ChatGPT + Noteable (Jupyter) = Mind-blowing! 

Thanks so much!

It's not usable.

I'm having great fun with it; thanks for releasing it so quickly!

@@LeviDeHaan What is not usable? The ChatGPT interface? Noteable? Something else?

Done some pretty good analysis already…have to check with very large dataset and I am bit concerned about the security of the data before I test them

@@History_Mystery_Crime Noteable's public/free systems are on shared computers, and they're small Amazon ECS instances. So you can't put too much data on them, and there's a chance that someone else could see/use your data. (I'm guessing it's a small chance, but I'm not your lawyer or head of security!) Their enterprise pricing is meant to avoid both of those problems. (This is from a meeting I had earlier this week with a senior person there.)

I know, right? Glad it was useful; I'm definitely going to be doing more with this as I explore it further.

Glad you enjoyed it - more to come soon!

Upvote for using "phenomenal" instead of the apish bogan lexicon which uses phrases such as "mind-blowing". Content makers take heed please. Stop using phrases that include "mind blowing", "blow your mind", "mind blown". It's tiresome and my first instinct is to pass on content with that type title. It "literally" blows my mind (in a bad way). You know I'm not being literal. Because I'm still typing, I think.

@@Tucanae515 Fair enough! I'm usually more mild-mannered in my speech and writing, but I spent a few days just completely awestruck by what I was seeing. Apologies that your gray matter is still fully inside of your head after watching the video!

@@ReuvenLerner Awestruck is good! I completely agree though, the pace of AI development is amazing. I had the 8088 Intel based PC when it first came on the market in the '80s. I thought Star Office was a whole different order of magnitude back then. I can only shake my head in disbelief re AI progress. Happy times. Can't wait for AGI (if it happens).

​@@Tucanae515 I'm totally with you on this; when I took a course on AI at MIT in 1990 from the head of the AI lab, the semester was basically telling us about their techniques for solving problems, and how nothing really worked. This was in the middle of the "AI Winter," as it was known. Fast forward 30 years, and all of the techniques they taught us have been blown out of the water by machine learning, and everything they wanted to do is now working beyond what they could have dreamed. A lot of that has to do with not just different techniques and approaches, but also the fact that computers can hold FAR more data in their memory. Which allows them to do the pattern matching and prediction. Fun times!

I'm delighted that you enjoyed -- looking forward to hearing what else can be done with these amazing tools.

@@ReuvenLerner Sure, will let you know!

Glad you enjoyed!

Amazing! Thanks so much for suggesting this.

Thanks! And yes, it definitely gets complex quickly; my next video shows what I had to go through in order to get ChatGPT to solve my Bamboo Weekly problems from last week. It wasn't smooth sailing, at all. And checking the code? Yeah, I show that, too - it's far, *far* from accurate all of the time.

Glad you enjoyed - more coming soon!

I've been playing with the API a bit, and will almost certainly do a video on that topic in the coming days. I'm actually thinking of doing a short online class in about two weeks on the subject, as well. I know, it's pretty amazing, right?

Thanks so much! I should have something new out very soon... I'm having so much fun playing with this!

@@ReuvenLerner You were interesting, but you were not great. Real men take constructive criticism and improve.

Aha! Thanks for the great tip.

Thanks for the warm words - you can totally learn programming with Python, and I hope that I can help you achieve that. Meanwhile... welcome!

My pleasure!

Welcome! I hope that I can show you lots of things you didn't know - the fun of this industry is that we're all always learning.

Yeah, this is pretty amazing stuff.

Glad you enjoyed!

Glad you enjoyed!

I know, I have been playing with this for the last few days, and I'm amazed by the possibilities. It's clear that just as accountants don't have to do simple calculations any more, and can concentrate on higher-level thinking, so too will data analysts be able to think at a higher level. But...finding a bug in your prompt isn't always easy. Finding a bug in the code that is generated is even harder. What will this do to teaching? To employment? To how we think about data? To... well, to enormous numbers of other things. For now, I'm just stunned by how much work that until two weeks ago could only be done by a trained coder is now doable via a handful of sentences. This is definitely revolutionary technology, but quite how it integrates into our existing paradigms, I'm not sure. And how I'll integrate it into my teaching... I'm definitely not sure, but am open to ideas!

