Search results for RLHF
AI & Enterprise
Frontier AI era’s ‘paradox of automation’ persists as progress remains slow
As AI models advance, companies are showing growing interest in autonomous AI agents that can carry out tasks without people. Yet complaints persist that AI-driven automation is still slow, even in Silicon Valley, highlighting a gap between expectations and on-the-ground experience. Some argue today’s large language models are built to assist humans rather than automate work, and that pursuing both goals can produce uneven performance. Surveys cited in reports show time saved by AI is often spent fixing or checking AI outputs.
AI & Enterprise
AI automation and work assistance cannot coexist. Why?
A new AI model called Jev from TypesafeAI, founded by Diogo Almeida, is drawing attention in tech circles for returning probabilities rather than text. Foundation Capital partner Jaya Gupta described it as a model focused solely on judgment. Almeida, who previously worked at OpenAI and co-developed GPT-4, InstructGPT and ChatGPT, said AI success stories usually involve human-in-the-loop assistance, while automation struggles. He attributed this to RLHF and said pursuing both assistance and automation leads to unstable performance.
AI & Enterprise
Typesafe AI\'s Jev probability model draws developers\' attention
Typesafe AI, founded by former OpenAI researcher Diogo Almeida (디오고 알메이다), has unveiled Jev, a new AI model that outputs probability values rather than generating sentences, TechCrunch reported on Sept. 18. The company says the model avoids hallucinations because users predefine outputs, and it highlights cost and speed advantages. Demand has been positive, and surging usage briefly strained its API service, it said.