Sources: ahead of its March 16 event, Baidu scrambles to ready its ChatGPT-style Ernie Bot, which struggles with basic tasks, to get ahead of its Chinese rivals
Employees working round-the-clock say the AI chatbot, dubbed Ernie Bot, is still in training mode before a planned March 16 rollout
Context & Ripple Effects
Baidu set its own clock on February 6 when it announced a ChatGPT-style service for search, sending shares up more than 15% in a single session on the announcement alone — before anyone saw the product. Within a day, Alibaba confirmed it was testing a rival ChatGPT-style tool, turning Baidu's reveal into the opening move of a domestic race.
This report, days out from the planned March 16 event, describes the bill coming due: employees working around the clock on a bot that sources say still struggles with basic tasks and remains in training mode. The gap between the February market reaction and the product's actual state is the story — Baidu is racing rivals it helped summon, on a deadline it chose itself.
First-order effects
- Baidu faces a compressed deadline it cannot quietly slip: the March 16 event was priced into the stock after February's surge, so a weak or scripted demo risks an immediate correction from investors who bought the announcement, not the product.
- Robin Li's team must present a half-trained model publicly, trading short-term embarrassment for first-mover positioning against Alibaba and other Chinese entrants already in internal testing.
Second-order effects
- Alibaba and other Chinese rivals get both a benchmark and a warning: Baidu's scramble shows that announcing early locks you into a public readiness contest, likely pushing competitors to time their own reveals against demonstrated capability rather than press releases.
- If the March 16 presentation leans on recorded material rather than live interaction — which is how Baidu ultimately staged the debut with Robin Li talking over a scripted video — the market punishes the gap between hype and hands-on proof, setting the credibility bar every subsequent Chinese LLM launch is judged against.
Third-order effects
- The episode sketches the structure of China's generative-AI race: national champions locked in a version-number arms race where each claims parity or superiority — Baidu later claimed Ernie 3.5 surpassed ChatGPT on general abilities and GPT-4 on many Chinese-language tasks — while adoption, not benchmarks, becomes the durable scoreboard, with Ernie Bot reaching 100M+ users by December 2023 and beyond.
- For enterprise buyers in China, the pattern points toward a small set of domestically sanctioned foundation models competing on distribution — search integration, enterprise clients — rather than raw capability gaps, making platform access the real battleground.
The trend: China's generative-AI race is being paced by announcement-driven deadlines rather than model readiness, forcing incumbents like Baidu to ship publicly and iterate under market scrutiny.