This is such an exciting development! A giant leap forward in what LLM models can do, now that Wolfram Plugin has entered the equation. This is history in the making. Thanks to the scientists at Wolfram and OpenAI who made this communication of two highly advanced systems possible. And thanks to all the presenters for the wide-ranging examples on what is now possible. Hope this feature is released to the public very soon.
This is truly revolutionary. Seeing the failures, and how experts handle it, was more interesting than successes for me. Also, people should understand this technology is growing exponentially in power and it grows more powerful MUCH faster than Moore's Law. This means we will see this technology become thousands of times more powerful in a matter of a few years and NOT decades as past innovations have progressed. It will surpass every human, in every subject, soon.
Regarding the pH problem that it got wrong on the first pass: If the initial prompt had required it to check its answer (as a mean 1950s nun would have insisted), would it have found its mistake and then re-done the problem?
I am not sure whether to feel impressed or feel disappointed. Chat GPT makes human-like errors: 1) Giving an obvious wrong input data into Wolfram interface. 2) Forgetting to do what's being told, despite promising to do it (ie chemical formula notations). 3) Not checking if the result makes sense or not ( -ve date when the two salaries are equal).
What about statistics? I am trying to test but too busy on the server. Even for simple calculations😢. But I pay for plus since possible. Nothing works for days. I just have access since this week to plugins but the give errors all the time. Maybe because I am based in the EU. It shouldn't as now I am a month behind with learning.
I don't know why but when I give him a probability problem he throws " Set::write : Tag Times in _A (P_not) is Protected", I give some probabilities information and then ask for it to calculate a particular probability based on the info i provided, but always getting this error, so chatgpt just try again and again and it stuck in a loop. unfortunately i had to start an new chat because you can't cancel the plugin either. for example: we have 4 nodes: A,B,C,D which: A have no parents B have no parents C have A as a parent D have A and B as parents we also know: P(A) == 0.1 P(B) == 0.5 P(C|A)==0.7 P(C|not(A))==0.2 P(D|A,B)==0.9 P(D|A,not(B))==0.7 P(D|not(A),B)==0.6 P(D|not(A),not(B))==0.3 calculate for me the next probability: P(not(A),B,not(C),D) wolfram alpha seems to stuck on it and I don't understand why.
Has anyone used the plugin yet?I am in the waiting line but no news. Without the plugin, Chatgpt pretends do do math. It fails at basic stuff like addition.
@@samuelluz9241 I think all the plugins are built on GPT-4. It is just that the system shown in the video is a internal version and using GPT 3.5 and also it is not upgraded to Plus subscription. So, GPT-4 might be even better.
Is anyone gonna do all those IQ tests and standardized exams on GPT-4 with and without plugins? These plug so many holes that talking about LLMs alone makes as much sense as talking about phones without app stores.
I've stopped using math and programming discussion sites almost entirely since I'm using Bing Chat and GPT-4 as a tutor. For one thing, moderators on some of these channels are just plain snotty. Like, how dare you ask a Newbie question that is just a basic misunderstanding. Then, they'll even delete the query as not worth anyone's time. Well, of course I have a basic misunderstanding. I'm a newbie, you jerk. Some of these arrogant moderators act like they were born knowing everything. AI, on the other hand, just answers the question patiently, with simple examples if you ask for that. And you can ask again and again for examples until something clicks. It will also correct itself if you point out an error after trying its result. Usually, you just add a bit more detail you didn't include. I've never had to do more than three tries on most things. Also, you can wait a day or two for answers from a discussion group, and then not get a good answer. As for using Help files, a lot of them are awful since programmers hate writing help files. So they'll scant basic examples, if they even give one. Then, it's on to examples so abstruse you already need to know the program and math to an advanced level to understand the examples. Mathematica is particularly awful in this regard, Again, infinitely patient AI tends to give basic examples, and you can ask for more. And I Never get a snotty reply. Humans suck 🤪
Apologies from chat create wrong impression than make you think you interact with all knowing entity which is not true, it's tool and tools don't apologize, it's a dangerous gimmick
This is perfect! Finally showing how ChatGPT can work with an accurate, vetted source of factual data. I love it!
This is such an exciting development! A giant leap forward in what LLM models can do, now that Wolfram Plugin has entered the equation. This is history in the making. Thanks to the scientists at Wolfram and OpenAI who made this communication of two highly advanced systems possible. And thanks to all the presenters for the wide-ranging examples on what is now possible. Hope this feature is released to the public very soon.
Really cool to listen to and watch a livestream of devs showing off a product I'm gonna spend many many many many hours in. Wonderful!
I need this so bad, you guys have no idea. I am so excited for this plugin!
This is truly revolutionary. Seeing the failures, and how experts handle it, was more interesting than successes for me.
Also, people should understand this technology is growing exponentially in power and it grows more powerful MUCH faster than Moore's Law.
This means we will see this technology become thousands of times more powerful in a matter of a few years and NOT decades as past innovations have progressed.
It will surpass every human, in every subject, soon.
But it will eventually slow down, because even Moore's law can't go infinitely 🤔, but it is still impressive 😎
running test for generated code is great idea so AI could auto correct itself until it finds working code instead me running it and see errors
Nice! I would love to try this. ❤
Any thoughts on how best to combine ChatGPT, Wolfram and something like Remnote to create flashcards?
