BLEU, or Bilingual Evaluation Understudy, is an automatic metric originally developed for machine translation. It compares short sequences of words in generated text with one or more reference translations, combines precision across several n-gram lengths, and applies a penalty when output is too short. BLEU is inexpensive and useful for comparing systems on the same dataset, but it does not directly measure factuality, fluency, meaning, or human preference. Valid alternative wording can receive a low score, while awkward text can match many reference phrases. Results depend on tokenization and implementation, so reporting should include the exact evaluation settings and, when possible, human judgment.
Anthropic’s revenue per employee dwarfs Pinterest, DocuSign and Zoom. The bigger risk may be old software’s pricing power as AI platforms take on more work.
Anthropic’s IPO filing puts Dario Amodei’s 2025 security bill at $2.75 million, about 4.2x his salary. Daniela Amodei’s protection cost another $1.29 million.
OpenAI will pool ChatGPT and Codex usage on Plus, Pro and Business plans in December. Instant chat is exempt, with credits available for eligible features.
OpenAI has about $11 billion in commitments against an original $30 billion goal, WSJ reports. Its revised $20 billion round remains open.
House Democrats want Trump’s AI conversations preserved after reports he consulted Grok about Venezuela, raising questions about military advice and oversight.
Some employers are restoring roles after AI-led cuts as experience, oversight and the work outside a job description prove harder to replace.
Google ran Gemma 4 on an orbiting TPU and received a haiku back. The Suncatcher test now moves toward radiation, heat and networking challenges.
Recall Foundation plans to burn 358.6 million RECALL as it winds down. Token holders will receive no buyback or treasury distribution.
Some Amazon engineers reportedly ran unnecessary AI tasks to climb usage rankings. The company is warning against turning token consumption into a target.
FOMAT means Fear of Missing Agent Time: the pressure to keep coding agents working while you step away. More runtime does not always mean more progress.