news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

"science and technology innovation board daily" september 18th (editor zhu ling)today, openai ceo sam altman said on the social platform x that although it took some time, openai performed very well in achieving the third technical goal.

he also shared an article published on june 20, 2016 titled "openai technical goals".

it was mentioned that openai’s third technical goal is to develop an agent that can perform complex tasks based on instructions in human natural language and can request explanations when tasks are not clearly understood.

"although there are some good algorithms that can complete language tasks such as question answering, grammatical parsing, and machine translation under human supervision, no algorithm can achieve more advanced language goals, such as the ability to have conversations with people, fully understand the content of documents, and complete tasks according to complex natural language instructions. openai hopes to solve these problems by developing new learning algorithms and models," the 2016 article reads.

it is worth noting that of the authors of this article, ilya sutskever, greg brockman, altman, and elon musk, only altman is still working at openai.

now, altman's announcement seems to mean that openai has made an important breakthrough in this field. however, he did not specify whether this breakthrough was achieved through the already released chatgpt, gpt4 or o1 (strawberry) model, or through a new project that has never been seen. this ambiguity has triggered widespread speculation and heated discussions among netizens.

some netizens believe that chatgpt can already use natural language to complete most of the work in the third technical goal. but some netizens also said that o1 is expensive and slow, and is not a practical intelligent entity at all.

it is important to note that in the 2016 article, the third technical goal emphasized intelligent agents, but none of the models currently released by openai emphasize intelligent agents.

according to a reuters report in july, openai has high hopes for the strawberry model’s capabilities, especially its ability to perform a range of complex tasks, such as browsing the web autonomously with the assistance of an agent that can take actions based on the findings of the large model. openai also plans to test its ability to assist software engineers and machine learning engineers in their work.

the us technology website the information reported in february this year that openai is stepping up the development of a new product that is almost as revolutionary as chatgpt - ai agent software, which is expected to become what altman calls a "super-intelligent working personal assistant."

one of the agents can automatically perform complex tasks by taking over human equipment, such as analyzing data in a document using a spreadsheet or automatically filling out an expense report and entering it into accounting software.

another agent can perform web-based tasks, such as gathering public data on a specific company, creating an itinerary based on a certain budget, or booking a flight.

shortly before the report was published, openai employee ben newhouse posted on x that he was recruiting for his team, "this may be an industry-defining zero-to-one product that leverages the latest and greatest in our upcoming models." openai vice president of product peter welinder also said that the product described by newhouse "will change everything."

it is worth mentioning that altman's announcement may also contain certain strategic considerations. by throwing out some vague information, people will speculate and hype, so as to raise the huge amount of funds they need for the company.