OpenAI, the manufacturer of ChatGPT, has launched a new version of their much-discussed chatbot with great fanfare. Six questions and answers about the benefits and risks of this new paradigm.
Is there really a successor to ChatGPT?
In fact, the first version of ChatGPT only recently came to light. The chat program launched at the end of last year is versatile and can answer questions, make summaries or write poems. ChatGPT immediately generated a lot of interest in education because students used it with enthusiasm.
OpenAI creator foot Tuesday with much fanfare, the successor to the AI model that powers ChatGPT. Now that is still GPT-3.5, the new version is GPT-4.
Greg Brockman (gdb) or OpenAI just experimented with GPT-4 to build a working website from a sketch image from his notebook.
It’s the coolest thing I’ve ever seen in technology.
If you extrapolate from this demo, the possibilities are endless.
A glimpse into the future of computing. pic.twitter.com/1QB6wbQkld
– McKay Wrigley (@mckaywrigley) March 14, 2023
What exactly has been improved?
Quite a few. In the words of OpenAI founder Sam Altman, it’s more creative, makes fewer mistakes, and is less prone to it prejudice (Incorrectly ingrained prejudices and stereotypes). And perhaps most surprisingly: for the first time, this version of GPT is also capable of interpreting images. In addition, vocabulary has increased significantly. All this together makes the new model even more versatile.
break 🚨
OpenAI just dropped GPT-4 and it could literally blow your mind 🤯
GPT-4 is a large multimedia form that can accept image and text inputs and produce text output.
GPT-4 capabilities in a thread 🧵👇 pic.twitter.com/MCfRf6KKec
– Shubham Saboo (@Saboo_Shubham_) March 14, 2023
How do these images work?
The idea is that a user can submit an image to ChatGPT and then ask a question about it. OpenAI itself provides several examples. For example, a photo of a phone with an old-fashioned cable. GPT-4 can explain what’s wrong with that.
Or a picture of some of the ingredients for which the program suggests recipes. Perhaps even more impressive is the example of a handwritten diagram of a website, with the designation GPT-4: Make this a functional website.
Unfortunately, not all of this can be tested in practice. OpenAI says it needs more time to make this new component immune to abuse, without going into detail about it.
Greg Brockman (gdb) or OpenAI just experimented with GPT-4 to build a working website from a sketch image from his notebook.
It’s the coolest thing I’ve ever seen in technology.
If you extrapolate from this demo, the possibilities are endless.
A glimpse into the future of computing. pic.twitter.com/1QB6wbQkld
– McKay Wrigley (@mckaywrigley) March 14, 2023
The first version of ChatGPT was also criticized. How is it integrated into the new version?
Soon after the initial surprise and excitement about ChatGPT, there was already opposition: the program convincingly talks nonsense, is basically completely stupid, makes bizarre mistakes and contributes to disinformation.
GPT-4 can’t remove all objections, but according to its maker, it’s far more capable than its predecessors and “demonstrates human-level performance on various professional and academic benchmarks.”
As evidence, OpenAI ramps up some of the testing, with GPT-4 now performing in the top 10 percent, instead of the bottom 10 percent of the previous version. Meanwhile, founder Sam Altman was noticeably modest this time around, noting that his brainchild is still flawed and limited, and “looks more impressive on first use than after spending more time with it.”
Here is GPT-4, our most capable and compatible model yet. It is available today in our API (with queue) and in ChatGPT+.https://t.co/2ZFC36xqAJ
It’s still flawed, it’s still limited, and it still looks more impressive when you first use it than it does after spending more time with it.
– Sam Altman (@sama) March 14, 2023
Has the criticism gone?
at least. Although GPT-4 “hallucinates” much less than its predecessors, according to its creator, you as a user can still be unsure that the program always tells the truth. Critics argue that this is a particular problem with a search service like Microsoft’s Bing – which already uses GPT-4.
Another concern is OpenAI’s lack of openness. He. She companion search From an AI company is a joke, according to Gary Marcus (a professor of neuroscience at New York University): OpenAI pretends to be scientific when it isn’t. “It’s a step backwards for science,” He states in his newsletter. We don’t know her size (model, Mr. Dr.) He is; We don’t know what the architecture is, we don’t know how much energy was used; We don’t know how many processors were used; We don’t know what he was trained for.”
I guess we can call it closed to AI “opening”: the 98-page paper introducing GPT-4 proudly declares that they disclose *nothing* about the contents of their training set. pic.twitter.com/dyI4Vf0uL3
– Ben Schmidt / @benmschmidt @vis.social (@benmschmidt) March 14, 2023
Can I actually use it?
OpenAI is already making GPT-4 available in ChatGPT, but only on the paid ($25 per month) Plus plan. And yet without pictures. Moreover, it seems that Microsoft’s Bing is already using the new model for its chat search service, and many applications using GPT-4 will undoubtedly be available in the near future. An example is the BeMyEyes app, which was developed for the blind and visually impaired.
“Twitter junkie. Lifelong communicator. Award-winning analyst. Subtly charming internetaholic.”