Because some of you asked, here is a discount code that will get you 50% off the annual price: ALEXZ . Note that it bypasses the 90-day trial if you use it.
That's awesome! Thanks for the discount code $72 for a year! Wish I saw your code before I signed up for the 90 day trial.. Hmm.. Maybe just use a different email address :)
JetBrains have recently introduced full line completions in their IDEs which use local models, would be interesting to see how this performs against Tabnine (obviously though it won't generate multi-line completions)
Permissive license doesn’t mean „free to train your AI with“. You usually have to include the license and copyright, which I have yet to see on AI assistants.
Copyright currently (according to precedent set so far) indicates that if you made the code public for people to read, then it's fine to train on it. They do not need your consent to read it, so they don't need your consent to train on it. Those running inference, however, the people sending prompts and getting results, which may sometimes be substantially similar to copyrighted code, they may be infringing if they save that code even if they now have absolutely no way to realistically know it came from you. So it is a legal nightmare, but not very likely to have anything to do on the training side, is more on the usage side. But all of this is just a demonstration about the insanity of information monopolies. Information cannot be owned. It cannot be contained. It exists in the minds of other people once they read it. So having a monopoly that says you can control any kind of information that can exist in substrate including the human mind is a right to some limited control on everyone else. It is ludicrous, and philosophically incoherent.
@@ThaitopYT Not really. The MIT license requires including a copy of the terms of the license and a copyright notice with all copies of the software or its substantial portions. The zlib license requires that the source of the software isn't misrepresented. Similar stories with BSD and Apache.
Thank you - I work for a fortune 50 company and we aren't permitted to use Generative AI while on our network for security reasons. This is something I'm going to check out.
@@user-ic6xf Why? I'm a noobie so forgive for this question...If there is a tool that gives you the opportunity to complete a task, wouldn't it be worth it?
@@TheXabl0 until a full risk assessment can be detailed. There's numerous cases where data is leaked and sensitivity data and algorithms are being stolen. I wouldn't trust OpenAI as far as I throw a ten ton elephant.
Sure point it out, but at the same time this is basic. The GPT model cannot run on your computer. It's not powerful enough and it has never claimed to do so. Thinking that an extension will be able to download and set up a massive AI model in your machine is kind of ridiculous
Y'all don't know the latest models then. A lot are much better than copilot/GPT4. And they run locally. Our computers are powerful enough (well depending on your budget and what you buy). The next Snapdragon Elite will destroy everything in AI task for a very reasonable price(45TOPS, vs something like 30TOPS for the much more expensive M4)
Actually a lot of people in embedded still use Eclipse. There are some great addons that make cross compilation easier. Yes, you can do it with VS Code, but it takes so long to setup and you have to maintain that environment yourself. Meanwhile your microcontroller vendor provides and maintains an Eclipse based IDE that works well enough. Not great, but well enough.
@@AZisk for embedded almost all main MCU manufactures still use Eclipse as a reference IDE from them. I haven't met any NetBeans implementations from chip vendors, but maybe there are some. Some of Eclipse IDEs are highly stripped version, some are normal with preinstalled adons. Market is changing really slow at that point, unfortunately.
Fun fact the co pilot series from many many offerers. Are using you to fine tuned the models. It's a Trojan horse type feel to it. What your doing when you use these co pilots guarantee's you won't be needed in that function permanently
Found this video while looking for a fair experience report for AI coding assistants. I doubt it now and it wasn't explicit it is is sponsored video. I always believe, you will always find something good or bad if you are looking. Anyhow, glad to find this channel.
For most of my life, everyone's name, address and phone number was on every street corner in a book everyone looked at. I don't understand why that is a problem now. It's never been a problem for me.
I have used Tabnine a few years back, before the AI Boom that time it was more like Intellisense auto-completion. Now that it's supercharged with AI will check it back once again. As always, good and informative video Alex!
I used co-pilot for a month, and then I left it. It is good for css, Javascript, but the moment I started to use advanced stuff in Go, Ruby, Python, it started to fail. It was not worth my 9.99 a month
This is/was the Apple idea, a IA running locally on the CPU/GPU I don't know if they got it now with m4 or it'll be a failure as they already starting a little late in this AI run.
Of course it sends your code to their servers.. that's how it runs inference on it to extend it ... And I actually mostly doubt Microsoft is using the code in a serious way beyond that, because the ramifications/risks for them if there is an instance of it with competitors software written using their tools or something like that are huge. So more than likely you can trust them, because their incentives are really not to screw this up. That said, I understand not trusting them, they are Microsoft, and I'm by no means sure of anything when it comes to anyone else, much less a big corp.
