From Openai to Nvidia, researchers agree: Artificial intelligence agents have a long way to go

Only in the Gulf region, on Saturday, you spend getting rid of artificial intelligence agents – along with 2000 students, researchers, and familiar with the technology who were crammed at the University of California in Berkeley – such as the completely regular weekend plan. When I picked up my badge at the AI Agency summit that lasted all day and watched the snake line through the Union UNION, it looked like an academic conference and more like the Silicon Valley version of a lunch spot in New York.

This was definitely because of the headphone collection, which was stacked with senior researchers and scientists of artificial intelligence, including Jacob Pashki, chief scientist in Openai; Ed Chi, Vice President of Google DeepMind; Bill Dali, Senior Scientist in Nafidia; Ion Stoica, founder in Databricks & Anyscale, as well as Professor Uc Berkelegi; Dawn Song, a leading professor at the University of California in Berkeley, focuses on artificial intelligence security.

Popular may also be due to the tumultuous theme-AII agents, which is generally defined as a system that works itself that can complete the tasks, often independently, using other software tools. Not only did it suggest the course of the holidays, but also the journey reservation and the hotel reservation.

As my colleague Jeremy Kan said In a recent article“This type of automation is a dream of perennial C-SUITE fever. Over the past decade, companies have adopted“ automation of automatic operations ”or RPA. This was the program that could Automation of repeated tasksLike cutting and paste between database programs. But traditional RPA systems are unable to deal with exceptions, and they can usually deal with only one narrow task. “Agency AI is supposed to be more flexible and powerful, and adapts to work needs.

In January 2025 Blog post“We believe that in 2025, we may see the first artificial intelligence agents to join the workforce” and change the directing of companies financially. ”

But despite the noise, the comprehensive message at the AIGEC AI summit was cautious and divided: the agents may be the most trend in artificial intelligence at the present time, but technology still has a long way. Unfortunately, artificial intelligence agents cannot always be relied upon. They may not remember what happened before.

For example, Google DeepMind emphasized the gap between the factors you can do in coordinated experimental offers for what is still required in the real world production environments. Pachocki highlighted concerns about safety, security and the merit of agents systems, especially when they are combined into sensitive applications or work independently.

“I still don’t think the agents have really lived for their promise,” said Sherwin Woo, head of engineering at Openai API. “Some general cases have succeeded, but my daily work does not really feel different with agents.”

While today’s agents may not currently rise to the huge noise level (consider Salesforce CEO Mark Beniof The last claim The shift to digital employment means that “the last CEO of Salesforce who only managed to manage human beings”), the speakers of the AICENC AI still have a lot of optimism to participate. Stokea Databricks has expressed her enthusiasm about infrastructure improvements that facilitate the construction of the agent’s systems. DALY suggested from NVIDIA to provide continuous devices will enable the most powerful and effective agent’s behavior. Several “narrow victories” indicated in specific areas, such as coding.

Today, artificial intelligence agents may still have increased pain, but given the crowded UC Berkeleg Hall, the industry maintains a prize: artificial intelligence agents who can work reliably in the real world. They think the reward will be worth waiting.

However, here is more news of artificial intelligence.

Sharon Goldman
[email protected]
Sharongoldman

Amnesty International in News

The American Agency Agreement. Reuters mentioned Today, the Public Services Administration, the Central Procurement Arm for the United States Government, added Chatgpt from Openai, Gemi GOOGLE, and Claude Antarbur to a list of adopted artificial intelligence vendors to accelerate the use of technology by government agencies. The tools will be available to agencies through a platform with the terms of the contract in place. GSA said that accredited artificial intelligence “are committed to responsible use and compliance with federal standards.”

A artificial intelligence spending can be real consequences for the American economy. According to Washington PostThe standard Big Tech Intelligence in AI-more than $ 350 billion this year from Google, Meta, Amazon and Microsoft-has become a major economic power, so that the broader American economy shows signs of slowdown. Although job growth cools, this tremendous artificial intelligence spending is stalked to build databases and lead demand for chips, servers and communication equipment – greatly enhances GDP growth by up to 0.7 % in 2025.

The Clay Intelligence Slievance Clay raises $ 100 million with a rate of $ 3.1 billion. the New York Times deal She stated that Clay, which helps sales representatives and marketers find new threads and convert them into customers, raised $ 100 million with a rate of $ 3.1 billion. The tour led Capitalg, an investment arm at Alphabet, the parent company of Google. Among the other participants include Mereitech Capital Partners and Sequoia Capital. About six months of initial funds raising $ 1.25 billion.

Eye on artificial intelligence research

The new Jenny “Genny 3” creates Google DeepMind in actual time. Google DeepMind has unveiled Genie 3, a powerful new system for Amnesty International that can generate rich interactive virtual worlds of simple text claims – which makes it possible to move in dynamic environments in actual time in 24 frameworks per second. But although it is tempting to jump immediately to the use of the model to experience the final games, it is actually the latest leap in the company’s long-term batch towards “global models”-or artificial intelligence systems that can learn how the world works and simulates realistic environments. This is a key to training advanced agents, and in the end the artificial general intelligence. Unlike previous video generators, Genie 3 allows users to move through environments that are created from artificial intelligence that remain visually consistent for several minutes-and even respond to orders such as “Make It Snow” or “Add a letter”. Currently, the DeepMind Genie 3 challenges a small group of researchers and creators while exploring the responsible publishing and risks.

Our wealth

North Korea’s IT infiltration has exploded by 220 % over the past 12 months, with the GEN AI weapon at each stage of the recruitment process Amanda Gerot

Artificial intelligence is doing work interviews now – but the candidates say they prefer to risk to stay unemployed instead of talking to another robot – By Emma Bourley

These plans show how China advances the United States in the race to run the future of artificial intelligence – Matt Himer and Nick Rap

You have a calendar

September 8-10: Fortune Brainstorm Tech, Park City, Utah. Apply to attendance here.

6-10 October: world Amnesty International The week, Amsterdam

October 21-22: Teddy San Francisco. Apply to attendance here.

2-7 December: NeuPIPs, San Diego

December 8-9: Fortune Brainstorm Ai San Francisso. Apply to attendance here.

Brain food

Could the “depth of thought” be a key to the logic of artificial intelligence?

The new artificial intelligence model is the challenge of what we know about how to learn models to the mind: researchers from the SAPIENT intelligence in Singapore recently released Hierarch thinking model (HRM), which is inspired by the process of thinking about the layers in the brain – and the results have the gossip of the artificial intelligence community. Although it is 100 times smaller than ChatGPT and training on only 1000 example (with no internet data or step -by -step guidelines), human resources management solves difficult logical problems such as Sudoku, Maze Maze, and abstract thinking tasks roaming in much larger models. Instead of simulating human language, the causes of human resources management internally – work directly through problems in hidden rings, such as the person who thinks about a mystery in his head. Its success hints to a fundamental shift in artificial intelligence: one where the depth of thought may be more than size.