The development of artificial intelligence in the US is gathering pace - new agent-based systems are changing the way programming is done, startups are racing for leadership and tech companies are looking for ways to survive strict rules from Brussels.

Americans take AI coding to the next level

Startup Moonshot AI introduced a new family of language models Kimi‑K2which focuses on the ability of artificial intelligence to perform more complex tasks. Version Kimi‑K2‑Instruct is trained to work with tools (e.g. internet search engines, calculators or databases) and achieves excellent results in programming, calculations or analysis.

  • Kimi-K2-Instruct beat other open models in tests such as LiveCodeBench (a test of real-time coding ability) and AceBench (an assessment of practical intelligence).
  • The model handles so-called multi-step problems - that is, problems that require sequential decision making.
  • Development is moving towards autonomous development assistants that can assist programmers as advanced colleagues.

Agent LLM systems are growing

Objevují se tzv. agentní systémy, which are types of artificial intelligence that no longer wait for commands from humans, but design solutions and perform complex procedures themselves. These systems can divide tasks, plan actions and evaluate their results.

  • Unlike regular "assistants", agent AI can plan and manage the entire work process.
  • Multiple "agents" (independent virtual workers) are used to communicate with each other and solve problems as a team.
  • This approach significantly speeds up software development and data analysis.

Silicon Valley is planning a Buildathon

DeepLearning.AI is hosting the first Buildathon - a competition where teams build a working AI app in just one hour. The goal is to show how effective today's agent tools are.

  • Participants will only be able to use tools based on AI models.
  • The team that can create the most functional solution within the time limit wins.
  • The event is intended to highlight how quickly AI can take you from idea to finished app today.

The US is keeping a close eye on European regulation

While the United States experiments with new types of intelligence, The European Union introduces a new legal framework - the so-called AI Act. This law is to ensure that AI is safe, fair and accountable. But it also affects US companies that want to operate in Europe.

  • The AI Act establishes obligations for all tools that may affect human rights or security.
  • US companies must prepare for rules such as data provenance, transparency or risk testing.
  • Pomoci jim má nový kodex správné praxe, which is the recommended way to comply with the regulations without legal hassle.

The Batch - DeepLearning.Ai by Andrew Ng / gnews.cz - GH