Bing Image Creator generates

AI Detector researchers say the model demonstrated 99% accuracy in distinguishing between human writing and ChatGPT-generated text, and significantly outperformed the GPT-2 Output Detector, an AI detection software provided by ChatGPT maker OpenAI.

The researchers also admitted that the sample data for testing the model algorithm of the competition at this stage is not large enough. They only tested articles from 13 chemical journals and 3 media. Through 20 text features and XGBoost, it can effectively distinguish between human collaboration and GPT- 3.5/GPT-4 generated content.

The researchers say these tests show their method is simple yet effective. They highlighted that the tool demonstrated 98%-100% accuracy in identifying AI-generated text, depending on the prompt and model.

Researchers say OpenAI’s current detector accuracy currently ranges from 10-56%, and the tool is more advanced than OpenAI’s.

The researchers further revealed that the goal behind this research and invention is to provide a way for the scientific community to assess “ChatGPT penetration into chemical journals, determine the consequences of its use, and quickly introduce mitigation strategies if issues arise” .

THIS IS A SPONSOR PROMOTION: >>>>>>>>>>>>>

Geekwills is an online shop that connects consumers with millions of products and brands around the world with the mission to empower them to live their best lives. Geekwills is committed to offering the most affordable quality products to enable consumers and sellers to fulfill their dreams in an inclusive environment.

The post With 99% accuracy, new model recognizes text content generated by ChatGPT appeared first on TechGoing.

[Introduction to New Intelligence] Some time ago, OpenAI CEO Sam Altman and CTO Mira Murati discussed AGI, the future development of GPT, and the impact of AI on humanity in an exclusive interview with WSJ.

“Why is OpenAI’s ultimate goal AGI? What is AGI?”

“What are ChatGPT and other language models used for?”

“How will the relationship between humans and artificial intelligence change in the future?”

At the 2023 Wall Street Journal (WSJ) technology press conference, OpenAI CEO Sam Altman and CTO Mira Murati discussed artificial general intelligence (AGI), the development of future GPT models, and the impact of artificial intelligence on humans Impact.

Nine years ago, Sam Altman said at a Wall Street Journal conference that artificial intelligence causing human unemployment was something that would not happen for a long time and that we did not need to worry too much about.

Less than a decade later, Altman and the company he co-founded, OpenAI, released an artificially intelligent chatbot called ChatGPT.

It can write emails, business plans, and even code that would have been unimaginable nine years ago.

When discussing the current impact and impact of artificial intelligence on human society, Altman remains optimistic, but more cautious than nine years ago.

OpenAI’s ultimate goal: AGI

AGI, this concept has been endowed with infinite beautiful imagination by people since its birth.

The same goes for Altman, who believes that AGI will be the most outstanding creation ever created by mankind.

Armed with this great tool, humans are able to solve the world’s problems today and create unimaginable new things for ourselves, each other, and the world.

At that time, humans will have more creative means of self-expression. But Altman is certain that these changes will bring huge benefits to humanity.

“In nine years, when the Wall Street Journal invites me, you may ask: Why do we think humans don’t want AGI to arrive?”

So when will AGI appear? And how can people tell when AGI is coming?

Altman defines AGI as something we don’t have yet. Ten years ago, people might have thought that something like GPT-4 or GPT-5 was AGI.

But now, GPT-4 is just seen as a pretty good “little chatbot”.

People’s benchmark threshold requirements for AGI are getting higher and higher, which requires people to put more and more efforts into artificial intelligence.

Altman said, “Humans are now close enough to the threshold of AGI that improving the ability of AI becomes less important. The problem we currently face is how to define AGI.”
GPT-5: Solve “illusion” and data copyright issues

OpenAI has released multiple versions of GPT since its inception, each version more powerful than the last.

In March this year, OpenAI released its latest model-GPT-4. But people are still full of expectations about OpenAI’s next model: Is GPT-5 under development?

Faced with this problem, Mira Murati, CTO of OpenAI, responded, “We are not there yet.”

