The authors of key papers that laid the muse for generative artificial intelligence (AI), resembling ‘Deep Reinforcement Learning’ and ‘Generative Adversarial Networks (GANs)’, have founded a startup. The corporate is developing an AI agent.
Business Insider reported on the seventh (local time) that AI agent startup Reflection AI attracted initial investment at a company value of $100 million (about 140 billion won) in a funding round led by Quoia Capital.
Reflection AI is developing AI agents that handle diverse and complicated tasks for businesses without human intervention.
It was co-founded earlier this 12 months by Google DeepMind veteran researchers Ioannis Antonoglu and Sergil Ozar, and Misha Raskin, who joined DeepMind after working at Berkeley AI Lab.
Co-founder Antonoglu participated in the development of DeepMind’s ‘AlphaGo’, and can be famous as considered one of the authors of the deep reinforcement learning paper developed by DeepMind in 2013 to enable the machine to learn Atari video games without human intervention. Until last 12 months, he was a key figure in leading reinforcement learning through human feedback (RHLF) in Google’s ‘Gemini’ development team.
Ozar can be considered one of the authors of GAN eye-tracking, which is as famous as ‘Transformers’. He left Google before Antonoglu and others and joined Tesla, but joined forces again to ascertain Reflection AI.
Their departure from Google received considerable attention, with news of their founding a brand new company breaking out in January.
Co-founder and CEO Raskin recently appeared on Sequoia Capital’s podcast and said he decided to start out an organization because “time was running out for me to remain at Google.”
“We imagine we’re only about three years away from having a model that approaches artificial general intelligence (AGI),” he said. “That’s what we call a universal agent.”
A general-purpose agent is an AI agent that may perform a wide range of tasks and process many inputs, moderately than simply one specific task.
“AlphaGo might be the deepest agent ever built, but it surely’s not very useful because it could actually only do one thing,” said CEO Laskin. In contrast, LLMs like Google’s Gemini, Antropic’s Claude, and OpenAI’s ChatGPT are general-purpose, but they’re not trained to be agents,” he said.
Reflection AI goals to construct “superhuman general agents,” that’s, AI agents with a big selection of abilities that surpass humans.
In Silicon Valley, startups developing AI agents like this are attracting attention from investors.
A representative example is Imbue, which is developing an agent focused on inference. After raising $200 million (about 280 billion won) in Series B funding in September last 12 months, it achieved a company value of $1 billion (about 1.4 trillion won).
French startup H (Holistic), which received $220 million (about 300 billion won) in investment from Amazon, Samsung, and others in May, can be a representative company that’s promoting the event of AI agents.
Reporter Park Chan cpark@aitimes.com
