"I have no idea at what stage is Google offer. Public version
is probably weaker than versions available to internal
developers and paying customers. Still, progress seem to
be rapid and there is something like a year from availability
of ChatGPT 3.5."
Well., I just a key and tried it out. I asked it to integrate sin(x) and it
returned back -cos(x)*sin(x)+c !
curl -H 'Content-Type: application/json' -d '{ "prompt": { "text":
"integrate(sin(x),x)"} }'
"https://generativelanguage.googleapis.com/v1beta3/models/text-bison-001:generateText?key=my_key_here";
{
"candidates": [
{
"output": "-cos(x)sin(x) + c",
"safetyRatings": [
{
"category": "HARM_CATEGORY_DEROGATORY",
"probability": "NEGLIGIBLE"
},
etc....
It also did not know how to integrate x it returned back 1/3*x^3
curl -H 'Content-Type: application/json' -d '{ "prompt": { "text":
"integrate(x,x)"} }'
"https://generativelanguage.googleapis.com/v1beta3/models/text-bison-001:generateText?key=my_key_here";
"candidates": [
{
"output": "1/3 x^3",
"safetyRatings": [
{
"category": "HARM_CATEGORY_DEROGATORY",
"probability": "NEGLIGIBLE"
},
etc....
And when I asked it to integrate zero, it returned x
curl -H 'Content-Type: application/json' -d '{ "prompt": { "text":
"integrate(0,x)"} }'
"https://generativelanguage.googleapis.com/v1beta3/models/text-bison-001:generateText?key=my_key_here";
"candidates": [
{
"output": "x",
"safetyRatings": [
{
"category": "HARM_CATEGORY_DEROGATORY",
"probability": "NEGLIGIBLE"
},
etc...
may be i am not using it right. My be the syntax I am using is not right.
First time I try it. I am running these commands from Unix terminal.
--Nasser
On Saturday, December 16, 2023 at 8:23:00 PM UTC-6 Waldek Hebisch wrote:
> On Fri, Dec 15, 2023 at 09:33:17AM -0800, 'Nasser M. Abbasi' via FriCAS -
> computer algebra system wrote:
> > fyi, on related subject, I tested the MIT integration test problems with
> > open AI chatGPT last summer. It did not do well. Here is the result
> >
> >
> https://12000.org/my_notes/CAS_integration_tests/reports/summer_2022/test_cases/11_MIT//reportchapter1.htm
> >
> > The result of solved % on this one file only (over 300 integration
> > problems) were
> >
> > Mathematica 99.38%
> > Fricas96.57%
> > Maple 95.33%
> > Rubi 94.08%
> > Maxima 92.52%
> > Giac 91.59%
> > Mupad 90.03%
> > Sympy 82.24%
> > ChatGPT 14.33%
>
> IIUC this was version 3.5. Newer version is supposed to be
> much stronger. Few weeks ago OpenAI people claimed to made
> "breakthrough" in handling of math problems. Of course, ATM
> it is not clear what they have. But some things are possible
> to guess. There were earlier reports that proper "didactics"
> helps: start training on easy problems and gradually move
> to harder ones. Supposedly learing "step by step" reasoning
> is easier than "end to end" approach which was used (and
> quite sucessful) on other problems. There is also an
> obvious workaround: since ANN-s seem to have trouble with
> algorithmically trival math one could couple ANN to a
> "calculator", that is conventional program for doing
> calculations. So there are various approaches. In particular
> neural nets have enough capacity to memorize something like
> Rubi rules. If "trainig" could teach it to chain rules, than
> ANN could get to level comparable to Rubi. It would probably
> get some side conditions wrong, but it could compesate by
> having more rules and via "beam search" (that is producing
> many candidate solutions and using external differentiator
> to remove wrong candidates). Better net could learn to
> internally discard wrong candidates.
>
> I have no idea at what stage is Google offer. Public version
> is probably weaker than versions available to internal
> developers and paying customers. Still, progress seem to
> be rapid and there is something like a year from availability
> of ChatGPT 3.5.
>
> --
> Waldek Hebisch
>
--
You received this message because you are subscribed to the Google Groups
"FriCAS - computer algebra system" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to fricas-devel+unsubscr...@googlegroups.com.
To view this discussion on the web visit
https://groups.google.com/d/msgid/fricas-devel/804b5a56-bd84-4808-b7be-f05d7e2161f9n%40googlegroups.com.
