Daily AI Newsletter: Former OpenAI CTO Mira Murati is reportedly fundraising for a new AI startup

In today’s email:

🔥 Solving complex problems with OpenAI o1 models
💀 ‘Musician’ Arrested for Using AI Songs and a Bot Army to Scam Spotify for Millions in Royalties
📚 The Ultimate Guide to Fine-Tuning LLMs
🧰 9 new AI-powered tools and resources. Make sure to check the online version for the full list of tools.

Former OpenAI CTO Mira Murati is reportedly fundraising for a new AI startup

Mira Murati, former Chief Technology Officer of OpenAI, is making significant moves in the AI industry as she seeks to raise over $100 million for her new startup. After playing a key role in major OpenAI projects such as ChatGPT and DALL-E, Murati has begun talks with venture capitalists to secure the funding necessary for developing proprietary AI models. This new venture marks her entrance into an increasingly competitive AI landscape, one filled with both commercial and open-source models that must constantly innovate to avoid obsolescence.

Reports suggest that Murati might receive support from Barrett Zoph, another former OpenAI executive, who left the organization around the same time as Murati. Together, they could bring a distinct approach to AI, emphasizing proprietary technology in an industry where maintaining an edge is a significant challenge. This choice reflects a strategic direction different from many other ventures in the space, potentially giving her startup a unique competitive advantage.

Murati’s new endeavor is part of a broader trend of ex-OpenAI executives establishing their own ventures. Figures like Ilya Sutskever and Andrej Karpathy have also started new initiatives in AI, focused on areas such as artificial superintelligence and education. Meanwhile, Jan Leike has joined Anthropic to continue his focus on AI safety. This wave of departures from OpenAI has sparked questions about the company’s current competitive edge in the push for artificial general intelligence, especially given the considerable capital now flowing into rival projects led by its former talent.

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 without having to search for them.

Unblocked gives you:

Instant answers about every aspect of your codebase
Relevant, historical context for any code open in an IDE
Automated answers to questions asked in Slack, removing interruptions to you

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

“Every developer now has the ability to tap into past discussions and decisions to fill their knowledge gaps, regardless of their tenure. We are moving faster and making more accurate decisions as a team as a result.” - Alex Mallet, EVP of Engineering at Forto

Start your 21-day trial today

Midjourney plans to let anyone on the web edit images with AI

Midjourney is preparing to release an upgraded web tool that will allow users to edit any uploaded images using its generative AI technology. This new feature, slated for release "early next week," will also enable users to retexture objects within images, essentially allowing them to change colors and details based on captions. The announcement was made by Midjourney CEO David Holz, who mentioned that the tool will initially be restricted to a subset of the existing community. The company plans to incorporate both human and AI moderators to curb potential misuse, but Holz acknowledged the challenges of managing such a release and mentioned that community feedback would help shape its deployment.

The use of AI for editing images has become a controversial issue, with platforms like Meta struggling to decide how to label AI-edited versus AI-generated content. Meanwhile, companies like Google are introducing AI features that do not necessarily indicate whether an image has been modified by AI. Although Midjourney has committed to using the IPTC's Digital Source Type property to embed metadata in AI-generated images, it hasn't adopted C2PA, a technology that tracks an image's complete history from creation to editing.

There are significant risks involved in releasing advanced editing tools without appropriate safeguards, as they can easily be misused to infringe on copyright or spread misleading deepfakes. Concerns have grown as the proliferation of deepfakes has made distinguishing between real and manipulated images increasingly difficult. A recent poll showed that 85% of Americans are worried about the spread of deepfakes, especially following events like Hurricane Helene, where fake AI-generated images of devastation circulated widely. The absence of a federal law criminalizing deepfakes in the U.S. has led states like California to introduce their own measures against AI-aided impersonation.

Midjourney has faced criticism in the past for its approach to responsible AI deployment, including ongoing lawsuits over the alleged use of copyrighted content in its training models. Nevertheless, the company has recently taken steps to limit deepfake proliferation, such as introducing filters to prevent political manipulation leading up to the U.S. presidential election.

ByteDance intern fired for planting malicious code in AI models

ByteDance, the owner of TikTok, recently confirmed that an intern was fired for planting malicious code that interfered with the training of its AI models. This confirmation followed widespread rumors that ByteDance had lost millions due to the intern's actions. According to ByteDance's statement, the intern, who was employed in the company's commercial technology team, was fired in August for serious disciplinary violations. ByteDance emphasized that none of its large models or commercial projects were affected, dismissing online claims of a massive financial loss as "seriously exaggerated."

Despite ByteDance's attempt to clarify the situation, many online commenters remain skeptical. Some questioned the distinction between the AI Lab and the commercial technology team, while others suggested ByteDance was downplaying the actual damage. One commenter claimed that the intern deliberately sabotaged model training over several months, resulting in significant losses in computing power and wasted research efforts.

ByteDance's struggle to keep up in the AI race is well-documented, with its chatbot, Doubao, largely considered a failure compared to rivals like Baidu. To compete with AI giants like Meta, Apple, and Google, ByteDance secured a $10.8 billion loan to support its expansion in AI research and e-commerce. However, retaining talent is proving difficult, especially as regulatory scrutiny over privacy and security concerns regarding TikTok intensifies. The company also faces the challenge of a potential nationwide TikTok ban in the US unless it is sold by early 2025, though legal experts believe the government will face significant challenges enforcing the ban.