Please explain the terrifying part. If you can.

​@@chrismachabee3128 Because of how it's going to change the nature of working with data. It's going to reshape the economy and no one is sure what the overall effect will be. An astronomical increase in productivity? Mass layoffs? Both? Etc....

​@@ReuvenLerner Actually, the issue you raised isn't an issue. GPT can already hunt bugs A LOT faster than humans. Hunting bugs have become a breeze with GPT for me. I just ask it to generate the code I want, then I review it, and keep asking it questions until I understand everything. Then, if there is a bug, I just post the error message, and in a few minutes of trial and error with me and GPT, we have caught all the bugs. I honestly don't see a need for consultant anymore, everyone can code, and I meant professionally, now, no matter their experience. This is indeed fascinating and terrifying.

​@@spacenerd113 I think that someone smart and determined, with time to do this sort of work, will indeed be able to much more than before without the help of consultants. But there are lots of people out there without time, and/or who can't even put together a good Google search. So I don't think all coders and consultants will go away, but there will be *far* less need for them, or they'll need to work at a higher level. But yeah, fascinating and terrifying. A great phrase.

I'm planning to do more videos about this, once I have a chance to breathe and think about it more, for sure! Thanks for the encouragement

Amazing, right? Glad to hear it worked!

Thanks so much!

We *all* have questions! I'm trying to understand these things as quickly as I can, and I'll keep coming out with videos/tweets/etc. as frequently as possible... but hey, whatever questions you have, please do share, and I'll do what I can to help.

Pandas can definitely connect to databases. So it's more a matter of figuring out how to describe, in a ChatGPT prompt, how to connect to the database server. But yeah, I'm sure you can do that!

Glad you enjoyed - welcome!

I think that I've seen/heard about some sort of interaction among chatbots. Can't remember where. By definition, these will only create generated text based on what they've been trained on. But that doesn't mean it'll always be of low quality or that it'll fail to amuse/entertain/inform us!

Your coffee machine is coming soon, perhaps via the zapier plugin! Don't taunt ChatGPT, because it's coming for you...

@@ReuvenLerner No doubt. Both my kids are fortunate to go to excellent private schools here in Oz. I hope they introduce hairdressing soon, as that may be the only job that can't be easily AI'd... have I spoken too soon? 🙂

@@deldridg 🤣

Thanks! I'm preparing another video to go along with this week's Bamboo Weekly, and will have a lot to show + tell about failure, etc. I've heard other people say that GPT-4 is giving them worse results than GPT-3.5 on certain things. Very interesting. I'm not sure what it means about the models, though!

No doubt, this is all in its very early stages, and there are all sorts of issues. I just spoke with someone from Noteable yesterday, and it seems that while the free version runs on small AWS instances (with limited RAM), enterprise customers can pay for heftier machines. I'm guessing that this might be connected to the problems you saw. Or it's just an annoying bug - hard to know at this point. I'm still trying to figure out what you can do, and how you specify things. And while this is truly amazing, I think that we're still far from non-technical people writing specs in English and seeing it run without problems. But it's still some of the most amazing stuff I've seen in a while.

One thing you can try is to ask ChatGPT to break up code into multiple cells. ChatGPT itself (not the plugin) can only work with so much text. When it errors due to size, the plugin doesn't even see it. As OpenAI rolls out larger context windows where this will be less of a problem.

@@KyleKelleyR Funny you say this - I have been doing just the opposite, shoving tons of text into ChatGPT, in order to stay under the 25-queries-per-3-hours limit. I do wonder if that has influenced the success of my queries, for good or for bad.

@@ReuvenLerner I don't know the internals at OpenAI but I definitely notice I can pass a lot of text to ChatGPT but sometimes while trying to make the plugin call ChatGPT will just stop short after a request gets long enough (especially if there were other calls). It's just a model though so I appreciate all the stuff we can throw at it right now to see what works. Experimenting and getting messy is a-ok with ChatGPT and ephemeral compute by our side. :D

Glad you enjoyed it!

Wow. That's astonishing.

