FSW WORLDWIDE

[Deleted member 1083629]

The most Canadian thing to do is to complain about how awful things are in Canada. Every immigrant will do the same thing after living over here for a few years. In reality, there’s no such thing as a perfect country. Every country has its own strengths and weaknesses. It’s up to the individual to decide what they value the most. For me personally, apart from the food and social life of my home country, India, there’s nothing that would make me go back. Canada is a million times better. I used to think the TTC was bad before visiting and using the NYC subway. The NYC subway, despite being situated in the richest city in the world and having the most amount of stations, the TTC is about a hundred times better in terms of frequency and cleanliness. Some of the subway stations in NYC are god awful and smell like shit. You would never get the impression that it’s the richest city in the world after visiting it. After seeing how run down and poor some of the cities and areas in New Jersey looked, I had a newfound appreciation for the run down areas of Canada. They are heaven in comparison. I realized that the west and specifically, the US is romanticized on television. Reality is far from what’s portrayed on television. I have a newfound appreciation of Canada after travelling to other places.

Coming from Moscow, I can talk so much shit about TTC lool. Anyone who visited/lived in Moscow, look at TTC like it's a pre-historic animal. Just google "Moscow Pushkinskaya subway". I am not even talking about the fact that the subway covers pretty much entire Moscow and Moscow regions.

 

[GandiBaat]

Of course, it will be fun. The biggest problem is that we need data from actual lawyers to see how they respond to PFLs. Don't forget that weekly there are new court cases that have an effect on how lawyers work. ChatGPT was trained on the data from 2021. No one has money to retrain that monster.

But I agree; it's a fun project.

All courtcases are public data. Most of them will have pfl as exhibits.

 

[GandiBaat]

All right folks, hit me with some positive things about Toronto. So far the discussions are depressing.

I will start - Picturesque fields just outside of Scarborough. Access to farms and fresh produce. Access to you name it grocery items.

Toronto : Constant reminder that you could have been living in Boston.

 

[Deleted member 1083629]

All courtcases are public data. Most of them will have pfl as exhibits.

Yes. I am not saying they are not. I am saying that for any AI model to give you up-to-date results, it needs to be retrained. Who will retrain ChatGPT?

 

[GandiBaat]

Yes. I am not saying they are not. I am saying that for any AI model to give you up-to-date results, it needs to be retrained. Who will retrain ChatGPT?

Not always. The "transformer" setup learns how to "transform". If you get a new pfl or decision, you can project it in its "embedding space" without need to retrain the entire network. And given how stupid canadian immigration laws are, we may just get away with a GPT2.

Heck I have even thought a name for that network too. DeepTHROAT or SUCKNet. Given how it sucks to deal with canadian immigration system.

 

[Deleted member 1083629]

Not always. The "transformer" setup learns how to "transform". If you get a new pfl or decision, you can project it in its "embedding space" without need to retrain the entire network. And given how stupid canadian immigration laws are, we may just get away with a GPT2.

Correct. I believe you are talking about warm-starting retraining. With how large ChatGPT, warm-start will still take money to update it. I am not talking about retraining it from ground 0; I am talking about talking it's 1.3 billion parameters (or how many are out there). Is it possible? Hell yeah. But it's expensive AF.

The biggest problem with PFLs is that recent law decisions can be crucial in responding. In order to get those latest cases and write responses with the latest information, it's important to always keep the model up to date. Thus, see my first paragraph.

 

[GandiBaat]

Correct. I believe you are talking about warm-starting retraining. With how large ChatGPT, warm-start will still take money to update it. I am not talking about retraining it from ground 0; I am talking about talking it's 1.3 billion parameters (or how many are out there). Is it possible? Hell yeah. But it's expensive AF.

Not really. You take a trained network but put your own data for decoding part. That is not too expensive. If everyone has to retrain or even domain-adapt the network, it will break the usefulness of the model for most customers except the ones with a lot of money.

 

[Deleted member 1083629]

Not really. You take a trained network but put your own data for decoding part. That is not too expensive. If everyone has to retrain or even domain-adapt the network, it will break the usefulness of the model for most customers except the ones with a lot of money.

I am arguing in Russian on Russian forum about some laws in Russia and then switch a tab and I have a conversation talking about ChatGPT to respond to PFL loool.

Anyways, I haven't seen ChatGPT to be used in any serious domains yet. If I start seeing that it's actually being used, I will take a look at it's architecture. IMO, in order for it to be useful in law, it needs to be updated. I assume "decoding" part means that one updates only the last layers. If so, then I can see it being useful in PFLs.

To be honest, I think that analyzing work experience letters to match NOC/TEER is much more useful use case for NLP. It's less risky -> more people will try it, it is easier to get it right (and faster) -> shorter time to market.

 

[GandiBaat]

To be honest, I think that analyzing work experience letters to match NOC/TEER is much more useful use case for NLP. It's less risky -> more people will try it, it is easier to get it right (and faster) -> shorter time to market.

That part is actually pretty easy.

That part requires distance metric learning. Train an embedding space in which samples of work experience letter from different NOCs are farther and same or similar NOC are nearer. Its how one way face recognition at scale works. I actual worked in a team that created a model to learn different vulgar and hate signs (gang signs, hate symbols etc) to pick-out images that may be illegal in some countries to even show. This is really how we did it. We trained it once and adding more signs was easy. Just project more samples and store their embeddings. Then we used FAISS do a fast similarity search.

I believe even a word2vec embedding will give you a good starting point for distance metric learning.

Ofcourse getting the dataset ready in first place was harder.

 

[Deleted member 1083629]

That part is actually pretty easy.

That part requires distance metric learning. Train an embedding space in which samples of work experience letter from different NOCs are farther and same or similar NOC are nearer. Its how one way face recognition at scale works. I actual worked in a team that created a model to learn different vulgar and hate signs (gang signs, hate symbols etc) to pick-out images that may be illegal in some countries to even show. This is really how we did it. We trained it once and adding more signs was easy. Just project more samples and store their embeddings. Then we used FAISS do a fast similarity search.

I believe even a word2vec embedding will give you a good starting point for distance metric learning.

Yup. More or less. I have built a simple NLP model but the hard part is always data. I got some letters from this forum and fed it in but to properly train it I would need more data. That's where ChatGPT can really shine; it's already trained and doesn't need new data to perform. Need to test it out this weekend to see how it works.

Damn it. Now I got all excited about coding and I won't be able to fall asleep lmaaao.

 

[Impatient Dankaroo]

lol. If I surround myself with rednecks, I might as well be paid top dollar for that. And Alberta ain't paying top dollar.

Doesn't tech have alot of remote opportunites?

 

[ASingh360]

I updated my job role as a Senior executive Title for foreign work experience whilst applying for a study permit back in 2020. Roles and duties are in accordance with NOC code B 6313. In my experience letter provided by the office, the Title is mentioned as team coach. No change in the duties. Can I upload this experience letter to count my experience or will it have any problem with regard to misrepresentation? Please suggest.

 

[sbali]

I updated my job role as a Senior executive Title for foreign work experience whilst applying for a study permit back in 2020. Roles and duties are in accordance with NOC code B 6313. In my experience letter provided by the office, the Title is mentioned as team coach. No change in the duties. Can I upload this experience letter to count my experience or will it have any problem with regard to misrepresentation? Please suggest.

No problem
Designations and title does not matter since there's no standardisation across companies / countries.

Hope it's clear

 

[Postgraworkperm]

Any draws today??

 

[AndyUK]

Any draws today??

There should be a draw today. Check in around 3-4 hours. The results are usually announced around 7-9pm GMT.