But she also said that OpenAI is always working on the next thing, like reducing model hallucinations: A future release of GPT-5 will work on solving the hallucination problem that now plagues models.

Mira said that although GPT-4 has made great progress on the hallucination problem, it is still some way away from completely solving the hallucination problem.

But OpenAI has been on the right track to solve the problem: reinforcement learning with human feedback (RLHF), allowing the model to output truly reliable content.

In addition, OpenAI also reduces the problem of model illusion by integrating multiple technologies, such as adding inspection and search capabilities to the model, and providing more factual data to the model to ensure that users can obtain more factual output from the model.

But for OpenAI, copyright has always been a controversy.

Not only the data used to train the model, but also the content generated by the model often involves copyright protection issues.

Many publishers and writers have protested against OpenAI’s infringement.

Altman discussed data usage and data ownership issues from another perspective.

In the future, OpenAI’s new model will become infrastructure that everyone can use, which means the way you think about data ownership and economic flows will change.

Now, OpenAI is working to establish partnerships with different data rights holders, but as models become more intelligent and capable, less and less data will be needed to train models in the future.

But current models still require as much data as possible produced by humans when training. However, Altman said that this will not be the long-term future development path of the model, because what really matters in the future is valuable data.

As OpenAI advances technologically, the conversation around data and ownership will shift.
The future of humans and AI

The first is the relationship between humans and AI.

On September 25, OpenAI added more personalization features to GPT-4. ChatGPT can now see, listen, and speak.

The newly added voice function of GPT-4 is quite user-friendly, and communication with people is very natural.

It is foreseeable that ubiquitous AI will soon become a reality of human life.

In the future, we cannot avoid interacting with artificial intelligence, and this brings up a question: how humans should treat their relationship with artificial intelligence.

And OpenAI and other companies that train the models have some control over the artificial intelligence that forms relationships with people.

This will be an unsettling future in which these artificial intelligences may become friends and even companions of humans.

But Altman made it clear that he doesn’t want people to develop a close relationship with artificial intelligence that goes beyond human friends: Artificial intelligence is different from humans. Maybe these systems are full of personality, but this has nothing to do with human nature.

Therefore, when communicating with artificial intelligence, you need to be different from communicating with humans.

“The reason why we named the model ChatGPT instead of a person’s name is to make it clear to users that they are communicating with an artificial intelligence, not a real human being.” Altman emphasized this point.

But just as people have many different relationships, people will also develop special relationships with artificial intelligence. But eventually, people will realize that artificial intelligence is different from humans, but our relationship with it will not be broken because of it.

On the other hand, the rapid development of AI also makes people worry about the uncontrollable risks it brings: the use of these systems to commit crimes and the impact on the job market.

Such as commanding AI to invade computer systems or design biological and chemical weapons.

This is not an unreachable future. Since the outbreak of generative AI, the use of AI to conduct fraud and cyberattacks has become commonplace.

But Altman believes that these negative impacts are inevitable as technology develops.

What we need to solve is the risks brought by technology, rather than giving up on development. The latter remains a moral failure for humanity.

In human history, almost every technological revolution will profoundly affect the job market, either completely subverting it or half of the jobs will disappear.

But in fact, when old jobs disappear, new ones are created. This is a testament to human progress, the question is how quickly our society adapts to change.

Within two or at most three generations, humans can adapt to almost any degree of job market change.

There may be people who are unwilling or unwilling to change their jobs, but the nature of their jobs will change.

For a primitive hunter-gatherer tribe, typing in front of a computer couldn’t be a real job.

“Work is just humans trying to entertain themselves with some stupid status game,” Altman said.

The real challenge is dealing with the transition to job market innovation.

Society needs to take action to ensure that people are not harmed in this transition. It is not enough to provide a universal basic income. People need to have the initiative and influence to participate in building the future.

This is why OpenAI is so determined to promote ChatGPT.

Although not everyone can use artificial intelligence technology, as more and more people participate, people will have the opportunity to think about and chart the direction of future development.

