Superpower Daily
Posts
Sam Altman says the pathway to AGI is now clear

Sam Altman says the pathway to AGI is now clear

The images of Spain’s floods weren’t created by AI. The trouble is, people think they were

Saeed Ezzati
November 12, 2024

In today’s email:

🧺 Physical Intelligence's first generalist policy AI can finally do your laundry
📉 Chegg Is On Its Last Legs After ChatGPT Sent Its Stock Down 99%
🧪 AI protein-prediction tool AlphaFold3 is now open source
🧰 11 new AI-powered tools and resources. Make sure to check the online version for the full list of tools.

Sam Altman says the pathway to AGI is now clear and "we actually know what do"

The discussion centers on the journey and insights of Sam Altman, co-founder of OpenAI, and his views on technology, startups, and the future. Altman reflects on his early belief in AI and how the field has evolved from being dismissed to a thriving area of innovation. He emphasizes the transformative power of conviction, citing OpenAI's commitment to AGI at a time when such ambitions were seen as unrealistic. This focus on scaling and deep learning led to breakthroughs like GPT models, with GPT-4 marking a significant leap in usability and commercial adoption.

Altman highlights the critical role of peer groups and environments like Y Combinator (YC) in fostering ambition and creativity among founders. He notes how YC's encouragement of bold, implausible ideas helped him shape his perspective and pursue impactful ventures. He also discusses the importance of conviction in startups, urging founders to focus on a singular bet rather than spreading efforts thinly. This strategy helped OpenAI navigate resource constraints and outperform competitors.

On the future of AI, Altman envisions advancements leading to abundant energy, near-limitless intelligence, and the discovery of new physics. He outlines a progression from chatbots (level one) to agents capable of solving complex, long-term tasks (level four) and eventually achieving organization-wide innovation (level five). He believes startups have a unique advantage in leveraging AI, given their agility and ability to iterate rapidly.

Altman concludes by advising entrepreneurs to embrace AI’s potential while adhering to fundamental business principles. He emphasizes the importance of building lasting value beyond impressive demos. Personally, he looks forward to the realization of AGI and the life changes brought by fatherhood, expressing optimism for the future of technology and humanity.

Finally, a developer tool that makes digging around for the answer a thing of the past.

Other developer tools can’t tell you how your codebase works and why. Unblocked can. We augment your source code with context from Slack, Confluence, Jira (and more), so your team gets helpful answers instantly.

Unblocked gives you:

Instant answers about every aspect of your codebase
Relevant, historical context for any file open in an IDE
A powerful Slack integration that answers questions in channels, removing interruptions to you

⌛Engineers who use Unblocked save an hour or more a day.

“Unblocked empowers every developer with the insight once reserved for the seasoned few. ” Mathew Spolin - SVP Engineering at AppDirect

Physical Intelligence's first generalist policy AI can finally do your laundry

The AI revolution has brought significant advancements in language and image generation, yet human intelligence remains unmatched in the physical world. Tasks like folding a shirt or cleaning a table represent monumental engineering challenges for AI, compared to easier, abstract tasks like playing chess. To bridge this gap, Physical Intelligence has developed a general-purpose robot foundation model, π0 (pi-zero), marking a step toward artificial physical intelligence. Trained on diverse text, image, and embodied robot data, π0 can output motor commands and control various robots, enabling it to perform tasks like folding laundry or assembling boxes. This model represents a paradigm shift, making robots more versatile and adaptive, much like language models in the NLP space.

Unlike specialized industrial robots programmed for repetitive tasks, π0's generalist capabilities come from training on the largest robot interaction dataset to date. This dataset includes dexterous tasks across eight robots, involving scenarios like table bussing, packing items, and handling deformable objects like laundry. By combining vision-language pretraining and a novel flow-matching approach for real-time control, π0 can autonomously adapt to complex, dynamic tasks. The model's ability to recover from interruptions or employ emergent strategies, such as stacking dishes or adjusting box folds, demonstrates its adaptability and dexterity.

Evaluation shows that π0 significantly outperforms other models like OpenVLA and Octo on complex, multi-stage tasks. While pretraining provides general physical understanding, fine-tuning enhances performance on specific, challenging tasks. For instance, π0 can unload a dryer, fold tangled clothing into neat stacks, or bus tables with high accuracy, highlighting its capability for real-world applications.

Looking ahead, π0 is an early step toward general-purpose robot models, with future research needed in long-horizon reasoning, safety, and autonomous self-improvement. Physical Intelligence is collaborating with robotics labs and companies to refine hardware, scale data collection, and adapt models to diverse platforms. They invite partnerships and are hiring researchers to advance this groundbreaking work.

The images of Spain’s floods weren’t created by AI. The trouble is, people think they were

A striking photograph from the aftermath of devastating floods in Valencia, Spain, went viral on social media, sparking claims it was AI-generated. The image, captured with a telephoto lens, depicted a chaotic scene of over 70 vehicles tossed around a narrow street by torrents of water from the “rain bomb” that dropped a year’s worth of rain in a single day. Despite its authenticity—confirmed through detective work involving Facebook, Apple Maps, and Street View—many viewers dismissed it as an AI creation. The sheen of the cars, peculiar shapes, and lack of visible damage may have contributed to the skepticism. However, the real culprit may be the growing distrust fueled by the surge of AI-generated content online.

This phenomenon, dubbed “AI slop,” is flooding social media with fabricated images and texts, making it harder for users to discern reality. Social platforms, rather than combating this trend, seem to embrace it. Meta’s CEO, Mark Zuckerberg, recently touted AI-generated feeds as a way to enhance user engagement and create new forms of content. The business logic is clear: more engagement means more profit, even if it means inundating platforms with synthetic media.

The problem extends to the creators of AI content, who also profit from this trend. Viral images, often emotionally manipulative, generate income through performance bonuses tied to likes and shares. This creates a feedback loop where platforms and content creators alike benefit financially, while users are left navigating a deluge of dubious material. Ultimately, this trend risks undermining trust in online media, pushing platforms toward what some fear is a state of terminal decline.