Chinese Search Giant Baidu To Launch ChatGPT-Style Bot (bloomberg.com) 24
Baidu is planning to roll out an artificial intelligence chatbot service similar to OpenAI's ChatGPT potentially China's most prominent entry in a race touched off by the tech phenomenon. From a report: China's largest search engine company plans to debut a ChatGPT-style application in March, initially embedding it into its main search services, said the person, asking to remain unidentified discussing private information. The tool, whose name hasn't been decided, will allow users to get conversation-style search results much like OpenAI's popular platform.
Baidu has spent billions of dollars researching AI in a years-long effort to transition from online marketing to deeper technology. Its Ernie system -- a large-scale machine-learning model that's been trained on data over several years -- will be the foundation of its upcoming ChatGPT-like tool, the person said. ChatGPT, OpenAI's artificial intelligence tool, has lit up the internet since its public debut in November, amassing more than a million users within days and touching off a debate about the role of AI in schools, offices and homes. Companies including Microsoft are investing billions to try and develop real-world applications, while others are capitalizing on the hype to raise funds. Buzzfeed's shares more than doubled this month after it announced plans to incorporate ChatGPT in its content.
Baidu has spent billions of dollars researching AI in a years-long effort to transition from online marketing to deeper technology. Its Ernie system -- a large-scale machine-learning model that's been trained on data over several years -- will be the foundation of its upcoming ChatGPT-like tool, the person said. ChatGPT, OpenAI's artificial intelligence tool, has lit up the internet since its public debut in November, amassing more than a million users within days and touching off a debate about the role of AI in schools, offices and homes. Companies including Microsoft are investing billions to try and develop real-world applications, while others are capitalizing on the hype to raise funds. Buzzfeed's shares more than doubled this month after it announced plans to incorporate ChatGPT in its content.
How will it be trained? (Score:2, Insightful)
Re: (Score:3, Funny)
A Chinese AI is always happy. A Chinese AI only does the right thing for the State. A Chinese AI outs you if you disagree with the State. What could go wrong?
Re: (Score:2)
I haven't seen that opinion much. Can you support it? I, for one, seriously doubt this Chinese AI has the computing power behind it to echo ChatGPT's capabilities, or the innovation to do it on less.
Re: (Score:2)
Of course there's an AI gap with China. And a missile gap with North Korea. And a chip gap with, uh, China again.
Re: (Score:2)
While tools derived from ChatGPT have proven to be effective in providing conversational-style search results, relying solely on Baidu's Ernie system may not be the best approach. Despite years of investment in AI research and development, the success of a tool like ChatGPT is largely dependent on the quality and diversity of the data it is trained on. If Baidu's Ernie system is trained on biased or limited data, it may produce results that are incomplete, inaccurate, or reflective of harmful biases.
There c
Re: (Score:3, Insightful)
All those 5 points above may be also true for ChatGPT
Re: (Score:3)
China is widely recognized as being 'ahead' in the AI 'race'.
I don't think that opinion is widely held. The opinion I believe is widely held is that China has the opportunity to race ahead of the US in AI implementations if they continue their current focus in the area. They cannot compete with the US in higher education and private research, because foreign students don't flock to Chinese universities and US companies simply pay more. But China can leverage US-based research which is made public (and that they can buy in acquisitions). They also likely won't have to
Re: (Score:2)
Show me the money (Score:5, Insightful)
My first reaction was almost to laugh. AI is not a field where you get cred by making claims about what you're going to do. (Wasn't AGI supposed to be solved by the 80s?) So until they actually do it, we should read this article as "Baidu has the ambition to launch a ChatGPT-style bot in its search results."
Re: (Score:2)
Re: (Score:3)
I do find it fascinating the ways tools like ChatGPT compensate for their weaknesses. For example, let's give ChatGPT a random math problem it's likely never seen before:
Q: What is 13.924 * 86.127?
A: 1203.4824868.
Now, one can look at that and say, "Aha! You see, the real answer is 1199.232348! This just demonstrates ChatGPT's weaknesses; it's just a language models, it can't learn 'skills' like how to multiply!
But I look at it rather differently. Yes, it's answer was off, but it was only off by less than
Re: (Score:2)
I think the math problem examples say more about us than they do about the machine. The goal of the field of AI is to study and develop algorithms that allow machines to perform more like people or animals do. So we deride ChatGPT by noting that it can't multiply numbers that we can't either.
You're right, it is fascinating that a language model can
a) learn to answer questions it can't answer by making shit up
b) apply that skill to arithmetic problems
c) maybe have learned some kind of estimating procedure th
Re: (Score:2)
And it's (C) that I find interesting.
Anyone can "make shit up". But the fact that it can get close with "some kind of estimating procedure" is to me quite interesting.
Re: (Score:2)
Yes, *anyone* can make shit up. That's why lying is an AI problem. But yes, C is most interesting.
Re: (Score:3)
answer:
13.924 * 86.127 = 1198.883068
Step by step:
Multiply the ones place of both numbers: 4 * 7 = 28
Multiply the tens place of both numbers: 2 * 7 = 14
Add a 0 to the end of 14: 140
Multiply the hundreds place of both numbers: 9 * 6 = 54
Add 140 and 54: 140 + 54 = 194
Add a 0 to the end of 194: 1940
Multiply the thousands place of both numbers: 9 *
Re: (Score:2)
But those steps are nonsense and the answer is still wrong. It resembles the steps to solving a multiplication problem without actually being them. Yet is close, which is what I find interesting.
Re: (Score:2)
I'm guessing it will be exactly as useful as ChatGPT, if only because it's actually just a user interface for ChatGPT.
Re: (Score:2)
ChatGPT has been demonstrated. Baidu is perfectly capable of training one. Google, Bing and a dozen others would be released tomorrow if any of them could figure out how the thing is supposed to actually make money.
Re: (Score:2)
Google would release it ASAP if they had it, mainly to prevent competitors from gaining market share. They could figure out how to monetize it later, as they did with most of their other products.
Re: (Score:2)
Remember the story a while ago about the chat system that convinced one of its engineers it was sentient? That was at Google. They've had one of these for some time. And no, Google is crazy, but they won't just turn on something that turns their profit margin negative.
Re: (Score:2)
Good point about the engineer, I stand corrected.
Re: (Score:2)
Actually it seems I was right about the motivation and effect but wrong about the timeline:
https://gizmodo.com/google-cha... [gizmodo.com]
Google is feeling the heat from ChatGPT and will be releasing their own chatbot as a result. So your concern about it costing them money may be true, but is not a deciding factor in their decision.
Race to the bottom (Score:2)
The obvious outcome of GPTChat/AI/Large Language Models is that the shit being produced, often called "content", will become rivers of diarrhea. What on the internet currently isn't a honeypot to gather information? What internet site hasn't become a maze of barely credible words, linked together in a circular fashion, to capture you, to lead you down the rabbit hole that they made? I haven't gotten any information that is of any value for a long time now. Everything