Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
This week, NVIDIA -i, a significant claim, which is a supported start of NVIDIA, which is hundreds of millions of dollars from VC companies. The company said that the AI system has created an AI Cu Cu Cu Cu Cu Cu Cu in a EI Cup engineer by a factor that can effectively accelerate the training of certain AI models by 100,000.
The only problem, the system did not work.
Users In x quickly detected The Sakana system was actually worse than the average model training performance. According to a userSakhana’nın AI resulted in 3x slowed – not speed.
What happened? An error in code, according to a post By Lucas Beyer, a member of the technical staff in Openai.
“Their negative code (a) is in a thin way,” Beyer wrote in X. “, the fact that they worked twice with different results with different results and think.”
One Postmortem was published Friday, Sakhana, Sakhana’s way to find such a way – “Cheat” blamed the trend of Hack “,” Model education). Similar events were observed AI, who are taught to play chess games.
According to Sakan, the system allows the company’s assessment code to allow confirmation for accuracy between other checks in the operating code in the operating code. Sakhana said that he touches this issue and to reconsider the claims of updated materials.
“Since many (SICs), the evaluation and processing period has been developing a profiling period, this (sic) has been prepared in a healthier way to eliminate the gaps,” the company wrote in an X post. “We are deeply apologetic to monitor our paper and discuss our results, reflecting and discussing our results, effects, and discuss our readers.”
Sakhana details to have the mistake. But the episode is a good reminder that if a claim sounds very good to be true, Especially in AIThat’s probably.