This is the most important thing we should pay attention to.

THIS IS A SPONSOR PROMOTION: >>>>>>>>>>>>>

Geekwills is an online shop that connects consumers with millions of products and brands around the world with the mission to empower them to live their best lives. Geekwills is committed to offering the most affordable quality products to enable consumers and sellers to fulfill their dreams in an inclusive environment.

The post Sam Altman: OpenAI Training GPT-5 Not Bad for Money, Humans Close to AGI Threshold appeared first on TechGoing.

OpenAI stated that the Assistants API provides three functions: “Code interpreter”, “Retrieval” and “Function calling” to help users handle “heavy work”:

     Code Interpreter allows developers to write and run Python code in a sandbox execution environment, generate graphical charts, and process files with different data and formats, and allows AI assistants to iteratively run code to solve challenging code and Math problems etc.

     Retrieval enhances the assistant with knowledge outside the model, such as proprietary domain data, product information, or user-provided documentation.

     Function Call enables the assistant to call a developer-defined function and incorporate the function response into its message.

OpenAI says the Assistants API is designed for flexibility, with use cases ranging from “natural language-based data analysis applications,” “code assistants,” “AI-based vacation planners,” “DJs,” “smart canvases,” and more.

It is reported that a key update introduced by the Assistants API is to provide persistent and infinitely long threads (Threading), allowing developers to hand over thread state management to OpenAI and solve the problem of context window length constraints.

OpenAI developer experience director Romain demonstrated the Assistants API live. He claimed that developers only need to enter the name and introduction of the chat assistant, select the model to be used, and then select the required tool to automatically generate the relevant AI assistant.

OpenAI says that, like other parts of the platform, data and files uploaded to the OpenAI API are never used to train its models, and developers can delete the data as they see fit.

It is found that OpenAI also claimed that the Assistants API will be open to beta today, and users can experience it on the Assistants Playground homepage without writing any code.

THIS IS A SPONSOR PROMOTION: >>>>>>>>>>>>>

Geekwills is an online shop that connects consumers with millions of products and brands around the world with the mission to empower them to live their best lives. Geekwills is committed to offering the most affordable quality products to enable consumers and sellers to fulfill their dreams in an inclusive environment.

The post OpenAI Launched Assistants API, Allowing Developers to Customize AI Assistants with One Click appeared first on TechGoing.

Satya Nadella did not attend for long and mainly talked about the rise of artificial intelligence and its impact on the company’s data infrastructure such as Azure.

Satya Nadella said in the interview:

 Workloads, workload patterns, and training operations need to be advanced simultaneously, and the amount of data and parallel processing are more complex than before.

 So the first thing we've been doing is collaborating to build the underlying systems, from power to data centers to racks to accelerators to networking, from the way we think about it to actually running it, the way Azure operates has happened A sea change.

 To support these models that OpenAI is enhancing, the company is changing rapidly. So our first job is to build the best system so we can build the best models and then make it all available to developers.

In his keynote speech, Sam Altman said:

 I think we have the best partnership in technology. It's clear that both OpenAI and Microsoft are happy with their collaboration, and the relationship will be even stronger for the foreseeable future, especially as more developers sign up for Azure to use OpenAI's tools and services.

During DevDay, OpenAI announced a plan to allow developers to create their own ChatGPT services without coding. It also launched a preview of its GPT-4 Turbo model, which has “knowledge of world events as of April 2023” and is cheaper to use than ChatGPT 4.

THIS IS A SPONSOR PROMOTION: >>>>>>>>>>>>>

Geekwills is an online shop that connects consumers with millions of products and brands around the world with the mission to empower them to live their best lives. Geekwills is committed to offering the most affordable quality products to enable consumers and sellers to fulfill their dreams in an inclusive environment.

The post Satya Nadella attended OpenAI’s first developer conference appeared first on TechGoing.

