Chainfeeds Summary:
Why was Anthropic able to realize in 2021 that coding might be the most important direction?
Article source:
https://mp.weixin.qq.com/s/sA20Zc74FYWxKOu9Nu-T1Q
Article Author:
Overseas Unicorns
Opinion:
Overseas Unicorns: First, strategically speaking, OpenAI has always been more like a company that "wants everything." In terms of model capabilities, OpenAI is simultaneously developing in areas such as math, science, coding, reasoning, multimodal computing, and architectural innovation. At the product level, businesses such as Codex, browsers, robotics, enterprise platforms, smart hardware, chips, and data centers are also progressing concurrently. It's said that OpenAI once had as many as 300 projects internally. In stark contrast, Anthropic is the only one of the three giants that abandoned the multimodal approach early on, and has never emphasized architectural innovation, nor has it heavily promoted concepts like reasoning models, RL, or continuous learning. They focus on only one thing: perfecting language model scaling and heavily betting on coding, prioritizing mastering the most critical capabilities. The market is now increasingly clear about why coding is so important. There are three core reasons: 1) Coding is the path to everything. The vast majority of tasks in the digital world can ultimately be expressed through code. 2) Coding is one of the most suitable capabilities for model learning. Because the results are highly verifiable, the feedback loop is short, and data generated during user usage is more easily used to feed back into model training. 3) Coding is the core accelerator for AGI development. Leading AI labs have now entered an accelerated cycle, with the improvement in model capabilities within a single quarter this year even exceeding that of the previous year. Anthropic's early investment in coding was half vision and half luck. Anthropic's early fundraising was not smooth. Lacking sufficient funds, they had to find a more efficient path towards AGI. The team carefully studied the options: if they could only invest in one direction, coding might be the best choice. It has the potential to create a flywheel: first train a stronger coding model → provide it to customers → obtain usage data in real-world engineering environments → then feed back into model training. Anthropic's growth manager mentioned that he had read an internal document written by the co-founders in 2021, the core of which was why we should focus on coding. The key point is that at that time, the entire industry didn't even know how large the real market opportunity for coding truly was. However, as its funding improved, Anthropic began experimenting with more general-purpose model foundations. The real turning point came after the explosive success of ChatGPT. Anthropic realized that the consumer market had been captured by OpenAI, so it shifted its focus to B2B and gradually strengthened its coding capabilities. More unique than its strategy itself is Anthropic's organizational culture. In the past six months, amidst the fierce competition for AI talent, Anthropic's employee turnover rate has been far lower than other AI labs. Data shows that for every 10.6 people who jump from Google DeepMind to Anthropic, only 1 goes back to DeepMind. And for every 8.2 people who go from OpenAI to Anthropic, only 1 goes back to OpenAI. Furthermore, the percentage of Anthropic employees who remain with the company after two years is as high as 80%, even slightly higher than at DeepMind. Many employees who joined Anthropic mention that the company's most unique aspect is its culture. Anthropic has a very strong mission-oriented atmosphere. They truly believe that AGI could either save the world or destroy it, and Anthropic's mission is to help humanity safely navigate this transformation of transformative AI.
Content source 
