The debut of OpenAI ChatGPT has been one of the most talked about topics in AI tech today.
Text-davinci-003 is a new model in the GPT-3 family of AI-powered large language models that the company unveiled on Monday. It is a member of what it refers to as the “GPT-3.5 series,” and it reportedly outperforms its forerunners by handling more complex instructions and generating higher-quality, longer-form content. On the topic, Twitter user and data scientist @minimaxir tweeted:
GOD DAMN IT @OpenAI STOP ANNOUNCING THINGS I AM TOO BUSY https://t.co/F7Xd511FAf
— Max Woolf (@minimaxir) November 30, 2022
The new model improves on ChatGPT by integrating reinforcement learning with human feedback to more closely match language models with instructions from humans. Davinci-003 is a genuine reinforcement learning with human feedback (RLHF) model, in contrast to davinci-002, which employs supervised fine-tuning on human-written demos and highly scored model samples to enhance generation quality.
In the demo, OpenAI ChatGPT can complete complicated tasks such as:
- Q&A
- Grammar correction
- Summarizing a conversation
- Text to command prompt generation
- English to other languages translation
- Classifying items
- Translating Phyton to natural language
- Classifying tweets
- Extracting keywords from a text
- Create a spreadsheet
- Creating a study note
If you would like to try out OpenAI ChatGPT yourself, you may click the link here.
Promising demo of OpenAI ChatGPT
The research release of ChatGPT, a recent OpenAI blog article states that “the latest step in OpenAI’s iterative deployment of increasingly safe and useful AI systems.” The safety mitigations in place for this release have been guided by several lessons learned through the deployment of prior models like GPT-3 and Codex, including a significant decrease in harmful and untruthful outputs gained by the application of reinforcement learning from human feedback (RLHF).
In the blog, we see big improvements to the OpenAI ChatGPT. AI could write down a limerick (a humorous five-line poem with a rhyme scheme aabba.) and a summary of the subject it was talking about with the user.
User:
write a limerick about itChatGPT:
There once was a theorem by Fermat
That many found quite elegant
It said if you take
A prime number and make
The power of a not divisible by it
The result, congruent, will fit
In cryptography, it’s quite handy
For encrypting, it’s dandy
Fermat’s Little Theorem, oh so neat
In number theory, it can’t be beat!User:
summarize the conversation so farChatGPT:
Fermat’s Little Theorem states that if pp is a prime number and aa is an integer not divisible by pp, then a^pa
(aa to the power pp) is congruent to aa modulo pp. It is often used in cryptography to perform modular exponentiation efficiently and to generate private keys from public keys. It is also a useful tool in number theory for proving other theorems. I also wrote a limerick about Fermat’s Little Theorem.Conversation between user and ChatGPT – from OpenAI blogpost
Limitations of OpenAI ChatGPT
Despite this, OpenAI described ChatGPT’s “limitations” in its blog post, including the fact that occasionally, answers are plausible-sounding yet mistaken or absurd. Thinking that ChatGPT is still an early demo, these kinds of problems are expected. In their blog post OpenAI stated:
“Fixing this issue is challenging, as: (1) during RL training, there’s currently no source of truth; (2) training the model to be more cautious causes it to decline questions that it can answer correctly; and (3) supervised training misleads the model because the ideal answer depends on what the model knows, rather than what the human demonstrator knows.”
ChatGPT will “sometimes respond to harmful instructions or exhibit biased behavior,” according to Open AI. Although we anticipate some false negatives and positives, for the time being, we are leveraging the Moderation API to alert users or prohibit specific categories of hazardous content. In order to help us in our continued efforts to improve this system, we are glad to gather user input.
And that covers all of the current updates on the OpenAI ChatGPT demo. The topic of AI and human conversations, which has recently come to the fore with Meta Galactica AI recently, is really interesting.
Are you into AI tools? Check out other articles:
- Free MyHeritage Historical AI alternatives
- Interior AI: Redesign your room effortlessly
- What if you were a Viking? MyHeritage AI Time Machine knows
- Meta’s Cicero AI achieves human-level performance in board game Diplomacy
- How long has AI been helping us?
We wish you a good day; keep reading us for more AI news.
