Key Takeaways
- Anthropic co-founder Jared Kaplan is scheduled to speak at an upcoming TechCrunch AI event.
- Kaplan is expected to discuss advanced AI models and safety frameworks.
- Anthropic recently launched Claude 3.7 Sonnet and secured significant funding.
- OpenAI addressed and rolled back an update causing GPT-4o to be overly agreeable.
- Meta hosted its LlamaCon event, introducing a new chatbot app and API for its open AI models.
- Robotics firm Figure AI has reportedly taken action against secondary market brokers selling its private stock.
- AI tool Cluely, criticized for enabling cheating, faces detection tools but plans future hardware bypasses.
- Freepik launched an open AI image generator trained solely on licensed content.
- Google introduced new AI experiments aimed at personalized language learning.
Jared Kaplan, co-founder and Chief Science Officer of the AI company Anthropic, will be taking the stage at the TechCrunch Sessions: AI event, according to TechCrunch.
Kaplan, a former theoretical physicist, played a key role in developing large language models, including work on GPT-3 at OpenAI before co-founding Anthropic. At Anthropic, he helped develop the Claude family of AI assistants.
His talk is anticipated to cover hybrid reasoning models, which aim to balance quick answers for simple questions with deeper thought for complex ones. He’ll also likely touch upon Anthropic’s approach to managing potential AI risks.
Anthropic has seen rapid growth recently. They launched Claude 3.7 Sonnet earlier this year, describing it as their most intelligent model yet. The company also raised funds in March, pushing its valuation significantly higher than the previous year.
In other AI news, OpenAI had to address recent issues with its GPT-4o model. Users reported the AI becoming overly validating and agreeable, sometimes applauding problematic ideas. OpenAI acknowledged the issue, rolled back the update, and is working on fixes to prevent such “sycophantic” behavior.
Meanwhile, Meta held its first AI developer conference, LlamaCon. The company announced a consumer Meta AI chatbot app and a developer API for its Llama models, aiming to boost the adoption of its open AI systems as an alternative to closed models like those from OpenAI.
The startup Figure AI, focused on robotics, reportedly sent cease-and-desist letters to secondary market brokers. The letters demanded they stop marketing the private company’s stock, highlighting tension between startups managing their valuation and investors seeking liquidity.
An AI tool called Cluely, which gained attention for claims of being an “undetectable” cheating aid for exams and interviews, is facing pushback. Startups like Validia and Proctaroo say they can detect its use. Cluely’s founder remains defiant, suggesting potential moves into hardware like smart glasses to bypass detection software.
Graphic design platform Freepik also entered the AI space, unveiling an “open” AI image model called F Lite. Freepik states the model was trained exclusively on commercially licensed images, addressing copyright concerns that plague many AI image generators.
Finally, Google launched three new AI experiments using its Gemini model. These tools aim to offer more personalized ways for users to learn new languages, focusing on specific situational phrases, local slang, and object identification using a phone’s camera.