▲Image source OpenAI

OpenAI said it launched these features to give users “some control” over ChatGPT. For example, businesses can create a special assistant that only employees can access, and parents can create a solution designed to teach their children how to solve tough math problems. question tool.

Currently need to subscribe to ChatGPT Plus or ChatGPT Enterprise for $20 per month (currently about RMB 146) to use the “GPTs” service. After that, users can go to their personal account and select the “Create GPT” option at the top of the page, and then the GPT Builder tool built into the website will help users set up their own GPT model step by step.

▲Image source OpenAI

It is reported that at the functional level, the names and avatars of “GPTs” can be customized. Users can also upload their own configuration files to further configure the “personality” of the model. Users can also enable additional functions, such as allowing “GPTs” to network Or integrate the DALL-E Vincent diagram tool.

▲Image source OpenAI

▲Image source OpenAI

OpenAI claims that they will launch a “GPT store” in late November that will showcase “verified user creations.” People can search the store for a variety of chatbots created by others in a variety of fields, including productivity and education, and users who publish their “own GPT” will receive some financial rewards.

In addition, OpenAI also said that they will launch a preview version of GPT-4 Turbo, which is built based on data from April this year. This model can “better understand current events”, and for developers, Turbo Versions will also be cheaper to use.

THIS IS A SPONSOR PROMOTION: >>>>>>>>>>>>>

Geekwills is an online shop that connects consumers with millions of products and brands around the world with the mission to empower them to live their best lives. Geekwills is committed to offering the most affordable quality products to enable consumers and sellers to fulfill their dreams in an inclusive environment.

The post OpenAI launches user-customizable GPTs service, a subscription is required appeared first on TechGoing.

It is reported that GPT-4 Turbo supports 128k context, the input price is three times cheaper than GPT-4, the rate limit is doubled, the knowledge base is updated to April 2023, JSON Mode is also added, and multiple function calling capabilities are updated. Compared with previous GPT-4, it has stronger performance, cheaper price and lower development cost.

According to OpenAI, compared to the previous GPT-4’s 8k context length, GPT-4 Turbo has a 128k context length. Calculated based on an article of about 1k text, GPT-4 Turbo can process a total of 128 articles at the same time.

In addition, OpenAI has also launched a new feature called Json Mode, which ensures that developers can more easily make models call Json and XML content, making it easier for models to return consistent output results, thereby achieving repeatable output, which is useful for controlling model behaviour and writing Model unit test content plays a role.

GPT-4 Turbo also supports image input and integrates DALL・E 3, which can automatically summarize the content of images and documents with images. OpenAI also plans to provide visual support for the main GPT-4 Turbo model.

In terms of price, developers can specify models through the OpenAI image API, and the price starts at $0.04 (currently about RMB 0.292) per generated image.

The input token price of GPT-4 Turbo is 1/3 of the previous GPT-4, 1k tokens only costs 0.01 US dollars (currently about RMB 0.073), and the output tokens price is 1k tokens 0.03 US dollars (currently about RMB 0.219 ).

Sam Altman said that uploading an image with a resolution of 1080×1080 pixels to GPT-4 Turbo will cost $0.00765 (currently about RMB 0.056).

In addition, OpenAI also announced its next-generation open-source speech recognition model. This model provides six preset sounds and two model variants. Users can generate more natural speech files through the text-to-speech API, and every 1,000 characters are input. The starting price is $0.015 (currently about RMB 0.1095).

OpenAI CEO Sam Altman claimed that GPT-4 Turbo is now available for all paying developers to try out by passing gpt-4-1106-previewAPI, and the relevant stable version is expected to be released in the next few weeks.

THIS IS A SPONSOR PROMOTION: >>>>>>>>>>>>>

Geekwills is an online shop that connects consumers with millions of products and brands around the world with the mission to empower them to live their best lives. Geekwills is committed to offering the most affordable quality products to enable consumers and sellers to fulfill their dreams in an inclusive environment.

The post OpenAI launches GPT-4 Turbo, cheaper than previous generation appeared first on TechGoing.

