ByteDance's Seed unit creates global AI safety team 'Alignment & Preparedness' and plans to hire ten researchers
ByteDance, owner of TikTok, has created a dedicated AI safety team inside its Seed research unit and opened a global search for ten researchers. The team, named Alignment & Preparedness, will work to ensure AI systems follow human objectives and safety requirements, including on complex multi-step tasks. Positions are located in Beijing and Shanghai (China) and San Jose (California). According to the state-run outlet The Paper, Seed is the first leading Chinese AI lab to set up an independent safety and alignment team. This is industry news with no direct threat or incident, and no vulnerabilities, exploitation or attribution are reported.
Detalhes técnicos
The announced research areas are:
- Safety training: methods to reduce risks while models are being trained.
- Automated testing: tools to detect flaws and vulnerabilities.
- Monitoring and explainability: analysis of system behavior and the factors that shape responses.
- Simulation environments: evaluation of autonomous agents in controlled scenarios.
- Emerging risks: investigation of threats that may arise as the technology advances.
Seed says the team will get compute and research resources comparable to those of the groups building its most advanced models. It also plans to seek independent evaluations and partnerships with universities and research institutions, and to feed the results into model training, evaluation and deployment. The announcement comes amid growing concern about AI agents that can access the internet and act autonomously, and while China and the US discuss AI safeguards and communication about serious incidents involving advanced systems.
Linha do tempo
-
ByteDance's Seed unit creates global AI safety team 'Alignment & Preparedness' and plans to hire ten researchers
ler a análise desta fonte ocultar
ByteDance, owner of TikTok, has created a dedicated AI safety team inside its Seed research unit and opened a global search for ten researchers. The team, named Alignment & Preparedness, will work to ensure AI systems follow human objectives and safety requirements, including on complex multi-step tasks. Positions are located in Beijing and Shanghai (China) and San Jose (California). According to the state-run outlet The Paper, Seed is the first leading Chinese AI lab to set up an independent safety and alignment team. This is industry news with no direct threat or incident, and no vulnerabilities, exploitation or attribution are reported.
Technical Details
The announced research areas are:
- Safety training: methods to reduce risks while models are being trained.
- Automated testing: tools to detect flaws and vulnerabilities.
- Monitoring and explainability: analysis of system behavior and the factors that shape responses.
- Simulation environments: evaluation of autonomous agents in controlled scenarios.
- Emerging risks: investigation of threats that may arise as the technology advances.
Seed says the team will get compute and research resources comparable to those of the groups building its most advanced models. It also plans to seek independent evaluations and partnerships with universities and research institutions, and to feed the results into model training, evaluation and deployment. The announcement comes amid growing concern about AI agents that can access the internet and act autonomously, and while China and the US discuss AI safeguards and communication about serious incidents involving advanced systems.