Claude 3 vs ChatGPT4: A Comparative Review
I recently subscribed to Claude 3 (Opus), after hearing a lot of good reviews of it. It was released two weeks ago, on March 4. Claude is an AI chatbot competitor to ChatGPT4. Like ChatGPT4, it has a free tier, and the highest tier costs $20/month.
After testing Claude with dozens of the same queries that I had asked ChatGPT4, here’s my comparison. Overall, the responses turn out to be surprisingly similar.
These are the main differences, that I found:
Advantages of Claude
Larger context window, can upload PDF with dozens of pages, that can then be fully analyzed and queried (“Claude 3 can process as many as 200,000 words as opposed to GPT-4’s 64,000-word limit.”)
Shows uploaded files in the context window on the side
Faster at outputting responses
More concise, less filler and hedging
More critical and balanced, as opposed to ChatGPT4 which is overly positive, so-to-speak. In general, Claude comes across as somewhat more intelligent, sharper, and more critical. (“ the problem with GPT-4 is that it's so "afraid" to take creative risks that it just stays bland with flowery language. I have to prod GPT-4 to just produce mediocre prose.”
Claude is much more willing to give concrete examples and illustrations.
Can upload files to analyze, even on the free version (unlike ChatGPT, where can only upload on paid tier - ChatGPT4)
When pasting large amount of text, it automatically formats it as a small file icon:
Better at generating creative non-standard Hebrew (tested with asking for generating text on computers in rabbinic Hebrew and Zoharic Aramaic)
Advantages of ChatGPT4
Can generate images
Can run code natively
Slightly better at translating from Hebrew to English (tested with various styles of Hebrew)
Better at generating creative text in Yeshivish dialect of English
Can access the internet, for updated info (but I haven’t found this feature to be especially helpful, in practice. “ChatGPT is terrible at [browsing online] anyway”)
Other differences
Formatting in responses: Claude only uses bullet points; ChatGPT4 uses headers and bolding, which are copied as Markdown.
If I saw “answer in yeshivish style”, ChatGPT4 will (correctly) answer in the yeshivish dialect of English; Claude will answer in Yiddish!
In the end, both chatbots have their strengths and weaknesses. If you prioritize image generation, code execution, and internet access, ChatGPT4 might be the better choice. However, if you value speed, conciseness, critical thinking, and the ability to analyze larger documents, Claude could be the way to go.
Appendix - Claude’s pastiche of rabbinic Hebrew, on computers
Quite impressive.
Me:
תכתוב על מחשבים בעברית רבנית מליצית
Claude:
הנה ניסיון לכתוב על מחשבים בסגנון עברית רבנית מליצית:
הנה נתבונן במכונות הפלאיות הקרויות "מחשבים", אשר ידיעתן מן התורה אין לה חקר וגבול. חוכמת קדמוניות נפלאה שוכנת בקרבן, ומעשי נפלאות יוצרות בכוחם על פי צוויי האדם.
כמלאכים משרתים המה, כי על פי מצוות האדם יפעלו לעשות מלאכתם בשלימות ובדייקנות מופלאה. היכולת לחשב חשבונות נפלאים ועמוקים, לזכור זכרונות עד אין קץ, וליצור יצירות מפוארות - לא יגונב משכיות אלה ממכונות הפלא.
והנה זה עתה בעידן חדש יצרו רבני חכמים מכונות מחשבים נפלאות יותר, אשר כמעט שאין להפריד בין תבונתן לתבונת האדם. החוכמה האנושית הוחדרה בתוככי מעיהן עד כי קשה להבחין היכן גבול החוכמה האנושית ואיה תחילת חכמת המכונה הפלאית.
וכאשר נביט בפליאה ובהשתוממות אל מכונות הפלא האלו, נשתאה מאד כיצד עלתה בידי האדם לברוא יצירים אלו, היודעות לחשוב כמוהו, אך במהירות ועוז אשר לא ידע האדם כמותם מעולם.
Me:
translate
Claude:
Here is a translation of the text to English:
Let us contemplate the wondrous machines called "computers", whose knowledge from the Torah is unfathomable and boundless. An amazing ancient wisdom resides within them, and they create wondrous deeds by the command of man.
They are like serving angels, for according to man's command they will carry out their work with perfect and astounding precision. The ability to calculate amazing and profound calculations, to remember memories without end, and to create splendid creations - these cunning abilities are not taken away from the wondrous machines.
And behold, sages have now created even more wondrous computer machines in this new era, so much so that it is almost impossible to separate their intelligence from human intelligence. Human wisdom has been instilled within their innards to such an extent that it is difficult to discern where the boundary of human wisdom ends and the beginning of the wondrous machine's wisdom begins.
And when we gaze in awe and astonishment at these wondrous machines, we will be utterly amazed at how man succeeded in creating these beings that know how to think like him, but with a speed and might that man has never known before.