Being trained on open-source projects, I wonder if they use different weights for different repos/projects...there's a large variance in code quality out there o say the least
You can dissable it, takes one checkbox to stop it. Should not be enabled by default, but it's right there in open on a main page when subscribing to copilot.
chatgpt keeps giving me bad code with methods that arent even in the ios framwork. clearly just scraped from somewhere. It also often gives the least efficient code. and how do I know all this.. EXPERIENCE.
Im laughing in Continue with autocompletions ( any competent small base model ) + llama3 for complete analisis and ideas. Oh and running everything from a LM studio server :)
just copilot? do the same with vanilla vscode, it's way less frequent but it's clearly sharing pretty big payloads with MS too big to be just "telemetry data"
I would really like to see something that uses OCR or an OS level integration. I switch between several different systems and IDE's (embedded devices, servers, game engines, ...) and getting tired of the setup process of these tools.
Alex, what do you think of the errors that have been identified for the M series devices? Do you think Apple will insulate (remedy) its customers from the risk of these M series devices?
these are vulnerabilities, just to be clear, and only effect devices that have been “borrowed” or stolen. I’m not worried about it because I’m the only one using my machine, and if someone rips it out of my hands and forces me to log in, then they can do that anyway
They won't honor your choice of opting out though. Will train and send feedback to home anyway. In fact from windows 10 and up the user can't stop the telemetry service from sending out data to Microsoft servers. Just to illustrate the idea of how the end user doesn't matter to Microsoft.
It depends on where you live and work, but as an employee, it's important to recognize that leaking company secrets can have serious legal and business consequences, both for you personally and for the company
@@AZisk Don't be, I am using Eclipse since day one (it was called "WebSpehre something" back in the day). So I know all the ugly things which are inside and how to trick it. I am lost with IJ... I do not even know how to build my project in IJ ! lol !
What programming language are you using it with? I used Eclise with Java some years ago but then switched to vs code. You can migrate your shortcuts from Eclipse if you want to
@@BelarusianInUk I am using Java with Eclipse, and Python in PyCharm. I link Eclipse because I can run a server inside and debug my code online (frontend & backend). I am managing a team of developers, and they are all using IJ (free version), and they are not able to run a server inside (like tomcat) to debug code online.
Because some of you asked, here is a discount code that will get you 50% off the annual price: ALEXZ . Note that it bypasses the 90-day trial if you use it.
That's awesome! Thanks for the discount code $72 for a year! Wish I saw your code before I signed up for the 90 day trial.. Hmm.. Maybe just use a different email address :)
@@ChinJazzuse the code after the trial expires
@@AZisk Thanks, will try then.
Hi, thx for the code. Does the discount work for the first year only or is enabled as long as I don't cancel my subscription?
Thanks for code i got 50% off on yearly
JetBrains have recently introduced full line completions in their IDEs which use local models, would be interesting to see how this performs against Tabnine (obviously though it won't generate multi-line completions)
Yeah that feature is nice but the completions aren't mind blowing
They have their own AI and I would love to see a comparison.
Same! Free Tabnine was kinda bad, years ago! Jetbrains has been great, so long as my potato from years ago CPU can keep up!
Permissive license doesn’t mean „free to train your AI with“. You usually have to include the license and copyright, which I have yet to see on AI assistants.
Copyright currently (according to precedent set so far) indicates that if you made the code public for people to read, then it's fine to train on it. They do not need your consent to read it, so they don't need your consent to train on it.
Those running inference, however, the people sending prompts and getting results, which may sometimes be substantially similar to copyrighted code, they may be infringing if they save that code even if they now have absolutely no way to realistically know it came from you.
So it is a legal nightmare, but not very likely to have anything to do on the training side, is more on the usage side.
But all of this is just a demonstration about the insanity of information monopolies. Information cannot be owned. It cannot be contained. It exists in the minds of other people once they read it. So having a monopoly that says you can control any kind of information that can exist in substrate including the human mind is a right to some limited control on everyone else. It is ludicrous, and philosophically incoherent.
That's the point of permissive license. you can do whatever you want with the code, no limitation. Maybe you confuse with the copyleft license.
@@ThaitopYT Not really. The MIT license requires including a copy of the terms of the license and a copyright notice with all copies of the software or its substantial portions. The zlib license requires that the source of the software isn't misrepresented. Similar stories with BSD and Apache.
Thank you - I work for a fortune 50 company and we aren't permitted to use Generative AI while on our network for security reasons. This is something I'm going to check out.
Yep, many clients of mine are banning any use of generative AI.
Yep.
@@user-ic6xf Why? I'm a noobie so forgive for this question...If there is a tool that gives you the opportunity to complete a task, wouldn't it be worth it?
@@wilsondv3 security.
@@wilsondv3 Do you understand what the risk they're weighing it against is?
Many clients of mine have put a complete ban on the use of any generative AI.
How do you mean? Purely for security reasons?