I think that it'll be *more* automated than it has been. But it's not like programming hasn't been automated already, to some degree. Compilers rewrite code all of the time to optimize it. No one writes assembly language any more, because compilers are better at it than people. At the same time, people are still needed to specify what they want, to make adjustments based on real-world feedback, and to check that the results are accurate. So we'll be concentrating on the high-level things, rather than the low-level things. Which doesn't seem like a bad place to be, except for the junior staffers who are doing low-level development work. Those jobs might be combined or eliminated.

Thanks - to be honest, I didn't know at that point what the maximum size of data that they could handle at Noteable. I now have a better idea, and even though it doesn't translate directly to Pandas size, some of those data sets would have been just fine. Bottom line: If you want 2GB of data, then you'll probably need the enterprise version of Noteable, which they can sell to individuals, but they're really gearing for big businesses. (That said, I might end up buying it anyway.) You can use their multitenant version for free, and that has a few GB of RAM you can use.

​@@ReuvenLerner Hi Reuven, thank you for the reply. Actually, I thought what you were doing as a demonstration was great. Now what stuck in my mind was you dropped the 2GB set of data. I mean I just wanted to see you work with it like the other datasets you showed. So, I'm trying to think how you can work with that data without actually eating the whole cookie. In other words, is there an intermediary place where you could download, get a look at the data, and perhaps clean it, and then choose a section you'd like to work with? The best would be if you were getting the data from a place that offered a programmatic solution, as in, offering the 2GB data in chunks, because perhaps everyone doesn't have to work with the entire data set, and if there was a chunk of data that had the information that was downloadable in a reasonable amount something could be accomplished. So many smart people. I think there has to be a buffer area, excel where you could land the data, segment to what you need to use, and allow the rest of the data to fade away. When I crop in Word I can take an entire two different web pages and crop down to an area that I want. I click on the outside and those two web pages disappear and the crop portion remains. So, that's what I mean if you could get a look at the data to see, you then would know what you wanted. Another thing is Google Sheets or Docs, those apps handle more data than suite apps, I don't know if they could 2GB. But here is an idea. I think something like Dropbox is where you could download data, and I am positive you could get a look at would you need and just get that data. Want to hear a funny? I used GPT-4. Went and got a bunch of plugins the other day. It works it worked for me, and it works for the guy I see using the plugins in their video. I go to use it today. It has no capability, It only has information up to 2021. OK, now I am looking at World News and Web pilot on my GPT-4 screen, and for some reason, GPT not making the connection. It is supposed to connect, but after a few moments, today it didn't, and then the other night, with GPT Plus I got a too many people online notice. Wait, I'm not supposed to get that, and it’s the 2nd time. Be aware there appear to be rolling changes to capabilities. They just came up with 25 inputs only, in 3 hours. They must have gotten the air conditioning bill. OK, we're really only a few weeks into this. I'm studying Python now in Jupyter and so I was really interested in what you you were doing because I know it's groundbreaking and I want to know how to work with the program that way also.

​@@chrismachabee3128 Pandas is totally able to handle 2GB of data. However, I wasn't sure whether the Noteable environment could, or if they would restrict how much I could import. Turns out, I was right to worry, because they're using small AWS instances and I might have overloaded it. You're undoubtedly right that they estimated how many people could use it at once, and now they're overwhelmed. That said, they're sitting on a lot of cash, and want to maintain their lead, so they'll keep firing up new virtual machines as often as they can! Lots more coming. This is tons of fun. Thanks for your comments.

@@ReuvenLerner Hi Reuven, OK, however to bring a little light you know, the Notable folks may have a restraint on, because AWS has capability modify space on the run, if Ntable will pay for the escalation of data. Anyway. Sound like you got the handle on it. Take it easy.

I don't know much about Graphviz (last used it in my dissertation a few years ago), but I'll see what I can do on the 3.5 front. But right now I'm charging (no pun intended) full steam ahead with Noteable integration, because it's just so much fun... I'll have more for you as well, and I've got tons of other Python/Pandas content, too!

I agree; coders won't be going away, but we will be using AI systems to write our code. The analogy I like to use is that my accountant still has a job, even though he uses Excel; the creation of spreadsheets didn't make his work obsolete. At the same time, his office needs fewer workers to do the same amount as before. Right now, anyone who blindly depends on ChatGPT to do their work will likely end up fired or an amusing news story. (Or both!)

