AI Gets Agents: ChatGPT Now Has Deep Research with Agentic Capabilities

[AleksandarK]

Today, OpenAI has announced a new ChatGPT feature called "Deep Research," which is capable of performing complex, multi-step research processes entirely on its own. Using so-called agents, which are autonomous bots working on top of the AI model, this feature searches the web and curates all needed information. This agentic behavior was trained on real-world browser usage, accompanied by Python code execution. Deep search, like OpenAI's o1 and o3 models, uses reinforcement learning, which steps back to "think" and creates a chain of thought before delivering users an answer to their question. Depending on the topic, deep research can take 5 to 30 minutes to search the web, crawl through data, and compile it in a reader-friendly manner.

Regarding benchmarks of its performance, OpenAI put out a lot of interesting comparisons and evaluations. Compared to all previous models, deep research gives these models additional context to help AI with more information. Thus, in evaluation benchmarks like Humanity's Last exam, deep research scored 26.6%, whereas o1 and o3-mini scored 9.1 and 13%, respectively. Other evaluations showed a modest improvement, while concrete comparisons were made in UX, business, and medical research. Turning the deep research feature on delivered more information every time, and you can see it for yourself here.




However, as with every Transformer-based AI model and technology, it is prone to hallucinations. Specifically, it can create false references, pick up on rumors and treat them as facts, and not distinguish confidently on the information. However, it is reportedly much better compared to an average AI model in ChatGPT. Interestingly, OpenAI expects this to get annulled with more usage as deep research advances and learns more about information processing on user prompts. This officially marks OpenAI's level three of AGI. Level one was chatbots, which we got with ChatGPT; level two was reasoning models, which was o1/o3; and level three was agents, who can now perform their own tasks. Level four is next: an AI model that can aid in human development and invention.

View at TechPowerUp Main Site | Source

 

[Harthad]

 

[tpa-pr]

Interesting. Since its bots crawling the web, will they be made to respect robots.txt or some similar standard?

 

[ZoneDymo]

and I take it AI will credit the sources used for this ermmm "research document" ?

 

[Assimilator]

Please can we not turn TPU into yet another LLM company PR regurgitation mouthpiece, kthx.

 

[AleksandarK]

Please can we not turn TPU into yet another LLM company PR regurgitation mouthpiece, kthx.

Technology progress is always reported, no mather if AI or crypto or anything else

 

[MacZ]

So you would expect that people start to realize that having accurate and truthful informations on the web is important.

And for example, in the news business, remember that their role is to inform but also to act as a bulkwark against the BS, twists, manipulation and outright lies, especially of powerful/rich persons and organizations, and especially of their own governement.

And this require a bit of ability to be skeptical, and a bit of courage (because they will be retaliated against)

Otherwise :

1/ If this is just cheerleading for companies and technologies (for example), you just need a model with 3 inputs : the press release, sentiment and style. It will do just as fine and nothing of value will be lost. And don't think about learning to code : it's too late.

2/ The AI superintelligence will have some other ways to go haywire than just hallucinations.