ChatGPT对基组的胡言乱语

sobereva

今天看到有人在CCL上发了他从chatGPT上返回的信息



Hi,

the following is from chatGPT, what do you guys think ?


When using the 6-311G basis set, it is common to add diffuse functions to improve the description of electron correlation and dispersion interactions. One commonly used set of diffuse functions is the "polarization-consistent basis set" (pcS-n, where n is an integer), which adds diffuse functions to the valence and polarization sets in a consistent way.

For Fe2+ ferrous ion, a suitable choice of diffuse functions would depend on the level of accuracy required and the size of the system. As a starting point, you could consider adding diffuse functions up to the pcS-4 level, which would add 4 sets of diffuse functions to the 6-311G basis set.

To specify the pcS-4 basis set in Gaussian, you can use the following basis set specification:


```
# opt umn15l/6-311++g(d,p) scrf=(solvent=water) geom=connectivity

Fe2CO2_OPT

2 3
Fe                 2.74538330    8.28679554    5.00000000
O                  4.55208397    8.06717607    5.00000000
C                  5.30819317    9.07309328    5.00000000
O                  5.97838127    9.96470142    5.00000000

1 2 1.0
2 3 2.0
3 4 3.0
4

Fe S 4
6-311G(d,p)  Fe 4,4,4,4,4,4
****
C S 1
6-311G(d,p)  C 3,3,3,3,3,3
****
O S 2
6-311G(d,p)  O 3,3,3,3,3,3
****
```


In this input file, the "S" keyword specifies that diffuse functions should be added to the basis set, and the "4" after "Fe S" indicates that 4 sets of diffuse functions should be added to the iron atom. The same format is used for the carbon and oxygen atoms.



pc系列基组提出者Frank Jensen回复：It is difficult to find anything correct in the free flowing text below.

wjc404

需要计算化学工作者给chatGPT高质量的人工标注数据。。从很多不专业的文章上瞎学可不行。不过估计OpenAI不会把这方面的质量优化放在很高的优先级。
不过换句话说，什么时候聊天机器人智能到能够自动听课学习（而不是需要人们手动操作其中的数据准备和训练）了，这个问题就好办了。

朽木可自雕

跟550W比差了亿点

flyingchow

未来已来，珍稀还可以找茬的时光吧。

刘小英雄steven

flyingchow 发表于 2023-3-7 02:05
未来已来，珍稀还可以找茬的时光吧。

wzkchem5

wjc404 发表于 2023-3-3 01:36
需要计算化学工作者给chatGPT高质量的人工标注数据。。从很多不专业的文章上瞎学可不行。不过估计OpenAI不 ...

高质量的人工标注数据好解决，在学习的时候把咱们论坛的数据的权重提高10倍就行了

钠离子电池

主要还是训练的语料太少了吧。毕竟第一性原理计算、量化计算、分子动力学模拟不算大的科学方向。