Photo credit: www.forbes.com
Welcome back to The Prompt,
Recently, an incident involving Cursor, a prominent AI coding tool developed by the emerging startup Anysphere, raised eyebrows when its customer service bot provided a fabricated response to a user query. This situation sparked discontent among several users.
The world of AI coding tools is rapidly evolving, with Cursor emerging as one of the star players in the industry. Despite its popularity, the platform experienced a glitch that exposed the limitations of AI technology—an occurrence known as “hallucination,” where artificial intelligence generates incorrect information presented as facts. Notably, an incident occurred when a developer discovered that switching devices resulted in being logged out of Cursor. Upon contacting customer support, they received a response from an AI agent named “Sam,” claiming this logout behavior was due to a recent policy change.
However, this policy change did not exist; the information was generated by the AI support bot. Michael Turell, CEO of Anysphere, confirmed the bot’s erroneous response on Reddit, prompting backlash from many programmers who utilize multiple machines for their work. Following the fallout, several users expressed intentions to cancel their subscriptions. Turell publicly apologized on Hacker News, ensuring the affected user was refunded and that future AI responses would be clearly identified as such.
Now, let’s delve into today’s key stories.
BIG PLAYS
OpenAI has unveiled its new AI models, o3 and o4-mini, which boast capabilities to analyze and manipulate images, conduct web searches, and process information from uploaded documents. According to the company, these models represent their most advanced reasoning systems to date. Unlike earlier versions, the new models require increased processing time and computational resources to interpret inputs effectively. They also utilize external tools, allowing them to perform tasks autonomously.
Users have begun experimenting with these models by uploading various photographs to ChatGPT to determine their exact locations, complete with GPS coordinates and navigation directions. Some have found it particularly enjoyable to challenge the model as if it were a game of GeoGuessr, where players identify locations based on imagery.
Simultaneously, a new social media craze has emerged, with users employing ChatGPT to design AI-generated action figures and Barbie dolls that feature their own appearances. These ChatGPT-inspired Barbies sport an array of professions, including real estate agents, tech specialists, and business coaches. In response, some artists have taken to hand-drawing their interpretations, as reported by the New York Times.
POLITICS + ELECTION
As the Canadian elections approach, Amazons’s virtual shelves have seen an influx of AI-generated books discussing figures like Prime Minister Mark Carney and others, as reported by Bloomberg. Notably, some of these titles include factual inaccuracies regarding the political leaders. In response, Amazon has removed most of the flagged publications. This incident highlights a broader trend—numerous titles have been generated using AI tools and are available through Amazon’s Kindle Direct Publishing.
AI DEAL OF THE WEEK
Anthropic has announced its first investment into a startup called Goodfire, based in San Francisco. Goodfire has successfully raised $50 million in a Series A funding round led by Menlo Ventures, with contributions from Lightspeed Ventures and more. The startup aims to tackle the black box problem of AI, addressing the significant gaps in understanding AI model behaviors. Their goal is to enhance comprehension of neural networks to facilitate new training methodologies and improve model performance.
DEEP DIVE
Before turning Cursor into an astoundingly successful AI tool, Michael Turell was just a student at MIT, feeling disenchanted during his internship at Google at the young age of 18. In 2019, while seated at the Computer History Museum’s café in Mountain View, he encountered Ali Partovi, who was scouting talent for his prestigious Neo Scholars program, which seeks to identify promising technology students.
Partovi, an early backer of tech giants like Facebook and Airbnb, engaged Turell in discussions about AI research and the dynamics of startup culture. To assess Turell’s skills, he presented him with a handwritten coding challenge, which Turell completed swiftly. Impressed, Partovi marked Turell’s dossier with a star, indicating his readiness to back any future endeavors Turell might undertake.
Eventually, Turell was selected as a Neo Scholar, and three years later, Partovi became the inaugural investor in Anysphere, the creator of Cursor. This startup reportedly achieved a remarkable $100 million in annual revenue within just a year—a milestone that has since doubled—placing it among the most rapidly expanding companies in history.
“Ali essentially endorsed us when we were only at the ideas stage. We didn’t have much else at that point,” Turell shared with Forbes. “He’s known for investing in potential rather than merely proven results.”
Founded in 2017, Neo selects 30 scholars annually, providing them with access to workshops, industry networking, and recruitment opportunities at leading tech firms. Similar to the Thiel Fellowship established by billionaire investor Peter Thiel, which offers significant financial support to entrepreneurs bypassing college, Neo Scholars can opt to pause their studies for one semester to pursue their projects, although it comes with a $20,000 grant, making it a less intensive commitment compared to Thiel’s program.
To explore the complete story of Neo, check it out on Forbes.
WEEKLY DEMO
In February, Perplexity’s CEO Aravind Srinivas announced that rather than spending millions on a Super Bowl ad, his AI search firm would award $1 million to a lucky user who downloaded the app, posed five questions, and referred friends. However, recent findings from the app testing firm Appknox revealed that Perplexity’s Android app harbors 10 significant security vulnerabilities, enabling impersonation attempts, account takeovers, and data theft, as reported by Forbes. These vulnerabilities could allow malicious actors to replicate the Perplexity app and access sensitive user information. This revelation emerges amidst talks of Perplexity integrating its AI assistant into smartphones manufactured by Lenovo and Samsung, as highlighted by Bloomberg.
MODEL BEHAVIOR
In the ongoing competition between humans and robotic capabilities, recent events demonstrated that there is still considerable ground to cover. In a half marathon held in Beijing, 21 humanoid robots participated alongside 12,000 human runners. Ultimately, only six robots managed to complete the race, as reported by Wired. Many of the robots faced challenges, such as overheating, resulting in their failure to finish. The fastest among them crossed the finish line in 2 hours and 40 minutes, but not without some setbacks, including falling down and requiring three battery changes. This race highlights that robots are still in the early stages of development.
Source
www.forbes.com