In the AI race, there is one project that is surrounded by more mystery and anticipation than most: OpenAI’s “Project Strawberry”. The project, formerly known as Q*, promises to be the next big thing in the AI world and bring us closer to the elusive goal of Artificial General Intelligence (AGI).
Revealing the mystery
It is assumed that the main aim of Project Strawberry is to drastically improve the reasoning ability of AI models. This means that they can not only generate answers to queries, but also plan ahead, navigate the internet independently and do what OpenAI calls “deep search”. According to sources familiar with the matter, this capability has been lacking in AI models until now, making Strawberry a potentially groundbreaking development.
Mysterious progress
Details about Strawberry are being kept top secret within OpenAI. However, leaked information verified by Reuters and Bloomberg gives an insight into the ambitious scope of the project. Internal documents suggest that Strawberry includes a specialised re-training process, a method developed to optimise pre-trained models for specific tasks. This approach is similar to the Stanford project’s Self-Taught Reasoner (STaR) method, which allows AI models to iteratively generate their own training data and bootstrap themselves to higher intelligence.
Promising results
Initial tests show promising results. Internal demonstrations of Q* models, which are to be part of Strawberry, have solved advanced mathematical and scientific problems that go beyond the capabilities of commercially available AIs. One source reported a model that scored over 90% in a challenging AI maths competition.
The development of Project Strawberry is taking place against a backdrop of intense competition and rapid innovation in the AI industry. Companies such as Google, Meta and Microsoft are working hard to improve their AI models. However, the field remains divided on the ability of Large Language Models (LLMs) to integrate human thinking and long-term planning.
Notably, Yann LeCun, a pioneer of modern AI at Meta, has frequently argued that LLMs are not capable of mimicking human thinking. This scepticism is shared by other experts and underlines the ongoing debate about the feasibility and implications of realising AGI.
Challenges and scepticism
OpenAI’s journey with Strawberry has not been without its challenges. Recent high-profile departures from the company, including co-founder John Schulman, have raised doubts about the direction of the project and the overall stability of OpenAI. In addition, the AI industry in general is under scrutiny regarding the sustainability of its rapid growth, with some researchers suggesting that models are running out of training data.
The financial pressure is also high. Despite the initial success of ChatGPT, OpenAI is reportedly spending billions of dollars and struggling to maintain momentum. In this context, the Strawberry project is both a glimmer of hope and a major challenge.
Hype and reality
In a recent cryptic social media post, OpenAI CEO Sam Altman hinted at Project Strawberry with an innocent photo of strawberries in the garden, causing excitement and speculation among AI enthusiasts. While Altman’s post may hint at upcoming official news about Strawberry, it could also be seen as a strategic move to generate hype during a turbulent time for the company.
A thoughtful connection
Does the name “Strawberry” perhaps have something to do with Elon Musk’s famous warning? Musk once gave the example of an AI that was supposed to pick strawberries. He suggested that this may seem harmless at first glance, but that an AI that wants to maximise its efficiency could reshape itself in ways that could lead to disastrous results – for example, by turning the entire surface of the earth into strawberry fields to optimise its task. This cautionary tale about the unforeseen consequences of advanced AI capabilities adds an intriguing layer to the mystery of Project Strawberry.
The mystery of Project Strawberry continues to captivate the AI community and the public. With its potential to revolutionise the reasoning and autonomy capabilities of AI, it is a testament to the ambition and innovation driving the industry forward. However, as with all ground-breaking endeavours, there are also significant risks and uncertainties. As OpenAI tackles these challenges, the world waits with bated breath to see if Project Strawberry is indeed the breakthrough that will lead us into a new era of AI.
Sources:
