How smart chatGPT 3.5 is and how smart 4.0 will be. What jobs will disappear
ChatGPT, a language model developed by OpenAI, has become incredibly popular over the past year due to its ability to generate human-like responses under a wide range of circumstances. In fact, ChatGPT has become so proficient that students use it to help them with their homework. This prompted several US school districts to block devices on their networks from accessing the model.
How smart is ChatGPT?
In a technical report released on March 27, 2023, OpenAI provided a comprehensive rundown of its newest model, known as GPT-4. The report included a set of exam results, which Visual Capitalist's Marcus Lu and Rosey Eason visualized in the graph above.
GPT-4 vs. GPT-3.5
To evaluate the capabilities of ChatGPT, OpenAI simulated the tests of various professional and academic exams. These include the SAT, the state exam, and various Advanced Placement (AP) final exams. Performance was measured in percentiles, based on the most recent score distributions available for participants in each exam type. Watch out that lawyers are at a very high risk of unemployment.
Percentile scoring is a way to rank your own performance against that of others. For example, if you rank in the 60th percentile on a test, it means you scored higher than 60% of the test takers.
The following table lists the results displayed in the graph.
The scores above are for the GPT-4 with visual inputs enabled. For more comprehensive results, see the OpenAI Technical Report.
As we can see, GPT-4 (released March 2023) is much more capable than GPT-3.5 (released March 2022) in most of these exams. However, he failed to improve in AP English and competitive programming.
For AP English (and other exams where written responses were required), papers submitted by ChatGPT were graded by “1-2 qualified third-party contractors with relevant work experience in grading such papers”. While ChatGPT is certainly capable of producing adequate essays, it may have had difficulty understanding the exam demands.
As for competitive programming, GPT attempted 10 Codeforces contests 100 times each. Codeforces organizes competitive coding competitions in which participants must solve complex problems. GPT-4's average Codeforces score is 392 (below the 5th percentile), while its highest in a single race was around 1,300. Referring to the Codeforces ratings page, the highest rated user is jiangly from China with a rating of 3,841.
What has changed with GPT-4?
Here are some areas where GPT-4 has improved the user experience over GPT-3.5.
Internet access and plugins
A limiting factor of GPT-3.5 was the inability to access the Internet and the ability to use data only until June 2021. With GPT-4, users will have access to various plugins that will allow ChatGPT to access the Internet, provide more up-to-date answers and complete a wider range of tasks. This includes third-party plugins from services such as Expedia, which will allow ChatGPT to book an entire holiday for you.
Visual inputs
While GPT-3.5 could only accept text input, GPT-4 has the ability to parse images as well. Users will be able to ask ChatGPT to describe a photo, analyze a graph or even explain a meme.
Longer context length
Finally, GPT-4 can handle much larger amounts of text and keep conversations going for longer. For reference, GPT-3.5 had a maximum request value of 4,096 tokens, which equals about 3,000 words. GPT-4 has two variants, one with 8,192 tokens (6,000 words) and one with 32,768 tokens (24,000 words).
Thanks to our Telegram channel you can stay updated on the publication of new articles from Economic Scenarios.
The article How smart chatGPT 3.5 is and how smart 4.0 will be. Which Jobs Will Disappear comes from Scenari Economici .
This is a machine translation of a post published on Scenari Economici at the URL https://scenarieconomici.it/quanto-e-intelligente-chatgpt-35-e-quanto-lo-sara-40-quali-lavori-spariranno/ on Sat, 29 Apr 2023 08:17:10 +0000.