According to Beating, MIT Kaiming He’s team recently released ELF (Embedded Language Flows), a language diffusion model that departs from the autoregressive “predict next token” approach used by GPT-style models. Instead, ELF performs text generation in a continuous embedding space, converting to discrete tokens only in the final step.
In OpenWebText unconditional generation benchmarks, the 105M-parameter ELF-B achieved approximately 24.1 generation perplexity (Gen. PPL) with 32-step sampling, outperforming multiple discrete and continuous diffusion language model baselines. Notably, ELF-B required only approximately 45 billion training tokens, roughly one order of magnitude fewer than comparable methods which typically exceed 500 billion tokens.
Disclaimer: The information on this page may come from third parties and does not represent the views or opinions of Gate. The content displayed on this page is for reference only and does not constitute any financial, investment, or legal advice. Gate does not guarantee the accuracy or completeness of the information and shall not be liable for any losses arising from the use of this information. Virtual asset investments carry high risks and are subject to significant price volatility. You may lose all of your invested principal. Please fully understand the relevant risks and make prudent decisions based on your own financial situation and risk tolerance. For details, please refer to
Disclaimer.
Related Articles
Andrew Ng Dismisses AI Unemployment Crisis, Cites 4.3% U.S. Jobless Rate and Strong Software Engineering Hiring
According to Beating, Andrew Ng issued a statement today dismissing widespread claims of AI-driven mass unemployment, describing the narrative as unsupported by data. Citing software engineering—the field most exposed to AI coding agents—Ng noted that hiring remains robust and the U.S. unemployment
GateNews9m ago
Korea Exchange Deploys AI in Core Market Operations on May 13
According to The Chosun Daily, on May 13, Korea Exchange deployed AI in core market operations with Fair Labs, a South Korean AI startup it acquired in February. The first use case is a system that scans local and overseas news for disclosure issues at listed firms that could trigger trading
GateNews51m ago
Tesla Plans More AI Products for China, Leverages 400+ Local Suppliers
According to China Daily, Tesla plans to bring more AI and intelligent products to China, with Vice President Grace Tao highlighting the company's deep supply chain integration in the country. As of March 2026, Tesla works with over 400 Chinese suppliers, more than 60 of which supply its global
GateNews1h ago
SoftBank Reports $45 Billion in Unrealized Gains from OpenAI Investment as of End-March
According to ChainCatcher, SoftBank reported cumulative unrealized gains of $45 billion from its OpenAI investment as of the end of March. In April 2026, the company deployed $20 billion in borrowings primarily for OpenAI investment, of which $2.5 billion was repaid in the same
GateNews1h ago
Chinese Ambassador Xie Feng Calls for Healthy AI Competition With U.S. Ahead of Trump Visit
According to the U.S. Embassy in China, Chinese Ambassador to the United States Xie Feng discussed U.S.-China relations and President Trump's upcoming visit to China in an interview with Newsweek senior diplomatic policy correspondent Tom O'Connor, released ahead of the
GateNews1h ago
EU Discusses Joining US-Backed Pax Silica Tech Alliance in Coming Weeks
According to Bloomberg, the European Union is in talks to join Pax Silica, a US-backed alliance for semiconductors, AI infrastructure, critical minerals, and data centers, as part of efforts to reduce reliance on China. The European Commission plans to send a senior official to the US in the
GateNews1h ago