Great idea!

The data size didn't matter from ChatGPT's perspective, because it was loaded from/into Noteable. It wasn't sent to/from ChatGPT at all, so we were fine there. I've found that ChatGPT tells me (roughly) what the limits are on its chat screen. I think that the API docs say even more about it. I'm pretty sure there isn't any rollover over tokens if you don't use them.

You have to opt-into it in your settings because plugins are still technically beta.

You need to turn plugins on, by choosing GPT-4 and then clicking the checkbox where it says "plugins." Then, after doing that, you'll be able to choose ChatGPT plugins (i.e., go to the plugin store) just under the GPT-3.5/GPT-4 buttons. That said, these things are changing rapidly, and you might need to look around.

Thanks for the translation for the younger, hipper viewers!

The code isn't the issue (in my opinion) nearly as much as the model's training data. Which might eventually be open source, or there might be open source alternatives, but ... I dunno, it also requires a ton of computational power to run. Regarding your login issues, someone else I know had that problem earlier today because of his custom domain. When he switched to a Gmail account, it worked. Maybe that's your issue? I dunno!

Fair point -- if I had *any* idea that this video would have been so popular, you can be sure that I would have been a bit more deliberate, that's for sure. I'm planning some more videos, and am indeed being more careful about what and how I show things, rather than just sharing my excitement with the world.

This video pops up in the recommendation, after watching I also got mind blowing. Actually I’d rather like this free style along the way of exploration, this gives me a sense as if I am with the author discovering things together. The content is also inspiring and I’d probably try this myself. I think this is the magic and value of sharing, which I believe is the most important reason for me coming to RU-vid.

​@@DSoong Thanks; a few days after posting this, I posted something more organized and methodical - which still managed to surprise and amaze me. That's me, and that's the nature of this kind of technology where it's going to give you different results each time. Thanks for taking the time to comment!

Personally, I feel like I get more value from seeing someone troubleshoot through a process rather than just giving me the final solution.

Once you have ChatGPT Plus, a new chat will (at the top) ask if you want the GPT-3.5 model or GPT-4. Choose GPT-4. Then you can choose which variation on GPT-4 you want. One of those options lets you install plugins. Once you've selected that plugins should be active, you can then install one or more plugins from the plugin store. Any number of plugins can be installed, but only three can be active at any given time. Activate the Noteable plugin, and you should (I hope!) be good to go.

Yeah, I didn't do it enough this time around. Whoops!

I haven't tried it yet, but there's a package on PyPI that works in the other direction -- you use Pandas in Jupyter, and it uses the OpenAI API to talk to ChatGPT. It's at github.com/gventuri/pandas-ai, and I'm hoping to try it in the near future.

I have a few ideas, but they're just guesses: (1) I've seen that Noteable's free plan has been a bit slower lately, especially at peak usage times. Maybe that's contributing to the timeout? (2) If you are loading a large data set from somewhere, that would probably compound the issue you're having. Maybe try pre-loading the data set into the Noteable notebook/project. (3) What if you uninstall and re-install the Noteable plugin? I hope that one or more of these will help...

@@ReuvenLerner I did sign up for the pro version of Noteable, and it did seem to help some but not entirely. I did find that it usually works to tell Chat GPT about the error and ask to continue where it left off and it usually works. I am using Starlink for internet, I am wondering if that has some impact

In theory, I have to assume that if earthquakes are predictable in some way, that AI could indeed predict them. But I don't know what factors/inputs are needed to make such predictions, and I have no idea if earth scientists know yet, either.

Sorry, but I really know nothing about Java Spring. But a notebook can execute any Python program you want, which means that you can almost certainly download files of any type and size.

@@ReuvenLerner I am looking for a way, where, chatgpt tool, can store generated files into a folder, either in shared locaitons or, urls like jupyter notebook or github or any shared folder or in my pc. some how, looking for a way, chatgpt can store the generated files

@@shivachaturvedhi8840 Maybe there's a plugin for it to store files; I don't know. Noteable notebooks can definitely store files, so there might be a way for you to store them in that way, either creating or downloading them into the "local" directory.

Good point! I'm so used to making the text bigger when I give courses, but less so when doing videos for RU-vid. I'll try to pay closer attention to this in the future; thanks for the feedback!

