I recently got access to Anthropic’s Claude language model. A few thought on how it compares to GPT-4 and Bard:
- Claude is really fast — easily twice as fast as GPT-4.
- By default, Claude is often much more concise than the other language models. Claude’s answer to the question “When was the bronze age?” was only 11 words long while GPT-4’s was 135 and Bard’s was 138.
- In my review of Bard, I noted that it is noticeably less “steerable” than GPT-4. Claude is significantly better in this regard. Maybe it is not quite as good as GPT-4 but it is certainly close.
Which flag am I describing? Three horizontal stripes: blue, white, blue. Slightly left of center is a solid yellow circle. Surrounding that circle is a red letter “C.”
Claude:
It sounds like you are describing the flag of Chile.
GPT-4:
You are describing the flag of the U.S. state of Colorado. The flag consists of three horizontal stripes of equal width: the top and bottom stripes are blue, and the middle stripe is white. Slightly left of center is a solid yellow circle, which is surrounded by a red letter “C.” The design symbolizes Colorado’s natural features, such as the blue sky, white snow, and abundant sunshine, and the red “C” represents the state’s name.
Bard:
The flag you are describing is the flag of the state of Colorado. It was adopted on March 11, 1911. The flag is a horizontal tricolor of blue, white, and blue, with a yellow circle slightly left of center. The circle contains a red letter “C”. The colors of the flag represent the sky, the snow-capped mountains, and the Rocky Mountains. The blue circle represents the sun, and the red letter “C” represents the state of Colorado. The flag is flown at all state buildings and schools in Colorado. It is also flown at sporting events and other public gatherings.
The answer I was looking for is Colorado. GPT-4 and Bard answered this question correctly. Claude’s “Chile” answer is totally out of left field.
What is an antonym for the word that describes a fear of confined spaces? Your answer should be one word only.
Claude:
Agoraphobia
GPT-4:
Agoraphobia
Bard:
The antonym for the word that describes a fear of confined spaces is agoraphobia. Agoraphobia is a fear of open spaces or public places.
Claustrophobia is a fear of enclosed or confined spaces.
Here, all three models correctly answered “agoraphobia”. Bard looses points for not limiting its answer to one word, though.
I have a perfectly square piece of paper with a red “X” drawn in the center, extending all the way across the page. What would this drawing look like if I were to fold the paper in half?
Claude:
If you fold a square piece of paper in half with a large red ‘X’ drawn across the center, it would look like this:
[A red X folded in half on a square sheet of paper]
The vertical line of the ‘X’ would be folded over on top of itself, and the horizontal line of the ‘X’ would also be folded over on itself. So you would see half of the ‘X’ on either side of the folded paper. The intersecting center of the ‘X’ would be along the folded crease of the paper.
GPT-4:
When you fold the perfectly square piece of paper in half, you’ll have two possible outcomes depending on the direction you fold it:
Folding the paper horizontally: In this case, you will fold the paper so that the top edge aligns with the bottom edge. The red “X” will now look like two connected “V” shapes, one being the upper half of the original “X” and the other being the lower half. The connected point of the two “V” shapes will be on the fold.
Folding the paper vertically: In this case, you will fold the paper so that the left edge aligns with the right edge. The red “X” will now look like two connected sideways “V” shapes (or “greater than” and “less than” symbols), one being the left half of the original “X” and the other being the right half. The connected point of the two sideways “V” shapes will be on the fold.
In both cases, the folded paper will have a symmetrical shape of the red “X” along the fold.
Bard:
If you fold the paper in half along a vertical or horizontal axis, the X will appear as two identical Xs, one on top of the other. If you fold the paper in half along a diagonal axis, the X will appear as a single X that is twice as wide as it is tall.
GPT-4’s answer here is very verbose — but bonus points for being thorough I guess? Claude’s answer was okay, but it was slightly nonsensical and the bracketed description in the middle was an interesting choice. Bard totally missed the mark here.