Is the hotly popular generative AI ready for enterprise applications?

Just a few months ago, OpenAI released ChatGPT, which instantly became a global sensation. Today, it has more than 100 million active users, relying on generating human-like and grammatically accurate response content. Similar technologies can even generate artwork and programming code based on input descriptions, with equally impressive results.

If the initial response is unsatisfactory, you can continue to interact with the AI, such as asking additional questions or adjusting the image or code content through supplements and clarifications to make the results more in line with your expectations. The whole process does not require domain experts, artists or programmers to work, it can be realized by speaking.

But generative AI also has its own problems, such as the source of data used to train the underlying AI model may be controversial, how the training data should be circulated, lack of access to source data, bias in the model, and More importantly - the accuracy of the content of the response, especially "serious nonsense".

But these problems have not stopped software companies from trying generative AI at all. Enterprises see the huge business potential and user enthusiasm behind this, and they certainly don’t want to fall behind.

Salesforce, Forethought and Thoughtspot have all recently announced their own generative AI testing solutions.

Salesforce launches generative AI across its own platforms, Forethought hopes to build a new chatbot, and Thoughtspot hopes to use AI to improve data query. Each company adds some algorithmic tweaks to the underlying technology to reflect the unique needs of its platform.

Microsoft announced that the OpenAI Service for Azure enterprise users is now fully open as a managed service.

Throughout 2023, I believe more companies will join the fray. But the limitations we mentioned earlier are still real, so a new question arises: No matter how cool it looks on the surface, no matter how much developers emphasize that it is still in the early stages of development, is generative AI really ready to serve the enterprise? ?

Start with limitations

Enterprise customers are working hard to introduce AI for business purposes, but there are still too many unknowns.

Currently, the underlying models of generative AI are still trained using Internet data and do not fully meet various licensing requirements. Whether it’s a website, a book, or an article text, this widespread collection of training material has real consequences for everyone, but especially for companies that create content for commercial purposes.

Salesforce founder Marc Benioff said in a previous interview that this is indeed an obvious flaw, but it did not prevent Salesforce from launching Einstein GPT.

The CEO emphasized at the time: "We all feel the exciting capabilities of ChatGPT, but we can also see its limitations. It is actually the ultimate 'plagiarist', and everything it learns is It was stolen from others. So its boundary is the boundary of the content it can crawl."

More importantly, a lot of the content is simply unreliable, or at least contains certain errors. . OpenAI also publicly admitted this in its list of technical limitations, writing: "ChatGPT sometimes gives answers that seem reasonable, but are incorrect or even ridiculously wrong. And this problem is difficult to solve..."

Deon Nicholas, CEO and co-founder of Forethought, believes that the biggest problem facing generative AI is wrong answers. "ChatGPT still can't get rid of 'illusion', right? Ask it a question about a specific business, and if it doesn't know the answer, it will just make up something that seems reasonable but is actually completely wrong."

In addition, the information used for training by ChatGPT is due in 2021, which is also a problem for companies looking to create new businesses.

There is also the issue of bias, which requires careful attention to models and training data by a diverse team to truly mitigate it. Last week, Neha Bajwa, Microsoft’s head of customer experience product marketing, also talked about the importance of AI bias in an interview.

"At Microsoft, we call this 'Responsible AI', which means ensuring that there is no bias in the data by paying attention to bias and inclusivity, so that AI has a reasonable ethical perspective and responsible characteristics. It must Be aware that data can amplify bias."

Solution

But these limitations are not insurmountable. Recently, several generative AI tool developers have begun to adopt OpenAI's basic model and transform it into their own technology to solve some of these problems. Although it has not been completely eliminated, efforts in this area will definitely become the focus of the next stage of work.

O’Reilly Media founder, chairman and CEO Tim O’Reilly sees ChatGPT as a true third wave of the Internet, but said some fine-tuning may be needed to meet the business needs of content owners.

OpenAI CEO Sam Altman has contacted O’Reilly and hopes to use the O’Reilly book catalog to obtain a knowledge training corpus. But O’Reilly expressed opposition because this method currently lacks a reasonable author compensation mechanism.

O'Reilly said: "I said, unless you find... some kind of payment method, because the content has a clear subject and people are willing to pay for it." He suggested setting up a system, Make sure users pay to access this type of professional content.

"The payment will go to the owner of the source content. Maybe we will establish a business model for this so that people can access this more authoritative content."

One way of doing this The biggest advantage lies in broadening the breadth of generative AI capabilities, including text, art, and code. Nicholas believes that if future models can generate code based on workflow, or automatically create or adjust execution methods, it will definitely become a powerful help for technology application companies.

"I would add something that people may not realize yet. Generative models like GPT-3 can generate code and of course can be used to generate real-time workflows, and the results are quite good. So it represents It is not only an AI model that can talk and think with us like humans, but will also become an important basis for many jobs - including generating Python code and generating automated workflows."

Digital Experience and Dries Buytaert, CTO of Acquia, a digital strategy consulting company, once founded the open source Drupal content management system. In his recent blog post, he introduced the application prospects of generative AI in content management and daily business.

In a recent interview, Buytaert compared the development of AI technology to cloud computing. By opening up low-threshold access to computing resources, cloud computing has fundamentally reshaped the face of enterprise computing.

"What OpenAI builds is not just a product, but also a popularization trend of AI tools. It allows many people without doctorates in machine learning and AI to quickly build practical results. Such an empowering effect makes People are amazed."

Buytaert suggested that companies should at least show their development process and how the AI model reaches the answer. “They must show sincerity so that they can eliminate the controversy about AI models eating up network traffic and having various negative effects. For example, after asking questions and giving answers, we should know why the AI trusts its own source, including returning corresponding links. ”

This is just the beginning, search engine startup You.com has begun to open related functions in its chat search engine.

Although it is more difficult to solve bias, Microsoft’s Bajwa said that it is not impossible to achieve with the joint efforts of all parties. "Supervision and supervision must exist. Technology can only play part of the role. In the end, efforts must be made in organizational structure, processes and governance. This is also where generative AI technology comes into effect. Enterprises must set corresponding policies for specific use methods. Parameters, suggestions and processes."

Only by reasonably solving these problems can enterprises truly use generative AI with confidence and boldness. Although the prospects are promising, they must not be left to chance. Although short-term applications can provide quick results, behind all the advantages there are hidden dangers. Considering that the technology itself is not yet mature, generative AI is best able to "grow healthily" under the company and care of humans.

The above is the detailed content of Is the hotly popular generative AI ready for enterprise applications?. For more information, please follow other related articles on the PHP Chinese website!