@@ReuvenLerner Yes, I teach via Zoom a lot, so it's something I do as well. Thanks!

Noteable definitely lets you have local data files. So you could ask ChatGPT to connect to an existing notebook and use the data from there. In this particular case, with small data retrieved from FRED, I'd say that we're OK. But there's a reason why I didn't download the really big data, in part because I didn't even know at the time how much Noteable could handle in its storage.

According to their site at noteable.io/, the answer is "yes."

I don't think we're even near the point where AI can do the work itself. It's a lever, allowing you to do more and harder work than you could do yourself, not a truly intelligent or autonomous system.

(1) Under GPT-4, make sure that there's a blue checkmark next to the plugins option. (2) When that's done, look under the GPT-4 button for a menu that lets you enter the plugin store. (3) Choose one or more plugins from the store, and it should (I hope!) work.

@@ReuvenLerner Something I only recently found out -- some people have to enable it under settings in the lower left.

It was missing a few days ago, and when I contacted them on Twitter, they said that they were waiting for re-approval from OpenAI, because (I think) they had updated their software. So if it's not available, wait a few hours, and it should (if all goes well) be back.

@@ReuvenLerner Unfortunately if we make certain kinds of updates, OpenAI needs to review them, so we'll be out of the store for a few hours. This shouldn't happen too often.

Hmmm. You need to be (a) using GPT-4, as part of the paid ChatGPT+ subscription, (b) have the plugins activated, and (c) have the Noteable plugin turned on. Also make sure to have a Noteable account that is connected to your ChatGPT account. If all of that is true and you're still getting this error, I'm not sure what to suggest, sorry!

@@ReuvenLerner thank you for your suggestion, yes I am a plus user and had notable turned on, I always find difficult to trigger noteable even I specify ' do with noteable' I do have a noteable account, do I have to have my workplace open when I use gpt ? how does noteable know which workspace to interact with or which project to interact with

​@@zac1427 I know that I had to somehow connect ChatGPT with Noteable, and that I had to tell it which noteable workspace to use. Try telling it that, saying, "Use the noteable workspace ABCD." I think (hope!) that'll then connect it. Also, I think that I was able to talk about using Jupyter, rather than Noteable, and it knew what to do. Hmmm...

ChatGPT has been around for about six months, the GPT4 model has been around even less, and plugin access is super new. Maybe my point wasn't clear, but I was trying to say that for someone to get excited about ChatGPT only now is late to the party - but to get excited about GPT4 and the plugins? Well, that's a different story!

I don't really know much about creating dashboards, I'm afraid. It's one of those Jupyter things that I really need to learn one of these days...

Well, Python can retrieve info from other sites via scraping or APIs. Noteable notebooks are running Python, so they can do this, too! I know, it's pretty amazing.

Agreed, the CPT4 model doesn't seem as smart as it did a few weeks ago. I would chalk this up to me getting used to it or having high expectations, but enough people are saying the same thing that I'm starting to wonder what is happening.

@@ReuvenLerner yeah same. It has been kind of crazy lately, where things that worked just a few weeks ago and then voila, now it just refuses or just doesn't seem to understand. I have tried a bunch of different prompts too and in the end, it just ends up a waste of time as I could probably just have done it by myself in half the time...

@@IronMeDen1 The "Hard Fork" podcast just said that the OpenAI folks warned us that ChatGPT-4 seems less and less amazing over time, as you use it. Maybe that's what we are all seeing? I don't think so, but there might be an element of that.

@@ReuvenLerner interesting. I mean, it is hard to bring concrete evidence to the debate since I'm too lazy to actually look back at previous conversations... Plus I have deleted them haha. But it wasn't long ago that I was giving instructions to GPT, stating what I wanted and how and it gave me something working right away. Now, I can give short concise instructions and it will miss the one and only condition I asked to keep in mind. Hopefully it's just some temporary hiccup from some tweaking or something, but something does feel different...

Noteable uses a small instance of AWS, so you are limited by the memory and disk size there. If you want something bigger, you'll need a corporate account. (Or so they tell me - I haven't tried it myself.) But if your data does fit into such a system, then you should be fine. I know that you can load files onto your Noteable account, and I have to assume (but haven't tried) that if you pre-load a file, then you can get ChatGPT to see the file and use it. I'm not sure, though.

