The week in AI: OpenAI now confident it knows how to build AGI, expects first AI agents to materially change the output of companies in 2025

In partnership with

Welcome to The Dispatch! We are the newsletter that keeps you informed about AI. Each Thursday, we aggregate the major developments in artificial intelligence - we pass along the news, useful resources, tools and services; we highlight the top research in the field as well as exciting developments in open source. Even if you aren’t a machine learning engineer, we’ll keep you in touch with the most important developments in AI.

NEWS & OPINION

OpenAI CEO Sam Altman says “we are now confident we know how to build AGI”

-------------------------

OpenAI CEO Sam Altman made waves on Sunday with bold predictions about artificial intelligence's immediate future. In a blog post titled "Reflections," Altman declared that OpenAI now knows how to achieve artificial general intelligence and projected that AI agents will start broadly impacting work operations in 2025 (of course for many companies, they already are).

OpenAI defines AGI as "highly autonomous systems that outperform humans at most economically valuable work." Altman’s statements have drawn intense criticism from some industry experts, including Gary Marcus (professor emeritus of neural science at NYU and one of AI’s leading critics), who humorously challenged OpenAI's optimistic timeline. Marcus also pointed to what he believes are continued barriers to true AGI - but those seem more like to barriers to ASI, or superintelligence. Humans make mistakes and can lack common sense, too; replacing them doesn’t require a perfect autonomous system, just one that makes sense economically.

Adding context to Altman’s predictions is a recently revealed financial agreement between OpenAI and Microsoft, which specifically defines AGI achievement as an AI model generating at least $100 billion in profits. Altman himself has acknowledged the potential impacts of AGI, suggesting that universal basic income might be necessary to address the economic disruption that could follow.

It’s tricky to see through the (very obviously self-serving) Sam Altman hype machine sometimes - so Marcus’ perspective has some merit. However, it’s worth considering the speed of advancements here. o3 debuted just three months after o1. Additionally, we should understand the prevalence of the current synthetic/LLM-generated data paradigm - just check out China’s most advanced LLM, DeepSeek, self-identifying as ChatGPT more often than itself. That’s because so much of DeepSeek’s training data was generated by GPT-4o - the synthetic trend we are likely to see continue. Data generated by o3 itself now becomes the de-facto gold standard for pre-training frontier level models (with a caveat; more on that in the paragraph below) - and OpenAI owns that data. That’s a huge advantage to have in the AI training data scarcity age.

All that should make the hairs rise on the back of your neck just a bit: we’re already at the point where we are more reliant on data generated by LLMs than we are on data made by humans to push the needle. OpenAI will make their next model better by using data from o1 and o3. This data can and does still suffer from hallucinations and other issues, but it is contracted out to companies like Data Annotation, Outlier and many others who employ ‘human experts’ to review and fix the LLM-generated data based on their subject matter expertise so it can then be fed right back into the machine.

Well, for now they employ human experts to review it, anyway! The idea of more intelligent and robust agentic AI systems coming down the pipeline, combined with this sort of recursive data loop leads down some fairly dystopian paths - we could be teetering on self-improving AI sooner than expected.

Consumer Electronics Show 2025 is here (and yes, all about AI)

-------------------------

If you were walking the floor at CES back in 2017, everything on display probably looked like a bit of a joke. Seemingly every single product had Wi-Fi for the sake of having Wi-Fi, and everything required an app. There was a smart toaster, a smart rubber duckie and multiple app-connected trash cans.

Ha ha, you might have said - many of the items seemed little more than sight gags or just too far-fetched to be useful. But the smart appliance industry has shown incredibly steady growth, with an expected value of $100B worldwide by 2029. CES 2025, which kicked off Tuesday in Las Vegas, is likely to have a similar vibe. Except this time, everything is going to have artificial intelligence, whether it needs it or not. 

The over 4,000 companies presenting at CES are betting that AI isn’t just for work. One startup plans to show an AI-powered spice dispenser. SoundHound will be demonstrating use cases for generative AI in cars. There will be various AI-infused curiosities in the health care space, including a startup that says they perform 90-second eye exams and a company that makes an AI-powered hormone sensor. Nvidia kicked off the show with an impressively powerful $3,000 AI supercomputer.

It remains to be seen what kind of AI-powered gadgets consumers actually want. CES has long been home to products making audacious claims that are hard to verify from a show floor, and not all of those inventions even make it to market. When it comes to AI gadgets in particular, there’s often a big gap between what companies have promised and what the technology is actually capable of, let alone what consumers will actually use.

The show goes through Saturday, so next week we’ll have a full breakdown of the latest AI-infused gadgetry coming your way.

MORE IN AI THIS WEEK

• Sam Altman’s younger sister files lawsuit claiming he sexually abused her (Altman and family responded on X/Twitter)

• Apple urged to withdraw 'out of control' AI news alerts

• Google is building its own ‘world modeling’ AI team for games and robot training

• Anthropic is in talks to raise a new round of funding that could value the company at $60 billion

• OpenAI shuts down engineer who created ChatGPT-powered robotic sentry rifle

• Microsoft President Brad Smith’s blog post highlights $80B AI data center investment in 2025, outlines vision for American AI leadership

• Food and Drug Administration unveils the first comprehensive draft guidance for AI-enabled medical devices - with over 1000 devices already authorized

• Man who exploded Cybertruck in Las Vegas used ChatGPT in planning, police say

• AI-designed drug shows promise for inflammatory bowel disease

Want to get the most out of ChatGPT?

ChatGPT is a superpower if you know how to use it correctly.

Discover how HubSpot's guide to AI can elevate both your productivity and creativity to get more things done.

Learn to automate tasks, enhance decision-making, and foster innovation with the power of AI.

Download the free guide with actionable prompts and tips to make your work more productive.

TECHNICAL NEWS, DEVELOPMENT, RESEARCH & OPEN SOURCE

• Nvidia introduces a set of open-source LLMs, for optimizing AI agents with 4x faster performance and scalable deployment options - also launches Cosmos with open-source video models to generate synthetic data for robotics

• Microsoft open-sources Phi-4 under MIT license, a 14B parameter model with near GPT-4 caliber benchmarks

• Human study on AI spear phishing campaigns

• AI improves cancer detection rates by 17% in largest-ever real world implementation of AI for breast cancer detection

• Anthropic unveils roadmap for Model Context Protocol

• Adobe’s TransPixar: advancing text-to-video generation

That’s all for this week! We’ll see you next Thursday.