation of AI learning resources - I think some of the ones I've collected will be a better place to start for most people.
I categorized them into what kind of goal they're relevant for - building products, deploying custom models, or self study towards ai research science and research eng roles.
hey thanks, this is super helpful for me as a web dev trying taking time off work to hopefully transition to this space. I especially appreciated your descriptions of the roles.
That course is extremely good. Particularly the first six videos start from literally nothing (but basic math/coding), and build all the way up to AI models that were state of the art a matter of years ago.
The way Andrej explains things is brilliant - he'll write some code to visualize the data, then point out something that looks anomalous, suggest a possible cause, then write code to fix the problem, and then after it's all implemented say "oh by the way that function we just wrote is also a standard pytorch API". And you wind up understanding the API in a way you never would have if he'd started by introducing the API and then explaining what problem it solves.
The final video (on ChatGPT) skips ahead a fair bit, but is still a great explanation of how attention works. Incidentally I don't know any python but I had no trouble following along.
It’s really good! I especially like how he explains the code with the Attention is All You Need paper. His State of GPT talk at Microsoft is a great follow up. https://m.youtube.com/watch?v=s6zNXZaIiiI
I'm watching this now and maybe he explains in detail soon, but he starts off talking about loss functions and back propagation and tensors and this very much does not feel like a course for someone with python and vague memories of high school calc.
From the "toggle info panel" button on one of the courses.
> Can I take this course for free?
- When you enroll into most courses, you will be able to consume course materials like videos and documents for free. If a course consists of labs, you will need to purchase an individual subscription or credits to be able consume the labs. Labs can also be unlocked by any campaigns you participate in. All required activities in a course must be completed to be awarded the completion badge
Ah, the classic "California Corporate" style of speaking:
Q: Yes or no question about a specific case?
A: Four sentences of condescending and convoluted explanation about how the
rules work for the entire system, while simultaneously refusing to give a
clear "yes" or "no" to the specific case being asked about.
It seems passive-aggressive to demand that every single reader piece together the logic of their "labs" and "campaigns" and "badges" to figure out if a given course is free or not. I'm all for shipping early and often, but it does feel like adding 2 if statements would have made the whole pricing answer about 100x more user-friendly.
It's not that complicated. "Courses" are canned, one-way training material (comparison: books, lectures) that usually cost nothing. "Labs" are interactive exercises that require spinning up resources for each user (comparison: science labs with chemicals and beakers) so they usually cost something.
Yes, as a Qwiklabs user, you know this. But it is quite hard to decipher if you haven’t used the platform before. I agree with the GP that the current wording poses an unnecessary cognitive burden and should be more clear. Nothing wrong with e.g. “Mainly yes, but you’ll have to spend a few bucks on the hands-on labs”
Even after reading these comments I still don't know if the course is free or not. Either they charge for the labs, or the labs are free but require access to GCP resources. I still don't know which.
Fair point. The non-interactive parts are free. There are very few labs which require the purchase of credits.
The labs are plug & play with all required resources provided (and deleted afterwards), so you don’t need an own GCP account.
Because people too often jump to conclusions on hearing yes/no, imo. Not trying to defend the corporate speak though, obviously they should have put the yes/no up front and then try to explain later.
The first 8 courses cover only theory and are completely free. The last two require spinning up GCP compute resources and cost 1 credit and 3 credits respectively, where the baseline price of a Qwiklabs credit is $1 each.
Certain HN discourse are pretty useless. Anything remotely related to Google, Meta, MS, privacy/copyright is filled with often repeated tropes or completely orthogonal discussions. You have to ignore first 40% of comments to find anything meaningful.
First off, I doubt that. Second of all, that in of and itself is not a reason to discount everything and anything Google.
If Google created a way to end world hunger and did it for free would you automatically discount that because they removed 'don't be evil' from their code of conduct?
Fair enough. I didn’t mean to speak representative for the cohort. Just that we’re past the point of it being a legitimate market of identity. Nobody should be running around defending Google out of misplaced pride anymore.
And obviously you're taught how to deploy to Google cloud so you have to pay for the course, pay with your data—petsonal and work related—and pay for their cloud services...
I'm not sure what you're trying to build, but by the end of the second course, you should be able to create a customer service chatbot that is equivalent to what others have built. If you're interested in building/fine-tuning an LLM, that's totally beyond my knowledge.
MIT posts their AI/ML Degree requirements online, as well as the courses, for free. Shouldn't take you more than a year to finish it and start reading research papers.
This learning path is pretty thin. Here's what I used to go from a Physics undergrad to doing research in industry (not in a flashy company, but doing original research and modeling nonetheless):
Looks like the examples use TensorFlow, my understanding is that outside Google, most of the industry has centered around PyTorch for now. Do any practitioners have any opinions on which it's more valuable to learn and whether that makes a difference?
Everyone uses PyTorch both in industry and academic research.
Some labs use JAX/Flax.
Companies that ships ML products often use Tensorflow.
I want to say that it doesn't matter much which one you start with, but you will have to learn PyTorch anyway- in the future- if you are serious about DL.
And the experience of working with PyTorch is Astronomical Units better than Tensorflow.
PyTorch is much more pleasant to work with and you can do really custom stuff pretty easily.
Does anyone have recommendations for introductory text to deep learning for someone who learned math up through multivariable calculus/diff eq in school?
The books I’ve found so far are either too general (not focused on deep learning) or hand-waive over the math.
Learn Machine Learning first. Andrew Ng's course is great. Then learn Deep Learning through Andrew NG's course on Coursera.
Then go through fast.ai for practical projects.
Learn PyTorch well. There are many books around. I like the one from Manning and the one by Sebastian Raschka.
Then chart your own path from there.
Many people "learned Math" in college, but these were actually mostly mindless following of algorithms to solve problems by hand. If you want proper refreshers, go through ICL's Mathematics for Machine Learning Specialization on Coursera or 3blue1brown's series on Linear Algebra and Calculus.
As someone who self-taught essentially this over the past couple years, I'm not sure what benefit it would bring. Both developing and deploying models are already domains well established enough to not be covered by a couple courses (I'll be honest it doesn't seem like the material is 9 courses worth). You can hire an 'AI Engineer' or freely use an open source SOTA model developed by a team of PhDs.
SQL and XGBoost will solve 95% of your business problems. It's boring but it's true.
If you're trying to create the Terminator for real, that's when you start looking at the JAX/TensorFlow/PyTorch docs. I started with that first, and paid attention to the math along the way. If you go math-first you can quickly lose the forest for the trees. But you can find pretty in-depth tutorials and source code for any of those frameworks (and the math) on Google.
I find it weird that there are no images on this page, and they only use the colors white and blue.. I swear, the google design team is ignoring everyone who enjoys seeing more than 2-3 colors on a page. Icons with colors! Icons with colors! please. Azure knocks this concept out of the park, with AWS coming in a close second.
Personally couldn't care less what colors they use or didn't use. As long as the material is valuable. I am here to learn, not to take pictures like a foodie and admire colors.
For me, the colors they have on there are great. I can focus without distractions of moving backgrounds, gradient colors etc.
This is pretty cool. There are good video lectures around for much of the transformer parts (the huggingface stuff, other ml youtubers, etc) but deep dive stuff like this is appreciated.
It's worth noting the actual videos are just unlisted youtube links so you can add them to playlists.
Qwiklabs is a Google company. It handles most if not all of their technical training that has labs. It has been a Google company for more than 7 years.
Oh man I am with you on this in that, the fact that it's not a phishing site, to me, means google is sending the completely wrong message to users. Here is a site with a non-google.com URL that expects users to know that only google owns that TLD, has a non Google sign in page, uses google logo. Eek.
They literally are using their own TLD, ".google".
The comment you replied to is surprised that they did use their own TLD, thinking they should stick with their classic ___domain on not their own TLD (google.com)
the sign-in flow directs you to a normal google.com sign-in, or you can use the sign-up form on this site just like any other non-google site - they're not asking you to enter your google account credentials here.
It is Google's e-learning platform that provides courses and labs (step-by-step trainings in Google Cloud). Before cloudskillboost it was called Qwiklabs which was acquired by Google a couple of years ago. It is completely legit and has a OAuth login flow using your Google Identity.
For Google partners the trainings are free.
(Source: working at a Google Cloud partner)
Its very weird for Google to be charging for this yes. Definitely pricked my ears up but can’t see how this is spoofed unless google.com has been pwnd.
So you trust Google can afford Google.com but you specifically think a TLD of .Google is phishing and not literally the first TLD they would buy when the proliferation of new custom TLDs came out years ago?
Its a normal sign in flow. A safe way just in case is to log in to Google normally, close that tab, open the posted link then click the links in the oauth flow without entering credentials.
A normal sign in flow, that is not google.com (or a subdomain), or an oauth sign in screen, presenting as google, seemingly asking for my google creds? But is not google.com or affiliated. I'm just supposed to...trust it's google?
Again, seems sketchy. Even if not actually sketchy (I'm obviously unwilling to try).
This page doesn't fit on my (old and crappy) Android phone. The cookie banner overflows horizontally on the homepage, and the course pages don't fit at all.
I understand that it's hard to make a webpage that response well on all devices but you'd think Google could manage it.
It took a bit of dig but they were acquired by Google. https://www.crunchbase.com/organization/qwiklabs. Strange that they haven't fully assimilated their stack / org into Google / GCP proper.
Edit: seems like they were acquired in 2016. On the one hand, it seems unacceptable that this team is going rogue and autonomous in 2023. Disclaimer: I worked at Google acquiree and worked on projects to integrate into Google proper. On the other hand, they must be frustrated with all the redtape of GCP. Disclaimer: I worked on GCP and also worked with the docs org there. The GCP docs org was the most toxic, incompetent and difficult people to work with during my time at Google. It would not surprise me if they were gatekeeping and blocking on this initiative. They ignored user feedback, and also shutdown any feedback from GCP developers and people who actually used and understood the product.