Sorry it wasn't readable; try the follow-on video I did, which I think (hope) is larger and clearer: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-2WUZ0b-hUDU.html

No, it's definitely there. Check that you're looking through all of the plugins in the plugin list/store, rather than just the newest ones.

@@ReuvenLerner I wonder if it shows up for you since you already installed it. I just went through the complete list again, and I don't see it.

@@paulalvarez1758 Hmm, I dunno. Maybe it's a temporary glitch? Maybe my video overwhelmed their system. :-) It was there earlier today when I went through the plugin store, and Notable just announced earlier today that their plugin is one of the most popular ones, so... I'm not sure.

I just searched all 16 pages of gpt-4 plugins. I cannot find Noteable as well.

@@doonamkim7593 I just went back to one of my older chats, where I had used Noteable. It's now showing up as "unverified." My guess is that they found a problematic bug, and that the plugin was pulled from the plugin store for now. I'm also guessing that it'll be back in the coming hours, but that's just a guess.

You're right that it's a code interpreter plugin. You're right that it requires more input than code, and it requires some programming skills. But I've been playing with it a lot over the last few days, and I wouldn't call it primitive - for example, if you ask it to turn a plain-text file into a data frame, it'll try to write a custom function that interprets the contents of the file. If you ask it to explain what it sees in a graph, it'll do a not-too-bad job. And yet, saying that you don't need coding skills to describe/specify what you want, or to check that it's doing the right thing, is (I think) overblown. I think that it's still a tool that helps people who know the subject work faster and better than they could on their own, but it doesn't allow non-programmers to suddenly remove the need for programmers in the world. To repeat a metaphor that I've used elsewhere, my accountant is still in business, even after the invention of Excel; he can just concentrate on higher-level thinking.

@@ReuvenLerner You make a great case. I totally agree with you on all of your points, I could’ve expanded a bit more in my comment. And you’re right, it won’t remove the need for programmers, and yes definitely will help people learn faster. I’ll look more into noteable to do some more complex tasks you’ve described they sound really exciting!

@@ReuvenLerner That second part is right on. The sky is falling crowd will say we're all screwed but I agree with you, you still need some skill to use this stuff. I love your metaphor by the way.

Yup, you can scrape the data, then analyze it in Pandas. That's pretty standard; the impressive thing here is that I described what I wanted to do, and ChatGPT turned it into Pandas code, and then executed it.

nice way to learn, because i can see the code so i can like copy and reuse that code in another moment @@ReuvenLerner

Thanks for the kind words! As for the time it takes to learn something, different topics take different amounts of time to learn. I try to learn things in depth, then simplify them to the point where I can explain them in a few minutes, hours, or days, depending on the depth that people need.

In theory, the answer is "yes," you could ask ChatGPT to scrape sites, retrieve the data, and analyze it. You might need to use plugins other than Noteable; Zapier comes to mind. But again, in theory, it's possible. Why am I stressing the "in theory" part? Because specifying things with the precision required by a computer is a skill that takes practice, and that even many experienced developers, using a specialized programming language, get wrong. So if you don't have that experience, and you're using a human, ambiguity-filled language like English, it'll be harder. Not impossible, and things will get easier. But we're not yet at the point where non-programmers can perform analysis projects without coding knowledge. However, once the data has been loaded, then I can more easily imagine using English and ChatGPT to analyze stuff that's already there.

Maybe, I dunno if the Noteable people really care about having access to your data. But there's no doubt that they own the AWS instances, and can do whatever they want with the code and data there. If you're doing commercial stuff that needs to remain secret, then you shouldn't be putting it in ChatGPT *or* in any other system that doesn't have a good agreement and track record on keeping your data private and secret. And even then, mistakes happen...

