Surge AI

The page discusses Google's GoEmotions dataset, which contains 58,000 human-labeled Reddit comments across 27 emotions, and reports that 30% is mislabeled.

Surge AI's mission is to inject human values and intelligence into AI.

The page discusses AI red teams of creative data labelers who interactively test AI defenses to teach models to withstand counterattacks.

Anthropic cited two Surge AI benchmarks, GDP.pdf and Riemann-bench, in their Fable 5 and Mythos 5 release. A look at why expert-built evaluations matter at the frontier.

Surge AI lists roles in research, engineering, operations, partnerships, and finance and says positions are mostly remote.

The company says its mission is to raise AGI with human intelligence characterized as curious, witty, and imaginative.

Surge Labs, Inc.'s Privacy Policy explains how the company collects, uses, and discloses information.

The terms that govern use of Surge AI's website and services, including accounts, acceptable use, payment, intellectual property and liability.

Mecka AI raises a $60 million Series B led by Sequoia, with participation from Nvidia, Microsoft’s venture fund M12, and others. Founded in 2024, the startup pays people to record everyday tasks, such as making coffee or fixing cars, using body sensors and smartphones to produce data for training robots. The company aims to provide robotics with human-generated training data, similar to data-labeling firms serving large language models.

Oct 2026 · TechCrunchNews article

Mistral launches a public preview of Mistral Large 4, a natively multimodal model with 1 trillion parameters and 49 billion active parameters. The company says the model performs competitively in coding, agentic workflows, and multimodal understanding; it scores 82% on a vulnerability reproduction-and-patching test and 93% on Cybench. Trained from scratch on 3,800 NVIDIA Grace Blackwell GPUs in Mistral’s European datacenters, its weights are scheduled for release by the end of the month.

Oct 2026 · Mistral BlogNews article

Anthropic introduces Claude Sonnet 5.5, a faster, lower-cost model for everyday tasks, bug fixes, and creating polished documents, slides, and spreadsheets. It scores 70.6% on Terminal-Bench 4.0, generates outputs more than 30% faster than Sonnet 5, and can cost up to 30% less per task while keeping the same listed token prices. Anthropic says the model includes cybersecurity safeguards and fallbacks similar to those used for its most capable models.

Sep 2026 · AnthropicNews article

ZeroDrift launched Anchor 3.0, a family of small language models that checks AI-agent messages for regulatory and corporate-policy violations before they are sent. Available through its Enforcement API, the models support real-time checks, while ZeroDrift’s broader platform can flag, rewrite, block or route problematic communications. ZeroDrift says its flagship detected more than 95% of violations in a FINRA-based benchmark and matched GPT-6 Astra and Claude Fable 5.1 accuracy at over 34 times the speed and one-twelfth the cost; the company published the benchmark.

Sep 2026 · SiliconAngleNews article