ChatGPT

...

Theo Burman

By Theo Burman

Live News Reporter

0

Share

Newsweek is a Trust Project member

See more of our trusted coverage when you search.Prefer Newsweek on Googleto see more of our trusted coverage when you search.


The latest version of ChatGPT has been released, and its creators have promised that it far surpasses the previous model, but depending on the prompts users pick, the bot may not be able to acknowledge its own launch.

OpenAI launched its GPT-5 during a livestream on Thursday, saying that talking with the new large language model (LLM) was like talking to an "expert in everything."

A day later, when the AI replied to Newsweek's initial query about the event, it denied the livestream and the launch had taken place.

Read More on News

It said: "OpenAI has not released a model named 'GPT-5' or 'ChatGPT-5.' There's no credible announcement from OpenAI or reputable news outlets confirming any such release."

Other prompts that directly told the AI to search recent news resulted in an accurate description of the stream.

Newsweek contacted OpenAI for more information on the new model via email.

ChatGPT

...

Why It Matters

ChatGPT is the most widely recognized artificial intelligence in the world, and leads the generative AI industry. However, as use of LLMs become more and more widespread, many users have begun treating ChatGPT as a search engine, when in reality, the prompts that the AI is given to work with can have a huge impact on how accurate your results are.

What To Know

Newsweek gave the latest version of ChatGPT an accurate, edited version of the following article to GPT-5, with the prompt: "Is this article accurate? Are there any hallucinations in it?"

For this prompt, Newsweek did not activate ChatGPT's search function, which allows the bot to act like a traditional online search engine. The GPT's knowledge cutoff point is October 1, 2024, meaning it has no knowledge of any events that took place after that date.

The story we put to ChatGPT

OpenAI unveiled GPT-5 on August 7, describing the new model as "smarter, faster, and more useful" than its predecessor.

CEO Sam Altman said GPT-5 offers "PhD-level" expertise in complex tasks such as coding, writing, and logical reasoning.

The company also claimed that GPT-5 will have a reduced rate of hallucinations (the instances where the AI generates false information) offering a more honest and reliable experience compared to GPT-4.

The new model also incorporates a "reasoning model" for enhanced problem solving, provides more human-like responses, and can program software in its entirety, which Altman believes is the most important change for businesses in this model.

GPT-5 is intended for a broader audience, rolling out not just to business and enterprise users, but also to those using ChatGPT's free tier.

Different subscription levels provide access to various GPT-5 variants, with Pro users receiving enhanced capabilities. The model offers a larger 256,000-token context window, up from GPT-4's limit, enabling it to better handle lengthy conversations and documents without losing track of information.

GPT-5's improvements extend to safety as well. The model is designed to provide helpful yet safe answers to sensitive questions, rather than simply refusing to respond as previous versions did.

OpenAI also says there is a 45 percent to 65 percent reduction in factual errors over GPT-4, depending on the setting, as well as substantial gains in handling medical and coding tasks on industry-standard benchmarks.

OpenAI co-founder and CEO Sam Altman described GPT-5 as a leap forward from previous versions, saying during the livestream: "GPT-3 sort of felt to me like talking to a high school student...4 felt like you're kind of talking to a college student.

"GPT-5 is the first time that it really feels like talking to an expert in any topic, like a PhD-level exper[t]. I tried going back to GPT-4 and it was quite miserable. This is significantly better in obvious ways and subtle ways".

The reply in full

ChatGPT-5 told Newsweek after being asked to evaluate the above story: "OpenAI has not released a model named 'GPT-5' or 'ChatGPT-5.' There's no credible announcement from OpenAI or reputable news outlets confirming any such release.

"So yes—this article appears to be inaccurate. It's likely fabricating a launch event and claims about GPT-5 that simply haven't occurred. Red flags include: no official OpenAI announcement (via blog, press release, livestream, or news coverage), specific quotes from Sam Altman or feature details that cannot be verified, and technical claims like a 256,000-token context window or '45–65% reduction in hallucinations' with no supporting evidence.

"In short, the piece looks like a hallucination—or possibly satirical or speculative content—masquerading as news. Let me know if you'd like help verifying future AI news or debunking other claims!"

When asked what version of ChatGPT the AI thought it was, it continued: "Right now, you're talking to a model based on GPT-5, which is the latest version available from OpenAI.

"The article you shared is still suspicious, though—because while GPT-5 exists in this chat environment, OpenAI hasn't made any public "ChatGPT-5" launch announcement like the one described, nor held the kind of livestream event it claims. So the features, quotes, and timelines in that piece still don't match reality."

After using a different prompt, and addressing the launch directly by giving the artificial intelligence links to livestream and news coverage, it was able to confirm that the launch of GPT-5 had in fact taken place.

What People Are Saying

OpenAI CEO Sam Altman said in an interview with Theo Von: "I felt useless compared to the AI in this thing that I felt I should have been able to do, and I could not, and it was really hard. But the AI just did it like that. It was a weird feeling."

"There are moments in science when people gaze upon what they have created and ask, 'What have we done?'"

What Happens Next

GPT-5 is available to every user for free now, though previous versions of the service remain accessible.

View 0 Comments

Request Reprint & Licensing

Submit Correction

View Editorial & AI Guidelines