/sci/ - Science & Math

[Anonymous No.15284422]

GPT-4 can finally sovle physics problems correctly

[Anonymous No.15284427]

this is GPT-3

wrong answer

[Anonymous No.15284434]

>>15284422
This is very interesting. I would try giving it a more involved problem and see how it does. A good one is some problem that can be solved simply using complex mapping but is very difficult to approach through biot savart. For instance, finding the magnetic field created by a sheet current [math] K = K_z [\math] confined to the strip [math]-L<x <L [\math]. I wonder if it can solve the problem, or even recognize that the limiting form of the field far away is [math] \vec{B} \approx \frac{4 pi K L}{s} \hat{\phi}[\math]

[Anonymous No.15284477]

>>15284422
stop flooding the board with this tripe, post in the ai general or don't post at all

[Anonymous No.15284586]

>>15284477
Too stupid to be an Machine Learning Engineer? Is that why you take it out on others and use words like "tripe"?

[Anonymous No.15284594]

>>15284434
>>15284477

[Anonymous No.15284625]

>>15284594
stop posting soijaks, they trigger me because they hit way too close to home

[Anonymous No.15284627]

>>15284477
Make me bitch

[Anonymous No.15284646]

>>15284627
you're already are bitching and moaning and you're doing it in response to my previous post.

[Anonymous No.15284661]

>>15284477
>stop posting about this huge paradigm shift that is the most important historic event of our lives
>only one post about it is permitted, make way for 10 IQ posts, 3 Monties, and 5 about how electric energy just doesn't work, ok?
You are free to make posts about topics that interest you, and to ignore threads you don't like. Go and read the rules to see if they say something about backseat moderating.

[Anonymous No.15284716]

>>15284594
It's interesting to see whether ai can make the connection between asymptotics in mathematics and physical concepts. At least your basedjack bumps the thread

[Anonymous No.15284738]

>>15284477
offtopic, but what is it about AI tech that causes people to seethe so much? It's bizarre. You don't see this in any other kind of tech. So many people just get incoherently and unreasonably upset whenever it's mentioned or whenever it accomplishes something new.

[Anonymous No.15284748]

>>15284738
The anti-AI nuts aren't against AI, they're against AI nuts
When something gets shilled and overhyped by retards to this degree people get sick of the faggotry and swing too far in the other direction

[Anonymous No.15284749]

>>15284422
Doesn't gpt4 read in images? Try posting some standard exam questions from physics which refer to images. Like a circuit.

[Anonymous No.15284758]

[Anonymous No.15284808]

woah

[Anonymous No.15284828]

>>15284422
It cannot "solve" problems. It looks up question in stack exchange and copies the answer. Remind me when it solves an unsolved problem.

[Anonymous No.15284842]

>>15284422
The development of AI has shown great potential in solving various problems across a wide range of fields, including physics. However, while AI systems have made impressive strides in natural language processing, image recognition, and other areas, solving physics problems correctly requires a deeper understanding of the underlying physical principles and mathematical equations involved.

It is important to note that physics problems involve more than just language processing. They require a strong understanding of mathematical concepts such as calculus, differential equations, and linear algebra, as well as an understanding of the fundamental physical principles that govern the behavior of the universe. These principles include classical mechanics, thermodynamics, electromagnetism, and quantum mechanics, among others.

While AI has shown great promise in some areas of physics, such as identifying patterns in large data sets or predicting the behavior of complex systems, there are still limitations to what it can do. AI systems can struggle with problems that require a nuanced understanding of physical principles or that involve complex mathematical equations. In many cases, these problems require human expertise to identify the best approach and to interpret the results.

In addition, while AI systems can be trained to recognize patterns and make predictions based on large data sets, they may not always provide the most accurate or precise answers. This is because AI systems are only as good as the data they are trained on, and if there are errors or biases in the data, those errors or biases can be magnified in the AI's output.

In summary, while it is possible that future versions of GPT or other AI systems will be able to make significant contributions to solving physics problems, it is important to approach such claims with a critical and realistic mindset. AI is still a developing technology, and there are limitations to what it can do.

[Anonymous No.15284850]

>>15284842
lel, nice one chatgpt

[Anonymous No.15284851]

>>15284422
I actually think GPT-3 should probably solve this most of the time. This is a trivial problem. Magnetic field strength a distance from a straight line current. Can't get easier than that.

[Anonymous No.15284905]

>>15284422
>Not being embarrassed that you "AI" takes a page of text and several years of upgrades to solve a one-line problem that requires zero algebraic restructuring or calculus.
This shit is a SURVEY-COURSE-LEVEL problem, and the AI clearly still struggles with it. It's fucking pathetic.

[Anonymous No.15284913]

>>15284850
>Fucking faggot can't even keep his leftard politics out of his physics course.
No wonder the problem is so simplistic.

[Anonymous No.15285215]

Can it solve the hard problem of consciousness though?

[Anonymous No.15285949]

>>15284828
This is basically what it's doing. If it only has one match for a prompt in its training data it will send you the answer copied verbatim from the source. If it has multiple matches in its training data it will send you some combination of them glued together

[Anonymous No.15285960]

>>15285215
No. Or at least it's very unlikely. It can only give results based on it's training data. It does have a memory of sorts to keep track of context but it can't think. It can do fuzzy marching of your prompts to its data though which can potentially give you results you wouldn't have thought of asking, but it's still limited by its training data.

[Anonymous No.15285966]

>>15284913
doesn't it make sense?
scholars devote their time to being told how to think.

[Anonymous No.15286501]

>>15284422
Language models do not solve problems. They converge to an output vector of text which matches the input prompt.

[Anonymous No.15286887]

>>15285949
>>15285960
>>15286501
Turns out that's what human brains have doing all along, just with more neurons and asynchronous execution

[Anonymous No.15287056]

>>15285949
>>15285960
>>15286501
bros......

[Anonymous No.15288840]

>>15286887
Based and truthpilled.