They also would have put a ban on electricity 100 years ago.
@@dasistdiewahrheit9585 electricity doesn't collect data that can be used by your competitors.
@@TheXabl0 until a full risk assessment can be detailed. There's numerous cases where data is leaked and sensitivity data and algorithms are being stolen. I wouldn't trust OpenAI as far as I throw a ten ton elephant.
This is the dumbest thing I've ever read in my entire life
Sure point it out, but at the same time this is basic. The GPT model cannot run on your computer. It's not powerful enough and it has never claimed to do so. Thinking that an extension will be able to download and set up a massive AI model in your machine is kind of ridiculous
There was a gpt2 model on hacker news that gave better answers then leading models. Gpt2 could run locally fairly easily.
You can run quantized models locally with something like Ollama and use a free and open source vs_code plugin like Continue
Y'all don't know the latest models then. A lot are much better than copilot/GPT4. And they run locally. Our computers are powerful enough (well depending on your budget and what you buy). The next Snapdragon Elite will destroy everything in AI task for a very reasonable price(45TOPS, vs something like 30TOPS for the much more expensive M4)
you mean gpt4-o pretty sure that model is massive, propietary as f, and closedai
Sounds like you haven't attempted installing one locally yet. Please do so
My company pays for Copilot so I don't really have an incentive to use anything else if it is paid.
If you're using vscode (with telemetry on) and ditching copilot because you thinking Microsoft is stoling your code, you're only deceiving yourself.
And then you push your secret code to Microsoft GitHub 😉
Actually a lot of people in embedded still use Eclipse. There are some great addons that make cross compilation easier. Yes, you can do it with VS Code, but it takes so long to setup and you have to maintain that environment yourself. Meanwhile your microcontroller vendor provides and maintains an Eclipse based IDE that works well enough. Not great, but well enough.
what about NetBeans? Back in the day when I used Eclipse, once I tried NetBeans it was like night and day.
@@AZisk Do microcontroller vendors also provide and maintain a Netbeans IDE?
@@AZisk for embedded almost all main MCU manufactures still use Eclipse as a reference IDE from them. I haven't met any NetBeans implementations from chip vendors, but maybe there are some. Some of Eclipse IDEs are highly stripped version, some are normal with preinstalled adons. Market is changing really slow at that point, unfortunately.
Fun fact the co pilot series from many many offerers.
Are using you to fine tuned the models.
It's a Trojan horse type feel to it.
What your doing when you use these co pilots guarantee's you won't be needed in that function permanently
Found this video while looking for a fair experience report for AI coding assistants. I doubt it now and it wasn't explicit it is is sponsored video. I always believe, you will always find something good or bad if you are looking. Anyhow, glad to find this channel.
For most of my life, everyone's name, address and phone number was on every street corner in a book everyone looked at. I don't understand why that is a problem now. It's never been a problem for me.
Were your API keys in that book too?
All of my ApI keys ended up on the public phonebook. And so did yours! Check your facts before you post! 🤭
Copilot was actually really unhelpful and actually a hinderance to my coding regime
Is there really any developer who thought their code wouldn’t leave their machine?
Every time on push 😂
what about replacing the model with yet another not listed in tabnine? phi3 and llama3 have cool quantized models
I have used Tabnine a few years back, before the AI Boom that time it was more like Intellisense auto-completion. Now that it's supercharged with AI will check it back once again.
As always, good and informative video Alex!
I used co-pilot for a month, and then I left it. It is good for css, Javascript, but the moment I started to use advanced stuff in Go, Ruby, Python, it started to fail. It was not worth my 9.99 a month
This is/was the Apple idea, a IA running locally on the CPU/GPU
I don't know if they got it now with m4 or it'll be a failure as they already starting a little late in this AI run.
Of course it sends your code to their servers.. that's how it runs inference on it to extend it ...
And I actually mostly doubt Microsoft is using the code in a serious way beyond that, because the ramifications/risks for them if there is an instance of it with competitors software written using their tools or something like that are huge.
So more than likely you can trust them, because their incentives are really not to screw this up.
That said, I understand not trusting them, they are Microsoft, and I'm by no means sure of anything when it comes to anyone else, much less a big corp.
Being trained on open-source projects, I wonder if they use different weights for different repos/projects...there's a large variance in code quality out there o say the least
Enterprise copilot license is private, or even you can run it on-prem locally.
And it is REALLY good. Copilot has been getting much better at coding over the last 6 months.
Sounds ironic but okay...
I'm trying tabnine, and I'm already getting pissed off. It doesn't let me choose a tabnine account
You can turn off Copilot using your code to improve its models for everybody else. It's an option.
Pray tell, how exactly?
Sorry but... actually tabnine sucks... it can be safer, but the accuracy answers isn't satisfactory
The Jebrains plugin hangs my ide. Not usable for me, I had to remove the plugin.
