Florian Gaertner/Photothek via Getty Images/Reuters
- OpenAI says it's retiring several ChatGPT models, including the much-loved GPT-4o.
- Many users developed a strong attachment to 4o's personality and relied on it for emotional advice.
- OpenAI's Fidji Simo said newer models have more guardrails to prevent "bad attachments."
OpenAI said that today — once and for all — it is retiring GPT-4o. For some, it's like being dumped the day before Valentine's Day.
Many ChatGPT users have a strong attachment to 4o, which is known for its sometimes sycophantic conversation style. Users who relied on the model as an emotional crutch and creative partner have described it as a "vital accessibility aid."
OpenAI first tried to deprecate the model in August, only to reverse its decision 24 hours later after an enormous backlash.
The company gave users another heads-up in January, but — as it turns out — time does not heal all wounds. The backlash is back.
"Rot in hell," one user wrote on X in response to OpenAI's announcement on Thursday.
"Are you going to cover our bereavement leave from work?" another asked.
Fidji Simo, OpenAI's CEO of Applications, said earlier this week that these strong attachments to 4o marked the start of a new era — a time when users develop AI-based relationships.
"Humans are built to develop attachments to intelligent things," she told Alex Heath on his Access podcast. "And AI is getting pretty intelligent."
Those relationships can get ethically murky, however, when users are asking ChatGPT for advice like "Should I leave my wife?" Newer models, Simo said, have guardrails to prevent "bad attachments." She said that newer models will tell users that it's "not their place" to tell them to stay married or not, and instead talk them through the pros and cons.
"We know that losing access to GPT‑4o will feel frustrating for some users, and we didn't make this decision lightly," OpenAI said in a blog post on Thursday announcing the retirement of 4o, as well as GPT‑4.1, GPT‑4.1 mini, and OpenAI o4-mini. "Retiring models is never easy, but it allows us to focus on improving the models most people use today."
OpenAI added that only 0.1% of its users were still using 4o.
OpenAI CEO Sam Altman has also acknowledged that users are especially attached to 4o's ingratiating responses.
"As we've been making those changes and talking to users about it, it's so sad to hear users say, 'Please can I have it back? I've never had anyone in my life be supportive of me. I never had a parent tell me I was doing a good job," Altman said on the "Huge Conversations" podcast in August after OpenAI first tried to kill 4o.
At the time, Altman said 4o's approach was "too sycophant-y and annoying," and fixes were imminent.
The fix is in, but perhaps at a cost.
"People are in absolute crisis because the companion they've collaborated with for months is being wiped with ZERO recourse for the average user," an X user wrote on Thursday.
from Business Insider https://ift.tt/6N2Or5e
No comments:
Post a Comment