Pexels

Since releasing the ChatGPT and Whisper models through an API (application programming interface) in March of this year, the company now has more than 2 million developers, including 92% of Fortune 500 companies.

At the developer conference, in addition to announcing these numbers, OpenAI detailed a number of new features, including a platform that can build customized versions of ChatGPT to help with specific tasks. The company also unveiled a new model, GPT-4 Turbo, with a knowledge base updated through April 2023 that can contain the equivalent of more than 300 pages of text in a single prompt.

After launching nearly a year ago, ChatGPT is widely regarded as the fastest-growing consumer Internet application ever, reaching about 100 million monthly active users in just two months. By comparison, it took Facebook about four and a half years to reach 100 million users after launching in 2004. x (formerly Twitter) took more than five years, and Instagram took more than two years.

Microsoft said its Bing search engine, which earlier this year added generative AI capabilities powered by OpenAI GPT-4, passed the 100 million daily active users milestone in March of this year, more than a decade after its 2009 launch.

However, ChatGPT’s record has already been broken by Meta’s Threads, which amassed 100 million users less than a week after launching in July. But Threads usage seems to have declined in the months since then. As of October, there were fewer than 100 million monthly active users.

No matter which way you analyze these numbers, it’s clear that ChatGPT is still very popular, even though it’s been less than a year since it was launched to the public.

This isn’t the first time we’ve seen OpenAI’s chatbot with a large number of users. Back in February, Similarweb estimated that the tool had reached a milestone of attracting 100 million unique visitors in a month and 25 million visitors per day. It’s worth noting, however, that Monday’s figures came from OpenAI officials, not third-party estimates.

OpenAI seems to be releasing these figures to refute recent media reports that ChatGPT’s popularity has begun to decline since its release last November.

THIS IS A SPONSOR PROMOTION: >>>>>>>>>>>>>

Geekwills is an online shop that connects consumers with millions of products and brands around the world with the mission to empower them to live their best lives. Geekwills is committed to offering the most affordable quality products to enable consumers and sellers to fulfill their dreams in an inclusive environment.

The post OpenAI: GPT-4 has 100 million weekly active users, most powerful AI model appeared first on TechGoing.

OpenAI said, “We are convinced that the capabilities of cutting-edge AI models will exceed the current state-of-the-art existing models and may benefit all mankind, but they also bring increasingly serious risks.” The team will work to mitigate nuclear threats, biological, chemical and radioactive threats and other major risks that may be caused by AI, as well as potential risks brought about by behaviours such as AI “self-replication”.

Image source Pexels

OpenAI says companies need to ensure they have the understanding and infrastructure needed to secure highly capable AI systems. The team will be led by Aleksander Madry, formerly director of MIT’s Center for Deployable Machine Learning, and will develop and maintain a “risk-informed development policy” outlining the company’s efforts to evaluate and monitor large AI models.

The non-profit AI Security Center issued a brief statement at the end of last month: “Reducing the risk of extinction caused by artificial intelligence will be placed at the same level as social issues such as epidemics and nuclear war, and promoted as a global priority.” , this 22-word initiative has been supported by many AI industry leaders, including OpenAI CEO Sam Altman.

THIS IS A SPONSOR PROMOTION: >>>>>>>>>>>>>

Geekwills is an online shop that connects consumers with millions of products and brands around the world with the mission to empower them to live their best lives. Geekwills is committed to offering the most affordable quality products to enable consumers and sellers to fulfill their dreams in an inclusive environment.

The post OpenAI forming new team to guard against catastrophic risks related to AI appeared first on TechGoing.

In a statement, Microsoft said, “We are proud of the impact this incubation program has had for our customers, and we will continue to invest in Azure to provide a computing platform for virtual worlds in industry as well as for a variety of AI projects within our company.” “We are working closely with our customers to realize this transformation.”

Prior to this, Microsoft officially ceased support on October 19 for Project Bonsai, an AI development platform for building automated systems for industrial use. Both programs are considered part of Microsoft’s “industrial meta-universe.”

