Exclusive: OpenAI's Secret Project ‘Strawberry’ Set to Revolutionize AI Reasoning


July 12 - OpenAI, the renowned creator of ChatGPT, is advancing its artificial intelligence capabilities with a groundbreaking project code-named “Strawberry,” according to a reliable source and internal documents reviewed by Reuters.

The Strawberry project, details of which were previously undisclosed, signifies a major step forward in AI reasoning. This development is crucial as OpenAI, backed by Microsoft, strives to demonstrate that its models can deliver sophisticated reasoning abilities.

A Secretive Project

Internal documents from OpenAI, dated May, reveal that Strawberry is a collaborative effort within the company, aimed at enhancing AI’s ability to perform advanced research autonomously. While the exact timeline for Strawberry’s public release remains uncertain, the source described the project as an ongoing effort.

Details on how Strawberry functions are closely guarded secrets within OpenAI. However, the project’s goal is clear: to enable AI not only to generate answers but also to plan ahead, navigate the internet autonomously, and perform what OpenAI calls “deep research.”

Breakthrough in AI Reasoning

The Strawberry initiative was previously known internally as Q*, a project already considered a breakthrough within the company. Earlier this year, OpenAI staff demonstrated Q*’s ability to answer complex science and math questions that current commercial models struggle with.

During a recent internal meeting, OpenAI showcased a project with new human-like reasoning skills, though it is unclear if this demonstration was related to Strawberry. OpenAI confirmed the meeting but withheld specific details.

Transformative Capabilities

Strawberry aims to revolutionize AI reasoning by processing models post-training on vast datasets. This approach promises to significantly enhance AI’s reasoning capabilities, enabling models to plan ahead, understand the physical world, and tackle multi-step problems reliably.

Improving reasoning is seen as a pivotal step towards achieving human or super-human-level AI intelligence. Current large language models can summarize dense texts and generate elegant prose quickly but often falter on common-sense problems, resulting in “hallucinations” or incorrect information.

AI Industry Implications

OpenAI’s CEO, Sam Altman, has emphasized the importance of reasoning in AI development. Other tech giants like Google, Meta, and Microsoft, as well as academic labs, are also exploring ways to improve AI reasoning. However, there is debate within the AI community about whether large language models can incorporate long-term planning and reasoning.

The Future of AI Research

Strawberry represents a significant component of OpenAI’s strategy to address these challenges. The project involves post-training techniques, including fine-tuning, to refine AI performance. Strawberry’s methodology bears similarities to Stanford’s 2022 “Self-Taught Reasoner” (STaR), which allows AI models to iteratively create their own training data to achieve higher intelligence levels.

Among Strawberry’s ambitious goals is enabling AI to perform long-horizon tasks (LHT), which require extensive planning and execution over time. OpenAI is developing a “deep-research” dataset to train and evaluate these capabilities, though specifics remain undisclosed.

Additionally, OpenAI intends for its models to autonomously browse the web and conduct research using a “CUA” (computer-using agent) that can act on its findings. These advanced capabilities will also be tested for software and machine learning engineering tasks.

Strawberry is poised to be a transformative force in AI, pushing the boundaries of what these models can achieve and paving the way for future innovations in artificial intelligence.

Comments

Post a Comment

Popular posts from this blog

Faulty CrowdStrike Update Crashes Windows Systems, Impacting Businesses Worldwide

Businesses across the world have been hit by widespread disruptions to their Windows workstations stemming from a faulty update pushed out by cybersecurity company CrowdStrike. "CrowdStrike is actively working with customers impacted by a defect found in a single content update for Windows hosts," the company's CEO George Kurtz said in a statement . "Mac and Linux hosts are not impacted. This is not a security incident or cyber attack." The company, which acknowledged "reports of [ Blue Screens of Death ] on Windows hosts," further said it has identified the issue and a fix has been deployed for its Falcon Sensor product, urging customers to refer to the support portal for the latest updates. For systems that have been already impacted by the problem, the mitigation instructions are listed below - Boot Windows in Safe Mode or Windows Recovery Environment Navigate to the C:\Windows\System32\drivers\CrowdStrike directory Find the file nam...
Read more

APT33 Expands Operations Targeting Aerospace, Satellite, and Energy Sectors Across the U.S., Europe, and Middle East

A renewed wave of cyber activity linked to the Iranian threat group APT33 has drawn attention to the increasing sophistication of state-aligned cyber operations targeting strategic industries worldwide. The group—also tracked under names such as Elfin, Refined Kitten, Magnallium, and Peach Sandstorm —has been associated with a series of campaigns affecting organizations in the aerospace, satellite, and energy sectors across the United States, Europe, and the Middle East . Unlike purely espionage-focused campaigns, recent operations attributed to APT33 appear to follow a dual-mandate strategy : collecting strategic intelligence while maintaining the capability to escalate into disruptive cyber activity if geopolitical conditions demand it. This blend of espionage and latent disruption reflects the evolving role of cyber operations as a strategic instrument of state power. APT33: A Long-Running Iranian Cyber Threat Actor APT33 is widely assessed by multiple threat intell...
Read more

Stealthy BITSLOTH Backdoor Exploits Windows BITS for Covert Communication

A newly discovered Windows backdoor, dubbed BITSLOTH, is making waves in the cybersecurity community for its cunning exploitation of the Background Intelligent Transfer Service (BITS) to facilitate stealthy communication. This advanced threat poses significant risks to organizations and individuals alike, highlighting the ever-evolving tactics of cybercriminals. Overview of BITSLOTH BITSLOTH leverages the legitimate Windows service BITS, which is typically used for background file transfers and updates, to establish a covert communication channel. Key features of this backdoor include: Stealthy Operations: By using BITS, BITSLOTH can blend in with normal network traffic, making it difficult for traditional security measures to detect its presence. Persistent Backdoor: Once installed, BITSLOTH creates a persistent backdoor that allows attackers to maintain long-term access to compromised systems. Data Exfiltration: The backdoor is capable of exfiltrating sensitive data from the targe...
Read more