nutanc

Did you just say "All y'all look alike!" to non white people?

Very good explanation of the current state of the art in STT. I have also personally gone down this rabbit hole(especially the CTC bit) and agree 100% with what the author says. My personal viewpoints:

1. To have an ImageNet moment, we should have an ImageNet. LibriSpeech doesn't cut it. We should have an equivalent SpeechNet. Problem is visual language is one, while speech languages are many. So we should have an ImageNet for every language? And train for every language?

2. What is an ImageNet moment? Personally, though ImageNet has contributed a lot, similar to speech and NLP, even vision applications have promised more and delivered less in real world scenarios. And just like speech and NLP, only the big boys actually provide solutions in vision which shows in vision also they have access to better data which they don't share.

I would do the following:

- Manually scan through a couple of hours of data and setup a human baseline.

- Run standard algorithms and find their accuracy.

- Find errors in the model and analyze why the errors are happening. Is the model classifying some other object as a supervisor? Is the model not classifying the supervisor in certain lighting conditions or scenarios.

- Retrain the model with the failure scenarios so that it learns.

This is a good balanced article that gets a lot of things right. We should take a forgiving approach when we talk about AI systems. And as the author points out the problem is not that AI systems dont have understanding yet. The problem is with the hype which leads many to believe that we are close to building systems which can understand us.

That said, I have a small problem with the examples presented to say that already machines understand us :)

The article says 'For example, when I tell Siri “Call Carol” and it dials the correct number, you will have a hard time convincing me that Siri did not understand my request"

Let me try to take a shot at trying to explain that Siri did not "understand" your request.

Siri was waiting for a command and executed the best command that matched. Which is, make a phone call.

It did not understand what you meant because it did not take the whole environment into consideration. What if Carol was just in the other room. A human would maybe just shout "hey Carol, Thomas is asking you to come", instead of making a phone call.

If listening to a request and executing a command is understanding, then computers have been understanding us for a long time. Even without the latest advances in AI.

How much of the $1B is in Azure credits?

Hi, thanks for taking the time to explain in detail why you think the proof is wrong. I think this kind of discussion will go a long way in trying to find the solution. Will pass on the comment to the prof. He will like that someone took the time to critique. So far most of the time all he gets is,"the proof is not valid because no one has done it so far" :).

Nice detailed lectures on RH along with a suggested proof is here https://www.youtube.com/playlist?list=PLRsxymPrOKUAk3eXhK9FZ...

True that :)

The death of the phone call has been heralded for many years now. I am guessing it's gonna stick on for some more time, at least for business use cases.

We at Ozonetel have been in the cloud business phone system business for sometime now. A couple of quick observations:

1. Surprisingly, the Calljoy website does not have a call in number or a demo number. This is supposed to be a disruptive technology, put a demo number that people can call in and test.

2. The Google brand is not there anywhere on the Calljoy website(except at the very bottom).

3. Rather than release a free beta product, which is the norm, Google has launched a paid commercial product.

4. Haven't tested the call quality yet. I think they maybe launching this on the Google voice platform.

5. Not sure why it should be a separate product, would have made more sense as a Gsuite add on.

6. Businesses need support. The success of this product will depend on the 24/7 support that Google can provide to business owners.

It is surprising why Google just didn't offer this as a service. You could already build something like this using platforms like Twilio, KOOKOO(in India) etc. We at Ozonetel, already provide a business phone service similar to this for our customers using Google speech as one speech engine, though spam detection is not yet inbuilt. After looking at the launch, I just quickly put together a bot page if someone wants to try it out, https://ering.me/ozonetel/spa/

Note:I am a co founder of KOOKOO.

Thanks. Will share the details with my prof.

Agree with this. Don't think there are too many mathematicians who can really argue for or against a really important theorem. My prof has come up with one for Riemann hypothesis[1], but he is not finding anyone willing to engage in a discussion. Most mathematicians seem to believe that the Riemann hypothesis is unsolvable and I don't think it helps that my prof is currently not associated with any reputable university. It's as if proofs can come from only reputable universities and reputable mathematicians.

[1]https://medium.com/@nutanc/the-final-and-exhaustive-proof-of...

Though this talks about current trends, I would place my bets on a more radical future where the current algorithms for AI are overhauled and we get much better and faster algorithms which can even work on generic CPUs.

Thanks for the prompt response. Do you use data augmentation? Sorry for all the questions which could maybe be answered by looking at the code, but I am just getting started with pytorch etc and I am trying to understand the exact input format.

Does the solution use pre trained Resnet models?

TL;DR, Take your Twilio account. Get a Kookoo account. Copy settings and you have a full fledged cloud contact center on Twilio.

See you all at Signal Conference :)

Yes, we do that, but not on this product. This product is specific to service the US businesses. If you have some time, please check out our company site which lists all our products, http://ozonetel.com

This is something we discussed. Thats why $19.90 is an introductory price. We are actually losing money on that. But since our India business is profitable, we are offsetting that for the first few customers. The actual price is ~$50

Thanks for the detailed analysis. Will take this into consideration. Did not understand the point about taking care of the workers. Did you mean the employees? If so, then yeah, we have had almost zero attrition. There are people who have stuck with us for the past 8 years. So guess, they are well taken care of :)

But how would I present that on the website? Any ideas? Thanks!

Since you took time to give some inputs, please do let me know if you need a business phone number. Free for you :)

They are very happy customers :) and we have more too, HDFC life, Zomato, BigBasket, Practo, Intuit, HUL etc. But they are all Indian customers, so not sure if their testimonials will have real value for US customers. But I think your suggestion is good. Let me A/B test with that. Thanks.

This certainly helped. Thanks for taking the time to reply.

Hey, thanks a lot for the useful input. Will work on this now.

Agree with this. We have a successful cloud call center product Cloudagent in India and it has an enterprisy pricing structure. We wanted to launch in the US this year and we came up with a new simplified product getkookoo.com and went with a drastic price change to capture the market. We removed per agent pricing and said a flat $19.9 for unlimited agents and pay for only the calls you make. We thought the super simple and cheaper pricing is all we need. Turns out, its not. Though we have got customers, its been harder than we thought. In our case, it looks like the second point above, trust, is what we have not been able to communicate.

So yeah, pricing is just one very small factor. It is a factor, but only after all other factors have been taken care of :)

What are the odds, wrote this blog exactly 5 years ago,to the day, http://blog.kookoo.in/2012/04/wedding-ivr.html

This was built using KooKoo, the Twilio for India and is more of an IVR rather than SMS as India is a voice friendly country, and who does not like to hear the invite in the groom's own voice :)

Its got options for directions etc too, people always seem to call to know the directions to marriage hall :)

Yes,most providers also provide an option to route the call to physical phones too. But there will be extra call charges for such routing.

Thats why we have removed per agent pricing and charge only for minutes[1]. After some scale, it might make sense to go for fixed agent rental. But to start of,its better to have no agent pricing.

[1] http://getkookoo.com/getting-saas-pricing-right-pricing-clou...

Absolutely right. Though in India we have been charging the regular call center model for our cloud call center, http://cloudagent.in, we knew the model will not continue for long, especially in the US. So for our US product getkookoo.com we did away with the per agent pricing. We went with a zero agent rental and fixed monthly cost model with free minutes included[1]. [1]-http://getkookoo.com/getting-saas-pricing-right-pricing-clou...

Inspired by this blog by DHH, https://m.signalvnoise.com/why-we-never-sold-basecamp-by-the...

Zendesk for email

Intercom for chat

And we dogfood getkookoo for phone support