Microsoft acquired the AI startup Bonsai in 2018, which was seen internally as Microsoft’s response to Google’s acquisition of Deepmind, people familiar with the matter said. Project Airsim was initially launched as an open-source project in 2017 but has since shifted its focus to products for industrial customers.

Both Project Airsim and Project Bonsai were driven by Kevin Scott, Microsoft’s chief technology officer. He brokered Microsoft’s partnership with OpenAI, which incubated the projects to enable industrial customers to use new products from Microsoft’s cloud business.

Nadella has talked about Project Bonsai the same way he talks about OpenAI today, mentioning it as part of Microsoft’s AI future in staff meetings and public interviews, according to a person familiar with the project.

The person familiar with the matter said that Microsoft initially saw the projects as a means to attract application developers in the industrial space and help Microsoft’s Azure cloud be able to compete with Amazon Web Services AWS. However as Microsoft’s partnership with OpenAI has grown, Scott has become less interested in the programs.

In early 2023, around the time Microsoft announced its expanded partnership with OpenAI, the company also began touting its vision for an industrial metaverse. But the good times didn’t last long for the projects in question, as Microsoft terminated Project Bonsai and laid off the 100-person team in charge of the project this spring, just a few months after it had assembled the team.

The person familiar with the matter said that Microsoft retained Project Airsim because it believed the incubator had a large number of potential customers.

Gurdeep Pall, a former Microsoft vice president who was head of product incubation and business AI, responsible for Project Bonsai and, most recently, Project Airsim, left the company last month after 33 years of service.

The termination of Project Airsim is another example of Microsoft shifting resources toward OpenAI. Last month it was reported that Microsoft was abandoning experimental products like the Surface headset to focus on investing in AI.

THIS IS A SPONSOR PROMOTION: >>>>>>>>>>>>>

Geekwills is an online shop that connects consumers with millions of products and brands around the world with the mission to empower them to live their best lives. Geekwills is committed to offering the most affordable quality products to enable consumers and sellers to fulfill their dreams in an inclusive environment.

The post Microsoft axes industrial metaverse project Airsim shifts AI strategy to OpenAI appeared first on TechGoing.

At the end of September, OpenAI announced that its chatbot ChatGPT has added web browsing capabilities, can access the Internet in real-time, and provide website links to the source of information. At that time, the feature was currently only available to ChatGPT Plus and Enterprise subscribers and was in a beta phase.

OpenAI announced via its release notes on Tuesday that the web browsing feature has moved from beta to official, making it more convenient for subscribers to use it without having to toggle the beta switch.

According to the instructions, now subscribers only need to select “Browse the web with Bing” from the GPT-4 model selector.

ChatGPT accesses the Internet through Microsoft’s Bing web browser. The web browsing feature, powered by Bing, was initially announced at Microsoft’s Build conference in May and is rolling out to ChatGPT Plus users. However, a few months later, OpenAI pulled the feature after discovering that users were using it to bypass paywalls.

In a September announcement, OpenAI said the company had received helpful feedback from users since its initial rollout in May and incorporated it into new updates, including “Follow robots.txt” and “Identify user agents so websites can You can control how ChatGPT interacts with them.”

ChatGPT Plus membership costs US$20 per month. For users who need the latest information, this new feature may be worth paying because it can improve the accuracy and reliability of the chatbot.

For other users eager to take advantage of this feature, OpenAI said in the original announcement that the feature will be rolled out to all users soon. Until then, Bing Chat is a good alternative as it is also powered by GPT-4, has internet access, and is free.

THIS IS A SPONSOR PROMOTION: >>>>>>>>>>>>>

Geekwills is an online shop that connects consumers with millions of products and brands around the world with the mission to empower them to live their best lives. Geekwills is committed to offering the most affordable quality products to enable consumers and sellers to fulfill their dreams in an inclusive environment.

The post ChatGPT networking function officially launched, available to Plus members appeared first on TechGoing.