I just had an hour-long meeting with someone from Noteable yesterday, and I came away quite impressed with the product. You can think of it as a corporate superset of regular ol' Jupter, and their files are compatible with notebooks elsewhere. I haven't used Noteable for anything other than playing around here, but I'm likely to try it for some of my training, in the wake of my conversation. (And it looks like their free plan will be sufficient for my needs.) The problem you're describing might be in Noteable, but I think it's more likely somewhere in the bridge between ChatGPT and Noteable. I also had some experiences where it went into a loop - failing to do what I wanted, and then trying again and again and again, until I stopped it. I have to assume that this will improve over time, as the various pieces of software (all of which are being developed and deployed at breakneck speed) stabilize. I wouldn't be surprised if ChatGPT will eventually be able to make plain-vanilla Jupyter notebooks. But right now, Noteable is the only game in town, and I've been satisfied with it so far, given the very early stage things are in.

@@ReuvenLerner I hope it will work better soon. When I discovered the Noteable plugin, I thought I found the holy grail :D Thanks for your extended reply!

@@rakly3473 Please swing by our forums we'd be happy to help resolve those issues: community.noteable.io/c/issues-and-bugs/

@@noteable_io Thanks for replying, I had trouble finding my post back. I wanted to update that now everything is working fine!

Sorry -- I was pretty excited about what I saw, and was simultaneously playing with new thumbnails for the RU-vid channel... I'm going to keep trying to improve...

ChatGPT+, which gives you access to the GPT-4 model and plugins, does indeed cost $20/month. Yes, it's true that you, I, and other developers can do all of this stuff ourselves. Where the line will be drawn between what we continue to do ourselves, and what we have models do for us... that's anyone's guess at this point. Right now, a non-developer doing this sort of thing would be foolish or dangerous. But in the long term, using a model to do some of our coding will likely be the equivalent of my accountant using Excel for his calculations. I would still prefer to write my own code. But the fact that it can do some of my research for me, and then download and visualize models in seconds without me lifting a finger... that's not nothing!

$20 a month is a bargain. The capabilities of these models are unknown, and thus should be explored and exploited.

@@cyberft I’m happy with the free 3.5 version as it seems like they are pretty similar so far.

@@Lnd2345 got-3.5 has the intelligence of a 13 year old, whereas gpt4 has the intelligence of the top 10% of medical school/law school student.

@@Lnd2345 lol, no gpt-4 is miles ahead of 3.5. I guess it depends on the task. If you only need it to write some basic emails, then they are the same basically.

I know, right?

I'm pretty sure that ChatGPT has only been around since the end of 2022. GPT models have been around longer than that, of course.

Reply

It has certainly occupied a lot of my attention over the last few days! LLMs are a revolutionary technology. They aren't perfect. But they are an incredible advance. I'm a little apprehensive, and very curious to see where this all goes. But it's certainly consuming a growing amount of my thinking time, considering what this all means, and how we'll use it.

Yeah, it's a real bummer that they had to shut down. I'm not sure who else is doing such things -- and to be honest, I'm going to try to stick with open-source stuff that is less likely to fall victim to startup pressures. If/when I see something interesting, I'll let you know!

Sorry it didn't work for you! I've often been told that people like to see when I make mistakes, rather than cover them up and pretend everything is perfect. I'm planning to do some more videos on this subject in the coming days, though, and hope that they'll be more in line with what you expect and want.

Meaning what, a dark screen for Jupyter? I honestly didn't think people cared, but you're the second person to comment on this, so maybe I should? I dunno, I've had a white-screen Jupyter background for as long as I can imagine. Maybe it's yet another sign that I'm old...

Yeah, this was a very quick initial video that I never expected to blow up, reflecting excitement and discovery from a day or two with it. My next video on the subject is far far more organized, and I think you'll find it more useful and enlightening.

Tools have been replacing jobs for centuries. And AI has been replacing jobs for decades. I think that the big deal here is that (a) it's happening to white-collar knowledge workers, who were sure that their jobs were untouchable by technology, and (b) it's happening so fast. I share your fear that society isn't completely ready for this, but I'm more worried about it taking way lower-level jobs, not all jobs, and making sure that we provide people with the support they need. But stopping AI? That's not going to happen. This technology is here, and we need to figure out how to keep it safe and regulate it, not do away with it.

Yeah, the data that I had been playing with for a few days suddenly wasn't behaving... so I used data which showed the idea, but not very effectively. I'll have something new in the coming day or two that is much more carefully selected and curated, and which will hopefully make more sense.

Maybe I'm just old, but I like the white background. However, I seem to be in a minority...

Glad you enjoyed!