OpenAI pulls wrappers from GPT-4, without text in video

The large multimodal language model, GPT-4, is ready for prime time, although, contrary to reports circulating since Friday, it does not support the ability to output video from text.

GPT-4 can, however, accept both images and text input and produce text output. On a range of domains — including documents with text and photographs, diagrams, or screenshots — GPT-4 exhibits capabilities similar to text-only inputs, OpenAI explained on its website.

This feature, however, is in “search preview” and will not be available to the public.

OpenAI explained that GPT-4, while less capable than humans in many real-world scenarios, exhibits human-level performance on various professional and academic benchmarks.

For example, he passed a mock bar exam with a score around the top 10% of candidates. In contrast, the GPT-3.5 score was around the bottom 10%.

Jump to past models

One of the early adopters of GPT-4 is Casetext, creator of an AI legal assistant, CoCounsel, which it says is able to pass both the multiple-choice and written portions of the Uniform Bar Exam.

“GPT-4 exceeds the power of earlier language models,” Pablo Arredondo, Casetext’s co-founder and chief innovation officer, said in a statement. “The model’s ability not only to generate text, but also to interpret it, heralds nothing less than a new era in the practice of law.”

“Casetext’s CoCounsel is changing the way law is practiced by automating critical, time-consuming tasks and allowing our attorneys to focus on the most impactful aspects of the practice,” added Frank Ryan, President of Americas, DLA Piper. , a global law firm. Press release.

OpenAI explained that it spent six months aligning GPT-4 using lessons from its adversarial test program, along with ChatGPT, which yielded its best – though far from perfect – results on factuality, the maneuverability and the refusal to leave the safety barriers.

He added that the GPT-4 practice run was unprecedentedly stable. It was the company’s first large model whose drive performance was able to accurately predict in advance.

“As we continue to focus on reliable scaling,” he writes, “we aim to refine our methodology to help us predict and prepare future capabilities further and further in advance – something we consider essential for safety.”

Subtle distinctions

OpenAI noted that the distinction between GPT-3.5 and GPT-4 might be subtle. The difference appears when the complexity of the task reaches a sufficient threshold, he explained. GPT-4 is more reliable and creative and can handle more nuanced instructions than GPT-3.5.

GPT-4 can also be customized more than its predecessor. Rather than the classic ChatGPT persona with fixed verbosity, tone, and style, OpenAI explained, developers — and soon ChatGPT users — can now prescribe their AI’s style and task by describing these instructions in the message. “system”. System messages allow API users to significantly customize their users’ experience within certain limits.

API users will have to wait to try this feature first, as their access to GPT-4 will be limited by a waiting list.

OpenAI has recognized that despite its capabilities, GPT-4 has similar limitations to previous GPT models. More importantly, it is still not completely reliable. He “hallucinates” the facts and makes errors of reasoning.

Great care should be taken when using language model outputs, especially in high-stakes contexts, OpenAI warned.

GPT-4 can also be confidently wrong in its predictions, not caring to double-check work when it is likely to be wrong, he added.

Absence of T2V

Anticipation for the new version of GPT was fueled over the weekend after a Microsoft executive in Germany suggested a text-to-video capability would be part of the final package.

“We will be presenting GPT-4 next week, where we have multimodal models that will offer completely different possibilities – for example, videos,” said Andreas Braun, chief technology officer of Microsoft in Germany, during a conference of press Friday.

Video text would be very disruptive, observed Rob Enderle, president and principal analyst of Enderle Group, a consulting services firm in Bend, Ore.

“It could dramatically change the way movies and TV shows are created, the way news programs are formatted by providing a very granular user customization mechanism,” he told TechNewsWorld. .

Enderle noted that an early use of the technology could be creating storyboards from draft scripts. “As this technology matures, it will evolve into something closer to a finished product.”

Video proliferation

The content created by text-to-video apps is still basic, noted Greg Sterling, co-founder of Near Media, a news, commentary and analysis website.

“But text in video has the potential to be disruptive in the sense that we’ll see a lot more video content being generated at very low or near zero cost,” he told TechNewsWorld.

“The quality and effectiveness of that video is another matter,” he continued. “But I suspect some of them will be decent.”

He added that explainers and basic practical information are good candidates for video text.

“I can imagine some agencies will use it to create videos that SMBs can use on their sites or on YouTube for ranking purposes,” he said.

“It won’t be good — at least initially — for any branded content,” he continued. “Social media content is another use case. You’ll see creators on YouTube using it to boost volume to drive views and ad revenue. »

Not fooled by deepfakes

As discovered with ChatGPT, there are potential dangers to technology like text in video.

“The most dangerous use cases, like all tools like this, are garden variety scams impersonating people from relatives or attacks on particularly vulnerable people or institutions,” observed Will Duffield, political analyst at the Cato Institute, a Washington, DC think tank. .

Duffield, however, dismissed the idea of ​​using text in video to produce effective “deepfakes”.

“Where we’ve seen well-funded attacks, like the Russian deepfake of Zelenskyy’s surrender last year, they’ve failed because there’s enough context and expectation in the world to refute the fake,” he said. he explained.

“We have very well-defined notions of who public figures are, what they talk about, what we can expect from them,” he continued. “So when we see media about them behaving in an aberrant way, which doesn’t match those expectations, we’re likely to be very critical or skeptical of them.”