Positive users like the take that AI models show only superficial understanding in writing tests, finding amusement in how poorly they write readable text yet excel at coding.
Based on 1 visible X reactions from 1 accounts; directional sample.
Ask a question below.
Published answers will appear here.
I like this take! And I am also amused at how bad these models are at writing (that is worth reading) but then very good at coding! https://x.com/ScottWRobinson/status/2076758105057014270
On one end, this gives me a bit of peace of mind: the skill of thinking + writing not going away, not even online On the other: these things were supposed to be exceptional at writing, right? And the term "thinking" is associated with them. But I just don't see any evidence anything close to "thinking" is happening? Again, for these domains I am an expert in and I get a terribly wrong word salad from eg raw interviews based on which I have zero problem writing something coherent but SOTA models write things I judge as being wrong
If there is "understanding" it is superficial. I do find it useful to eg generate ideas, check if I missed something etc, but I still cannot find "thinking" in any domain I know very well... and a reason for me to keep getting to know domains very well
*GPT-5.6, not GPT-6, typo. I don't have any early access to any of these (and don't want it!)
I like this take! And I am also amused at how bad these models are at writing (that is worth reading) but then very good at coding! https://x.com/ScottWRobinson/status/2076758105057014270
On one end, this gives me a bit of peace of mind: the skill of thinking + writing not going away, not even online On the other: these things were supposed to be exceptional at writing, right? And the term "thinking" is associated with them. But I just don't see any evidence anything close to "thinking" is happening? Again, for these domains I am an expert in and I get a terribly wrong word salad from eg raw interviews based on which I have zero problem writing something coherent but SOTA models write things I judge as being wrong
If there is "understanding" it is superficial. I do find it useful to eg generate ideas, check if I missed something etc, but I still cannot find "thinking" in any domain I know very well... and a reason for me to keep getting to know domains very well
Positive users like the take that AI models show only superficial understanding in writing tests, finding amusement in how poorly they write readable text yet excel at coding.
Based on 1 visible X reactions from 1 accounts; directional sample.
Ask a question below.
Published answers will appear here.