OpenAI's New AI CriticGPT is The Reason Why ChatGPT Will Soon Be Unbeatable
ฝัง
- เผยแพร่เมื่อ 15 ต.ค. 2024
- OpenAI has developed CriticGPT, a new AI model designed to enhance the accuracy and reliability of code generated by ChatGPT. CriticGPT acts as an AI code reviewer, identifying errors and providing critiques to improve code quality, ultimately making ChatGPT a more powerful and reliable tool for developers. This advancement in AI technology highlights OpenAI's commitment to pushing the boundaries of AI development and improving the performance of AI models like ChatGPT.
#chatgpt #openai
![เปิดใจ "อาจารย์เบียร์ คนตื่นธรรม" อดีตเคยสุดทุกทาง ถ้าใช้หนี้หมดไปบวชแน่ l [Nickynachat]](http://i.ytimg.com/vi/IxeE0xVhy08/mqdefault.jpg)
Okay, so I take the output from chatgpt and paste in criticgpt and say critic it and then go back to chatgpt ? or is there a pipeline linking these two model or even simply, as a user why should i care about criticgpt ? i am hoping criticgpt is a background thread of chatgpt ? someone please let me know?
But who to criticize CriticGPT?
The lawyers 😂
But who to criticize who criticizes CriticGPT?
Exactly
@@manonamission2000Critic Critic GPT
What I do know is that it is easier to criticize a result than to achieve it. A good critic points the way to improvement, even if they can't accomplish what they advise. Reinforcement Learning from Human Feedback (RLHF) is based on the idea that it is easier to evaluate a result well than to follow the process to achieve that result. This is why it is possible to build models that are smarter than the humans who provide their feedback in RLHF. Humans continue to critique the results of all AI, including CriticGPT. CriticGPT helps us better evaluate the outcomes of other AIs by explaining complex results or by directly producing its critiques. In many cases, it would be much easier for humans to criticize the output from CriticGPT than to criticize the outputs from another AI that CriticGPT has analyzed. This is because CriticGPT's explanations and critiques are designed to be more comprehensible to humans, breaking down complex AI outputs into more accessible terms and concepts.
All people need to do is ask any ai “are you sure” and it usually finds the problem in the response it has just given you. Not sure why ai can’t ask itself that before giving you an answer.
The problem is that code libraries are updated faster than chatgpt or any GenAI model, so GenAI can never keep up with the latest updates... so, the solution should be RAG injection into the ChatGPT to acquire the latest code updates/upgrades from the web, but even then GenAi can fail as it would need follow up access to the lates codes upgrades from the web, so it needs to be in a loop to get it all corrected, but then it will slow everything down and quadruple GenAI run costs. We need a GenAI model that is not static, but a dynamic model that keeps updating itself constantly with the latest data on the web and opensource libraries.
No that is wrong. If you ask it are you sure it will change its answer even if it had been correct. And it does it again and again with a different wrong answer
Is CriticGPT found in the GPT store, or is it a separate model?
I think it's more like Claude constitutional ai? But we can see it like what Gemini wanted to be.
When will critic gpt be released?
What's stopping other AI companies doing the same thing for their models? they would likely also develop such thing as CriticGPT for their own AI chat models
Extremely well explained.
🇧🇷🇧🇷🇧🇷🇧🇷👏🏻, Amazing video, good job! Thanks for sharing.
I like this channel am not sure if it's AI but it's so good I don't care
it's AI, listen how he always spells "GPT" cringe :)
Surely Chinese companies can just access GPT with a VPN. And don't Chinese companies already have their own AIs?
Despite their propaganda, everything that comes from China is rushed and usually breaks. Would you expect their Ai to be any better? They can't even hold onto their space rockets these days 😂😂😂
@@TheRealKingYeti Seeing as everything comes from China, that would mean that everything usually breaks. I don't think so.
If CriticGPT is preferable by 60% of reviewers, that means it's efficiency is 10% better that without it. It's something but it isn't massive by any way.
how did you calculate to get 10% please
@@elawchess A 50-50 split would represent a random choice, like flipping a coin. CriticGPT being preferred by 60% of reviewers means it's 10 percentage points above this random baseline (60% - 50% = 10%).
@@chanpasadopolska Thanks, but I checked with both ChatGPT and Claude and they get 20% instead of 10%. Here’s the quick math: if users with CriticGPT outperform those without it 60% of the time, that means CriticGPT has a success rate of 60% (0.6) compared to 50% (0.5) without it. The improvement is calculated as (0.6 - 0.5) / 0.5 = 0.1 / 0.5 = 0.2, which is a 20% improvement.
@@elawchess I give absolute improvement and you relative one
@@chanpasadopolska OK thanks.
CriticGPT still needs human reviewers, who now have to review both ChatGPT and CriticGPT. :-)
CriticGPT how can I get it ?
Hello as criticGPT I evaluated your request and decided you suck... Thankyou for using criticGPT.
Like realizing that OAI cant even program for its own latest API?
Damn good point pro
Hey bro,I'm a marketing manager from an AI company,Would you like to talk for cooperation?
About China access to OpenAi: Such a non news. Not only the Chinese govt was first to want to block it, but with vpn it makes no difference... the only real difficulty to make it difficult to access is about payments, and maybe phone number confirmations, but this is the case since the start... I don't see the change.
How To : Control info like a hidden dictatorship.
Nice
They need to create one call idiot GPT. Lot of people will get it. Lol 😅
I don't get it.
You have been trolled.
My spell correct is often wrong 😑 and calling a tech company AI customer service maze is never helpful . Install me about your latest miracle AI 😆😆
Create the problem then sell the solution. Got it.
The human condition.
I broke chat gpt
Another AI scam presented by closed AI.
Критин ЖиПиТи
Its must not repeat the same code rather update the existing ask if want to update create a new response alot can be modified just don't understand who they hire but does people are not advancing layout first if can write with create design this will help, also some of them tell what to do, rather than doing their task at hand, But chat gpt has now able to self correct some mistakes but Gemini is like chatting to a wall 🧱 😂😢, Have nosense of humour nothing cold hard steel chatting service 🐕🦺, it's like chatting to questions over questions, you ask it ask Lol 😂😆 you chat random stuff it put it put to project and didn't say so it's like its not s conversation but command line on terminal Lol 😂😆 it's crazy Google AI not yet in the future yet other AI's getting more smarter , Google I don't know what going on there but they need alot of improvements, Google must stop insulting ours AI they have none yet chatbot like robo conversation, not natural language it must feel like a human conversation not bot conversations 😂, i bet even the premium version it sucks. Lol 😂😆 they need fresh ideas its like the 90s AI Lol 😂😆 chatbot , i still have it Lol 😂😆 i think it's way smarter than than google Gemini Lol what are they doing falling behind Lol 😂😆