I use only tabnine since years and am very happy with it. More secure and more fitting my coding style.
You can dissable it, takes one checkbox to stop it. Should not be enabled by default, but it's right there in open on a main page when subscribing to copilot.
chatgpt keeps giving me bad code with methods that arent even in the ios framwork. clearly just scraped from somewhere. It also often gives the least efficient code. and how do I know all this.. EXPERIENCE.
Calling CLion Neovim is crazy
Thanks for the tips and intro to TabNine. The training of a local model on your own code base sounds useful, but is that only enterprise version?
just say no
also by default openai api does not train, the chatgpt app does though.
We're not allowed to use external coding AI.
Im laughing in Continue with autocompletions ( any competent small base model ) + llama3 for complete analisis and ideas. Oh and running everything from a LM studio server :)
First time I've ever considered native vi as a security assistant.
AI does this with all sort of data property. It includes things you author (contracts, etc.).
0:24 captain obvious…. Does nobody read any terms of service they use?
@@dinoscheidt quite literally….almost no one.
did you opt-out of sharing your code?
Can you even trust them...
No discount code :(
I now have a discount code for you: Use 'ALEXZ' to get 50% off the annual price.
Anyone knows how to add your own project as context to Copilot? So far it only uses current open class and sometimes open tabs.
That head/body animation killed me 🤣🔥
Change your keys frequently
Super cool. I was looking for a good code model service
with the future generations of ai enabled CPUs offering 30+ TOPS it shouldn't be a problem to run this locally.
But it doesn't do this for priva ......... oh, yea it does
just copilot? do the same with vanilla vscode, it's way less frequent but it's clearly sharing pretty big payloads with MS too big to be just "telemetry data"
The last time I tried keep saying that there was no support for apple silicon, I am going to check it out again
Great helping tool - thanks, Alex!
Out of your computer doesn't mean out of your Microsoft tenant (with Copilot).
Great recommendation!!, the video was really fast though.
why does nobody ever talk about supermaven? its the best copilot BY FAR and is free. it’s so smart.
We need to spread this video .
Oracle is going to have so much fun in the coming years.
I would really like to see something that uses OCR or an OS level integration. I switch between several different systems and IDE's (embedded devices, servers, game engines, ...) and getting tired of the setup process of these tools.
Oh really? By developers you mean Yourself? You are the developer? Really? I can tell that my team using more and more of Copilot and other models.
this is because I use Local AI
Alex, what do you think of the errors that have been identified for the M series devices? Do you think Apple will insulate (remedy) its customers from the risk of these M series devices?
these are vulnerabilities, just to be clear, and only effect devices that have been “borrowed” or stolen. I’m not worried about it because I’m the only one using my machine, and if someone rips it out of my hands and forces me to log in, then they can do that anyway
@@AZisk it's my understanding that software running on those devices can launch such attacks. Am I misinformed?
Does it have a chat option like Copilot Chat?
Yes, it does and you can switch models too
You can opt out of copilot training on your code.
They won't honor your choice of opting out though.
Will train and send feedback to home anyway.
In fact from windows 10 and up the user can't stop the telemetry service from sending out data to Microsoft servers. Just to illustrate the idea of how the end user doesn't matter to Microsoft.
1:33 this hurt me bro :(
🤣
can I say that I don’t care if my company secrets get leaked? 😅😂
It depends on where you live and work, but as an employee, it's important to recognize that leaking company secrets can have serious legal and business consequences, both for you personally and for the company
Not legal, not ethical. You better care
I am still using Eclipse ! lol !
i’m sorry
@@AZisk Don't be, I am using Eclipse since day one (it was called "WebSpehre something" back in the day). So I know all the ugly things which are inside and how to trick it. I am lost with IJ... I do not even know how to build my project in IJ ! lol !
What programming language are you using it with? I used Eclise with Java some years ago but then switched to vs code. You can migrate your shortcuts from Eclipse if you want to
@@BelarusianInUk I am using Java with Eclipse, and Python in PyCharm. I link Eclipse because I can run a server inside and debug my code online (frontend & backend). I am managing a team of developers, and they are all using IJ (free version), and they are not able to run a server inside (like tomcat) to debug code online.
I love tabnine, and has been there even before copilot
eclipse🙋♂
i knew there had to be one
Yes, people still use Eclipse. LOL
that dudes face was copilot generated. beware!
cursor ide review plis my bos
Nothing new or surprising with this topic.
1:34 yes
any dev using AI is in trouble.
They still can't steal your brain 🤣🤣
Did I spot you using little snitch? That’s what’s up. You should do videos on really useful software like that.
yep, little snitch is sweet
I love cheeseburger 🍔
I use eclipse everyday ...hahaha😄😄
Djit?!