By Song Jiaonan, Tech Reporter
Just one week after putting out the fire over that whole “parents followed Doubao’s advice and only fed their newborn 60ml of formula per feed” saga, Doubao is back in the hot seat.
On June 5, Douyin Group VP Li Liang stepped up to address the viral story about Doubao allegedly misidentifying mushrooms and leading to a poisoning case. He confirmed that the Doubao team has already reached out to the person involved. According to their findings, the user snapped a photo of some wild mushrooms they picked from their residential complex, and Doubao flagged them as “Chicken Leg Mushrooms.” But here’s the catch: Doubao actually included a pretty clear warning right in the response. It explicitly stated that these are super easy to confuse with the highly toxic Chlorophyllum molybdites, which can trigger severe gastrointestinal distress if swallowed. It also strongly advised against eating them if they were foraged outdoors, emphasizing that telling apart wild fungi just from a picture is incredibly risky and that no image-based ID can guarantee you’ve ruled out poisonous lookalikes.
Li Liang didn’t mince words either, stressing that we’re still in the early stages of AI development. While Doubao is constantly getting sharper at recognizing things, when it comes to anything touching personal safety, he’s drawing a hard line: always treat AI outputs as a reference point, not a final answer. Cross-check everything with multiple sources before acting on it. He also threw a practical reminder out there: even if those mushrooms or fruits you pick around your apartment complex aren’t naturally toxic, they might be laced with pesticides or environmental pollutants. Honestly, he strongly recommends leaving them alone.
To back up his points, Li Liang shared a post from Weibo user “Plant Man Shi Jun.” Dropping into the #Don’tTrustDoubaoWithWildMushrooms thread, Shi Jun made a solid point: “Honestly, forget Doubao—even seasoned botanists can totally miss a callout when they’re in unfamiliar territory. Plus, a lot of these toxic mushrooms act like paraquat; once you swallow them, there’s basically zero chance of saving you.” For context, that’s Dr. Shi Jun, a Botany PhD from the Chinese Academy of Sciences and an active member of the Scientific Communication Committee for the Chinese Botanical Society.
For anyone who doesn’t know, “Chicken Leg Mushroom” is the common name for Coprinus comatus. It gets its name because it literally looks like a drumstick, and the flesh pulls apart like shredded chicken. It’s actually a commercially farmed delicacy that’s been gaining serious traction lately, often called the “rising star” of the fungal world. Packed with nutrients and known for its amazing umami flavor, it’s a total crowd-pleaser. Sure, young wild specimens are technically edible, but here’s the kicker: those popping up along city medians or neighborhood lawns tend to soak up heavy metals like lead and cadmium straight from the soil. That’s why experts consistently warn against foraging them in urban areas.
On the flip side, you’ve got the Chlorophyllum molybdites—aka the green-spored parasol. This guy is a notorious killer, showing up as the number one culprit in China’s annual mushroom poisoning stats. When they’re young, they’re stark white with thick stems and plain caps, making them look shockingly similar to those edible chicken leg mushrooms. For the average person trying to tell them apart with just naked eyes? It’s practically impossible. Countless ER visits happen every year because folks mistake this deadly twin for a tasty dinner.
After hearing out Li Liang’s official take, Dr. Shi Jun chimed in with a straightforward note: “Doubao’s reach is massive now. If you’re going to handle food safety queries, you absolutely need to push out clear, unmissable warnings.”
It’s worth remembering that earlier, back on May 28, news broke about a pair of new parents in Nanning, Guangxi. They blindly trusted Doubao’s tips and only gave their month-old baby 60ml of milk per feeding. Naturally, the kid was fussing constantly and wasn’t gaining any weight. A hospital visit put a pin in that theory: doctors immediately flagged the feeding schedule as dangerously low, explaining that healthy full-term, one-month-old babies typically need between 80ml and 100ml per session. Sixty milliliters just isn’t cutting it for proper growth.
In response, Doubao quickly released a statement brushing off the coverage as “misinformation.” Their team pointed out that after running multiple rounds of internal testing, the model definitely wouldn’t spit out that specific 60ml-per-feed recommendation under normal circumstances. In fact, Doubao’s standard protocol is to provide a daily total volume, remind parents to watch for the baby’s cues, and suggest bumping up feeds or calling a pediatrician if crying spells become frequent. And interestingly enough, other major LLMs would have given virtually the exact same guidance.
Doubao also got in touch with the treating hospitals and the doctors who spoke to the press. Here’s what the medical staff clarified: a parent had brought in a jaundiced infant for a checkup. During the consultation, the family mentioned that Doubao had suggested “60ml per feed.” However, they couldn’t produce the actual chat logs, nor could they confirm whether Doubao had ever outlined a daily total amount or provided broader context. The attending physician noted that without knowing the exact prompt history or conversation flow, it’s nearly impossible to reconstruct what really went down. That missing context leaves plenty of room for misunderstanding.
Alongside all that, Doubao reiterated its commitment to tightening accuracy across specialized verticals and beefing up its safety warning systems. They also drove home the golden rule for AI health queries: generated content is strictly for informational purposes. It’s never a substitute for real medical diagnosis or treatment. When it comes to your health, trust the doctor, not the algorithm.
The reason this keeps making headlines? Doubao is firmly planted in the top tier of China’s domestic LLM scene right now. Millions of users default to it for everyday questions, and unfortunately, that includes plenty of high-stakes, safety-critical scenarios.
The numbers backing this up are staggering. QuestMobile’s latest report shows Doubao hit 345 million monthly active users in Q1 2026. That’s miles ahead of runner-up Qwen (166 million) and third-place DeepSeek (127 million). To put that dominance into perspective, Doubao’s MAU actually equals the combined user base of ranks two through five.
Adding fuel to the fire, Tan Dai, President of ByteDance’s Volcengine division, dropped some heavy metrics last March. Doubao’s daily token consumption skyrocketed past 120 trillion—a complete double in just three months, and a mind-blowing 1,000x jump since its initial May 2024 launch.
But let’s keep it real: scale doesn’t magically fix the hallucination problem. Whether it’s Doubao or any other major player, AI still trips over itself when asked tricky questions. A recent study published in BMJ Open actually tested five top-tier AI chatbots on evidence-based medical queries, and half of their answers came out fundamentally flawed. Some models were racking up error rates as high as 58%. Yeah, it’s a messy space.
Trying to pivot toward more specialized use cases, Doubao is quietly gearing up to roll out a Pro version. This premium tier is slated to cover heavy-hitting services like software dev, advanced data analytics, professional design workflows, automation scripting, financial modeling, and scientific research. Right now, it’s still locked behind closed doors for beta testing.
If you want to peek at the price tags, Doubao laid out three tiers. The Standard plan runs 68 RMB monthly (or 688 RMB annually), the Enhanced tier jumps to 200 RMB a month (2,048 RMB yearly), and the newly minted Pro edition sits at 500 RMB monthly (5,088 RMB yearly).
But here’s the plot twist: data released Wednesday by global AI tracker Aicpb.com shows that right after Doubao rolled out its paywall, its May MAU took a nosedive, shedding a cool 6.1 million users overnight.
And it’s not just noise. Aicpb.com founder Li Bangzhu weighed in with a sharp assessment: “The era of free AI in China is nowhere near over. So yeah, Doubao pushing monetization this aggressively might just be jumping the gun.”
