# Post-Training Related Articles

HTX News Center provides the latest articles and in-depth analysis on "Post-Training", covering market trends, project updates, tech developments, and regulatory policies in the crypto industry.

DeepSeek V4 Official Version Arrives, New Capabilities Emerge, Value-for-Money King Enters the Fray

On July 31st, DeepSeek officially launched the public API beta for its DeepSeek-V4-Flash model. A key highlight is its performance on multiple Agent benchmark tests, reportedly nearing or even surpassing the level of the V4-Pro preview version from three months ago. Notably, the Flash model achieves this with significantly smaller scale (130B active parameters vs. Pro's 490B), suggesting that post-training optimization and data quality may be as crucial as raw model size. DeepSeek emphasized that the V4-Flash-0731 uses the same model architecture and size as its preview version, with improvements attributed solely to "re-trained post-training." The update also marks the official debut of DeepSeek's self-developed Agent framework, "Harness." The move signals DeepSeek's strategic push to position its cost-effective Flash model as a competitive base for Agent applications—scenarios requiring autonomous planning, tool usage, and complex task execution—where inference speed and cost are critical. By natively supporting OpenAI's Responses API format and adapting for code-generation scenarios, DeepSeek aims not just to be a cheaper alternative but to establish its own ecosystem in the Agent era. This release follows DeepSeek's record-breaking ~$50 billion fundraising round roughly two months prior, underscoring market confidence in its technology and commercialization prospects. The company is reportedly preparing for another funding round at a valuation of approximately $71 billion. The Flash model's advancement represents a step in fulfilling the high expectations that come with this valuation, setting the stage for the impending release of the V4-Pro official version and intensifying competition in the global Agent landscape.

marsbit07/31 08:01

DeepSeek V4 Official Version Arrives, New Capabilities Emerge, Value-for-Money King Enters the Fray

marsbit07/31 08:01

OpenAI Post-Training Engineer Weng Jiayi Proposes a New Paradigm Hypothesis for Agentic AI

OpenAI engineer Weng Jiayi's "Heuristic Learning" experiments propose a new paradigm for Agentic AI, suggesting that intelligent agents can improve not just by training neural networks, but also by autonomously writing and refining code based on environmental feedback. In the experiment, a coding agent (powered by Codex) was tasked with developing and maintaining a programmatic strategy for the Atari game Breakout. Starting from a basic prompt, the agent iteratively wrote code, ran the game, analyzed logs and video replays to identify failures, and then modified the code. Through this engineering loop of "code-run-debug-update," it evolved a pure Python heuristic strategy that achieved a perfect score of 864 in Breakout and performed competitively with deep reinforcement learning (RL) algorithms in MuJoCo control tasks like Ant and HalfCheetah. This approach, termed Heuristic Learning (HL), contrasts with Deep RL. In HL, experience is captured in readable, modifiable code, tests, logs, and configurations—a software system—rather than being encoded solely into opaque neural network weights. This offers potential advantages in explainability, auditability for safety-critical applications, easier integration of regression tests to combat catastrophic forgetting, and more efficient sample use in early learning stages, as demonstrated in broader tests on 57 Atari games. However, the blog acknowledges clear limitations. Programmatic strategies struggle with tasks requiring long-horizon planning or complex perception (e.g., Montezuma's Revenge), areas where neural networks excel. The future vision is a hybrid architecture: specialized neural networks for fast perception (System 1), HL systems for rules, safety, and local recovery (also System 1), and LLM agents providing high-level feedback and learning from the HL system's data (System 2). The core proposition is that in the era of capable coding agents, a significant portion of an AI's learned experience could be maintained as an auditable, evolving software system.

marsbit05/11 00:17

OpenAI Post-Training Engineer Weng Jiayi Proposes a New Paradigm Hypothesis for Agentic AI

marsbit05/11 00:17

活动图片

Bitcoinright-arrow

Trading Strategiesright-arrow