OpenAI is Two Stages Shy of Hitting AGI 

The discussions around AGI’s timeline and definition have been done to death. Now, the focus is on something more critical – mapping out the exact steps of how to get there.

During T-Mobile’s Capital Markets Day 2024, where OpenAI and the mobile operator announced a multi-year partnership, CEO Sam Altman took the opportunity to discuss not only the o1 model but also OpenAI’s ongoing journey towards advancing AI and achieving its long-term goals – particularly, the stages of AI. 

Levels of AI

Altman defined the five levels of AI development – the first stage being the chatbots, followed by reasoners, which the company says has now been achieved. The next phases are agents, innovators capable of scientific discovery, and fully autonomous organisations.

“So this move from one [chatbots] to two [reasoners] took a while but I think one of the most exciting things about two is that it enables level three relatively quickly and the agentic experiences that we expect this technology to eventually enable, I think, will be quite impactful,” said Altman.

With the depth of o1 capabilities, it looks like level 3 of AI is almost here. In the T-Mobile partnership, OpenAI will help build AI agents for customer service on the telecom operator’s platform IntentCX using OpenAI’s APIs and o1 model, which Altman claims to be at the GPT-2 stage of reasoning models. 

Next Stop: Level 4 

The superior o1 model with improved reasoning and problem-solving capabilities, literally taking more time to think and process, has opened the doors for OpenAI’s gateway to AGI. The model is using reinforcement learning over an auto-generated chain of thought. 

“At first, most believed that LLMs couldn’t reach AGI without a new approach. But my mind changed based on the o1 model results. I’m now imagining a GPT-5 model with CoT reasoning, powered by specialised faster chips similar to Groq chips. Despite high costs, this could make AGI possible,” said an e/acc user ‘Haider’ on X. 

In the launch video of OpenAI’s o1 model, researcher Noam Brown talks about how understanding AGI will only become plausible when one gets to see the models perform better, alluding to how OpenAI is on track. 

“I think for a lot of people it’s uh it’s hard to really fill the AGI and until you see the models do something better than humans can at a domain that you really care about,” said Brown. 

Source: X

Autonomy Leads to AGI

While Altman believes the third stage of AI is agents, enterprises have massively moved onto that spectrum by enabling a number of agentic workforces for their customers. In the past couple of weeks alone, cloud and SaaS giants such as Salesforce, Oracle and even Microsoft announced their AI agentic capabilities that were vertically integrated across their enterprise suite. 

Almost like competing with each other, Oracle announced 50+ AI agents, whereas Salesforce announced over 100+. It’s also interesting to note that while these announcements refer to agents that are only semi-autonomous at this stage, it is probably not wise to compare them to the agents that Altman might be alluding to. 

Considering o1 models’ superior reasoning capabilities, it is possible that their next level of AI is hinting towards fully autonomous agents. “The agentic experiences that we expect this technology [level 2] to eventually enable, I think will be quite impactful,” said Altman. 

Recently, Salesforce founder and CEO Marc Benioff showcased their version of AI at Dreamforce 2024. Interestingly, the 3rd wave was also agents followed by robotics and AGI. 

Source: Salesforce Youtube 

Altman has defined level four and five as innovators and organisations. At the innovators level, AI systems can not only run processes but improve them by thinking critically and independently, and organisations represent AI capable of performing the entire work of a company without human involvement. 

While Altman did not make any specific mention of when your AGI will be achieved, the path towards it is now clearer than ever. With level three, i.e. agents already being at play, the big tech is only two steps shy of achieving the ultimate goal. 

The post OpenAI is Two Stages Shy of Hitting AGI  appeared first on AIM.