Regarding the pH problem that it got wrong on the first pass: If the initial prompt had required it to check its answer (as a mean 1950s nun would have insisted), would it have found its mistake and then re-done the problem?
I am not sure whether to feel impressed or feel disappointed. Chat GPT makes human-like errors:
1) Giving an obvious wrong input data into Wolfram interface.
2) Forgetting to do what's being told, despite promising to do it (ie chemical formula notations).
3) Not checking if the result makes sense or not ( -ve date when the two salaries are equal).
It so happens that the wolfram plugin could be functional as an early warning system towards rogue Ai tendencies
The warnings are already here even without the plugin:
ua-cam.com/video/c02s4s34fyw/v-deo.html
What is the question cap with the plugin? 25 questions per 3 hours?
I am totally convinced that now we have a power to build a Jarvis.
Starts at 1:23
I can send images to Chatgpt4 but how do you send images to Chatgpt4 with Wolfram?
This will transform education
What about statistics? I am trying to test but too busy on the server. Even for simple calculations😢. But I pay for plus since possible. Nothing works for days. I just have access since this week to plugins but the give errors all the time. Maybe because I am based in the EU. It shouldn't as now I am a month behind with learning.
Can it do abstract algebra like give you some examples of good written proofs ?
Mathematica has proof algorithms, so probably!
I don't know why but when I give him a probability problem he throws "
Set::write : Tag Times in _A (P_not) is Protected", I give some probabilities information and then ask for it to calculate a particular probability based on the info i provided, but always getting this error, so chatgpt just try again and again and it stuck in a loop. unfortunately i had to start an new chat because you can't cancel the plugin either.
for example:
we have 4 nodes: A,B,C,D which:
A have no parents
B have no parents
C have A as a parent
D have A and B as parents
we also know:
P(A) == 0.1
P(B) == 0.5
P(C|A)==0.7
P(C|not(A))==0.2
P(D|A,B)==0.9
P(D|A,not(B))==0.7
P(D|not(A),B)==0.6
P(D|not(A),not(B))==0.3
calculate for me the next probability:
P(not(A),B,not(C),D)
wolfram alpha seems to stuck on it and I don't understand why.
Has anyone used the plugin yet?I am in the waiting line but no news. Without the plugin, Chatgpt pretends do do math. It fails at basic stuff like addition.
Cool ..... Wolphram is cool.....
does it still use images? When asking what's 2+2 or the population of France it doesn't give any graphs or image examples
keep in mind i'm only 9 minutes into the stream
Before all the ChatGPT math was 100% wrong. This meant, can I really trust the code it produces too?
That's a good question. Probably you should check it to trust it, as you check human answers whenever you ask something
Its so funny when it apologizes.
But also annoying after a while, it wastes a lot of time.
@@vsiegel why do I get the impression it’s just playin’ dumb and it’s pulling my leg sometimes ?
@@anonymous.youtuber Because it was told to play dumb, but maybe not how to perfectly hide it...
How much better does GPT-4 do?
GPT-4 does not have plugin ability yet. They are testing on the less powerful model (GPT-3.5) first.
@@samuelluz9241 I think all the plugins are built on GPT-4. It is just that the system shown in the video is a internal version and using GPT 3.5 and also it is not upgraded to Plus subscription. So, GPT-4 might be even better.
fake links is real problem got those often
still pretty rough
Chatgpt4 with Wolfram correctly did some calculus questions today but I don't know if you can upload pictures to Chatgpt4 with Wolfram running.
Is anyone gonna do all those IQ tests and standardized exams on GPT-4 with and without plugins? These plug so many holes that talking about LLMs alone makes as much sense as talking about phones without app stores.
Jason S (he /they) wtf?
Elon musk wants to create the borg on mars
I've stopped using math and programming discussion sites almost entirely since I'm using Bing Chat and GPT-4 as a tutor.
For one thing, moderators on some of these channels are just plain snotty. Like, how dare you ask a Newbie question that is just a basic misunderstanding. Then, they'll even delete the query as not worth anyone's time. Well, of course I have a basic misunderstanding. I'm a newbie, you jerk. Some of these arrogant moderators act like they were born knowing everything.
AI, on the other hand, just answers the question patiently, with simple examples if you ask for that. And you can ask again and again for examples until something clicks. It will also correct itself if you point out an error after trying its result. Usually, you just add a bit more detail you didn't include. I've never had to do more than three tries on most things.
Also, you can wait a day or two for answers from a discussion group, and then not get a good answer.
As for using Help files, a lot of them are awful since programmers hate writing help files. So they'll scant basic examples, if they even give one. Then, it's on to examples so abstruse you already need to know the program and math to an advanced level to understand the examples. Mathematica is particularly awful in this regard, Again, infinitely patient AI tends to give basic examples, and you can ask for more. And I Never get a snotty reply. Humans suck 🤪
I don't know, is it me? This felt more like "(not) using ChatGPT Plugin"...
Ну туповат конечно, но надо признать, что возможности поражают
Apologies from chat create wrong impression than make you think you interact with all knowing entity which is not true, it's tool and tools don't apologize, it's a dangerous gimmick
I think you are a dangerous gimmick
wtf