OpenAI Trains New AI Model, Creates Safety Committee to Mitigate Risks

OpenAI Trains New AI Model, Creates Safety Committee to Mitigate Risks

Artificial intelligence company OpenAI has announced the training of a new AI model that is set to succeed GPT-4, the model currently powering ChatGPT. OpenAI claims that the new model will exhibit the "next level of capabilities" but has not specified what these capabilities will be. As the company continues to push the boundaries of AI, it has also created a Safety and Security Committee to assess the risks that the technology may pose and provide an initial set of recommendations to mitigate those risks within 90 days. The committee will be led by board members Bret Taylor, Adam D'Angelo, Nicole Seligman, and CEO Sam Altman, and will include external experts such as Rob Joyce, former cybersecurity director at the National Security Agency, and John Carlin, a former Department of Justice official.

Deepfaked Fashion Designer's Body Used to Sell Counterfeit Products

Cassey Ho, CEO and founder of fashion brand Popflex, has raised concerns about the violation of intellectual property rights through the use of AI. Ho claimed that her images and those of her brand's models and customers were stolen and used to sell counterfeit versions of her products on platforms such as Amazon and Walmart. The fashion designer's body was also deepfaked using AI tools to create more deceptive counterfeit listings. Forbes spoke to Ho, who said she found the deepfake "incredibly violating." This incident highlights the potential misuse of AI technology and the need for effective measures to protect intellectual property and personal identities.

Nvidia Reports Record Profit, Stock Soars

AI chipmaker Nvidia had a successful week on Wall Street, with its stock price reaching a record high of over $1,064 per share on Friday. The company reported a 630% profit increase and a 270% increase in revenue compared to the previous year's quarter, making it Nvidia's most profitable and highest sales quarter ever. CEO Jensen Huang's net worth also increased by over $10 billion, propelling him into the top 20 of Forbes' richest people rankings. The strong earnings report highlights the growing demand for AI hardware and the success of Nvidia's AI business, which includes data centers, autonomous driving, and gaming.

AI in Real Estate: The Rising Role of Generative AI in the Industry

An AI assistant called Sidekick is being marketed to real estate agents to help them conduct property research, compare prices, and generate descriptions of listings from uploaded photos. Sidekick, which is built on GPT-4 and trained on a property database called MLS, aims to simplify the data analysis process for real estate agents, allowing them to focus on their sales skills. The current focus is on the Miami and San Francisco markets, with plans to expand to other cities in the future. This highlights the increasing adoption of AI in the real estate industry to enhance efficiency and improve the user experience.

AI Deals of the Week

Several AI deals have been announced this week:

  • Helsing, a defense AI company, is in talks to raise $400 million at a $4 billion valuation, led by General Catalyst.
  • DeepL, a Germany-based AI translation startup, has raised $300 million at a $2 billion valuation to focus on enterprise growth.
  • H, a French AI startup, has raised $220 million in seed funding to build AI agents for business applications.

These deals highlight the continued investment and growth in the AI industry, focusing on areas such as defense, translation, and business applications.

Deep Dive: Google's Monopoly in Search Despite Inaccurate AI Answers

This week, Google rolled out AI search overviews, but the initial results contained a series of misleading answers. Some responses suggested that users could stare at the sun for up to 30 minutes and that geologists recommend eating one rock per day. Users also discovered that the AI treated satirical websites as legitimate news sources. Following the backlash, Google removed some of the answers. Alternative search engines like DuckDuckGo and You.com expressed that while Google's AI may be inaccurate, the company's monopoly in search remains strong because it is the default search engine on popular devices and browsers. The situation highlights the potential challenges of using AI in search and the ongoing debate over monopolistic practices in the tech industry.

AI Index

  1. Out of every 6 responses generated by AI models for legal queries, 1 is incorrect, according to a new Stanford study.
  2. A previous study showed that 75% of responses to questions about a court's ruling created by general-purpose large language models like OpenAI's GPT-3.5, Google's PaLM 2, and Meta's Llama 2 were also incorrect.
  3. Elon Musk is embroiled in an online feud with this AI leader, who also attended Musk's AI conference in 2017.

Model Behavior: A new app called Style DNA uses AI to analyze a person's existing wardrobe to make styling and outfit recommendations. The UK-based company claims over 3 million downloads and 300,000 active users. It works by scanning a person's selfie to suggest color palettes for outfits. Founded in 2019 by Elena Volkova, the company aims to solve the "nothing to wear" dilemma using AI.

Read more