top of page
Writer's pictureRich Washburn

Strawberry Signals? Sam Altman Teases Q-Star


Altman Leak Q-Star

On August 7th, Altman posted a seemingly innocent photo of strawberries with the caption, “I love summer in the garden.” This tweet, however, has been interpreted as a significant clue hinting at the imminent release of OpenAI’s next major model, rumored to be codenamed “Strawberry” or “Q*Star.”


Altman’s tweet, while ostensibly mundane, was enough to send AI enthusiasts into a speculative frenzy. The term "Strawberry" is believed to be the codename for OpenAI's next Frontier Model, potentially GPT-5. The excitement is not unfounded; previous versions of GPT have set new standards in AI capabilities, and the community is eager to see what advancements the next iteration might bring.


Adding fuel to the fire, two anonymous models have recently appeared on lmsys.org, a platform known for testing AI models. This stealthy approach mirrors OpenAI’s strategy for previous model releases, dropping new models anonymously for the community to discover and test. While these models have been hard to track down, notable AI enthusiasts have reported their existence and shared initial impressions.


Jimmy Apples, a well-known figure in the OpenAI community, posted about a model named “anonymous chatbot.” From preliminary tests, this model appears to be a fine-tuned version of GPT-4, though its exact capabilities remain uncertain. Hater Atlow, a developer closely monitoring these developments, provided a detailed breakdown of what the AI community expects from Project Strawberry.


Project Strawberry, also known as Q*Star, aims to advance AI’s reasoning and planning capabilities. This project focuses on allowing large language models to think ahead, plan, and execute long-term tasks—a critical step towards achieving Artificial General Intelligence (AGI). Some of the rumored features include:


Autonomous Internet Navigation**: The ability to browse the web, perform deep research, and take actions based on findings.

Continuous Learning**: Unlike current models that are static post-training, Strawberry aims to continuously learn and improve.

Enhanced Reasoning**: Improved logic and reasoning skills, particularly in mathematical and strategic thinking.


In the wild west of model testing, users have put these anonymous models through their paces. One particularly challenging test involved reasoning through a logic puzzle: “There are three killers in the room. Someone enters and kills one. How many killers are left?” The model known as “sus column R” provided a detailed step-by-step reasoning, showcasing improved cognitive processing, though not without flaws.


Another tough question involved the location of a marble after a series of movements. While many models struggled, “sus column R” delivered an accurate response, further hinting at its advanced reasoning capabilities.


Bendu Ready from Abacus AI offers a tempered perspective, noting that while Project Strawberry aims to leap ahead, other AI labs like Google’s DeepMind have made significant strides in reasoning and math, as evidenced by their recent success at the Math Olympiads. This competition underscores the rapid pace of AI advancements and the high stakes for leading AI labs.


The hype surrounding Strawberry is palpable, fueled by community figures like I Rule the World Mo, who has been fervently posting about the project. Whether an insider or a master troll, his enthusiasm mirrors the broader anticipation for what could be a groundbreaking development in AI.


As we wait for official announcements, the question remains: Is Sam Altman simply teasing the community, or is the next frontier of AI truly upon us? Only time will tell, but one thing is clear—the AI world is on the edge of its seat.






19 views0 comments

Comments


bottom of page