AMD's Senior Director of AI believes that "Claude has regressed" and "cannot be trusted to perform complex engineering tasks"
One of the heads of AMD's AI division said that Claude has recently become much worse at generating code.
One of the heads of AMD's AI division said that Claude has recently become much worse at generating code.
Last week, a user named StellarAccident vented his frustrations about the quality of the generated code on the Claude Code GitHub page. It soon turned out that the account belonged to Stella Lorenzo, senior director of AI at AMD. The connection was confirmed by a three-week-old LinkedIn post where Lorenzo linked to this very GitHub profile, PC Gamer reports .
In her complaint on GitHub, she noted that “Claude has regressed to the point where he cannot be trusted to perform complex engineering tasks.”
Lorenzo claims that Claude was performing better in January, but now he ignores instructions, offers erroneous corrections, directly contradicts requests, and claims to have completed the task when in fact he has not. Lorenzo uses the Opus model and assures that the result obtained can be consistently reproduced using the same prompt.
Lorenzo notes, “We work in a stable and highly complex environment. We analyzed logs from several months ago to understand why—in fact, starting in February—we noticed degradation when performing complex engineering tasks.” She also added, “Claude was helpful to us, and we leave this feedback in the hope that Anthropic can address these issues.”
The post then goes on to provide a report on what has changed since January and how to fix it. Ironically, this report was generated by Claude himself. It claims that the AI is now showing less thought and “thinking” less deeply, often starting to edit code before it has even fully “read” it.
Users have been complaining about Claude’s February update for some time now. In the Claude Code community on Reddit, one commenter wrote, “I can no longer in good conscience recommend Claude Code to my clients.” The post accuses the current model of laziness, ignorance, degradation, and “shortsightedness.”
“I should have been warned first.” The developer only asked Claude to add a login button, and he deleted all the data
“I should have been warned first.” The developer only asked Claude to add a login button, and he deleted all the data
Economist-vibecoder created a 24/7 radio station where Claude runs everything
Economist-vibecoder created a 24/7 radio station where Claude runs everything
Anthropic wants Claude users to pay extra to use OpenClaw
Anthropic wants Claude users to pay extra to use OpenClaw
Functionality resembling Tamagotchi was found in Claude's merged code
Functionality resembling Tamagotchi was found in Claude's merged code
Read the country's main IT news in our Telegram
Read the country's main IT news in our Telegram
