ChatGPT-ийн алдааг харуулсан бичлэгүүд нь цахим орчинд түгсэний дараа хиймэл оюуны томоохон компаниуд контент бүтээгчтэй хамтрах хүсэлт илэрхийлсэн байна.
Мэтт Хаск лхагва гарагт болсон WIRED World Fair арга хэмжээний үеэр OpenAI, Anthropic, xAI зэрэг компани түүнтэй хамтран ажиллах санал тавьсан гэж ярьжээ. Түүний хэлснээр, xAI зөвхөн тус компанид зориулж бичлэг хийх нөхцөлтэйгөөр Tesla автомашиныг долоо хоног ашиглуулах санал хүртэл тавьсан байна.
Хаск ChatGPT-ийн дуу хоолойн загвараас 100 хүртэл тоолох, битбокс хийх зэрэг хүсэлт тавьж, хариултыг нь бичлэг болгон нийтэлдэг. Тэрээр Instagram, TikTok, YouTube-д нийлээд хоёр сая орчим дагагчтай болж, Unicorn агентлагтай гэрээ байгуулжээ. Бичлэгүүдэд ChatGPT даалгаврыг буруу гүйцэтгэх нь олонтаа; тухайлбал, нэг милийн гүйлтийг хэдхэн секундэд дуусгасныг нь ойролцоогоор 10 минут гэж тооцсон байна. Сэм Олтман дөрөвдүгээр сард энэ мэт цаг хэмжих алдаа нь танигдсан асуудал бөгөөд дахин нэг жил үргэлжилж болзошгүйг подкастад ярьжээ.
Хаскийн ярьснаар, OpenAI алдааг зассан хувилбарт эрт нэвтрэх эрхийг өөрийн түүхээ хуваалцах нөхцөлтэй санал болгосон ч тэр татгалзсан байна. Мөн тус компани хамтарсан бичлэг хийх санал болон бэлэг илгээх хүсэлт тавьж, гэрийнх нь хаягийг асуухад Хаск зөвшөөрөөгүй гэжээ. OpenAI мягмар гарагт ахуйн ажил, хүүхдээ цэцэрлэгт бүртгүүлэх зэрэгт туслах чадварыг үнэлэх жишиг үзүүлэлтэд загвар нь өндөр оноо авсныг зарласан ч Хаск хиймэл оюуныг талх худалдаж авахад туслуулж туршжээ: зарын 10 мянган ам.долларын үнийг 5 ам.доллар болгохыг хүсэхэд загвар 400 ам.долларын саналыг зөвшөөрсөн байна.
Хаск чатботуудыг олон нийтэд ёжилж эхлээд жил орчим болжээ. ChatGPT-д цөмийн бөмбөг дэлбэрсэн гэж хэлсэн бичлэг нь маргааш өглөө гэхэд 100 мянган үзэлт авч, улмаар энэ төрлийн бичлэгээ үргэлжлүүлэхэд хүргэсэн гэж тэр ярьсан байна. Түүнээс хойш CeraVe, Ninja брэндтэй хамтарч, The Onion-той хамт Zoom-ээр пантомим зэрэг ур чадвар сурах цуврал хийхээр болжээ. Бичлэгийн санаа хайхдаа хиймэл оюунтай ярьдаг ч санаа гаргуулахад ашигладаггүй, учир нь үр дүн нь хоорондоо хэт төстэй болох магадлалтай гэж Хаск хэлсэн байна.
Дэлгэрэнгүйг эх сурвалжаас харах
↓Эх сурвалжийг нээх ↓
OpenAI, Anthropic, and xAI have all asked creator Matt Husk to collaborate following the viral success of his videos trolling AI models.
During the WIRED World Fair on Wednesday, Husk told the crowd “all the big AI companies” have inquired about working with him, including Elon Musk’s xAI, which Husk claimed offered a Tesla for a week if he made videos exclusively for the firm.
The comedian has gained some 2 million followers combined across Instagram, TikTok, and YouTube, as well as a deal with the talent agency Unicorn, through videoing himself making deadpan requests to ChatGPT’s voice model—asking it to count to 100, or to beatbox, for example. His viral videos have exposed the weaknesses of OpenAI’s LLM, which comically botches the tasks and swings between sycophancy and gaslighting Husk, and earned him the title of “Sam Altman’s worst nightmare.”
In April, Altman, nodding and smiling, admitted on tech podcast Mostly Human that the model’s inability to set an accurate timer was a “known issue” and would be for another year, after being shown a Husk video in which he asked ChatGPT to time him running a mile and found that it measured a span of a few seconds as lasting “around 10 minutes.” When Husk says “I swear I was faster than that,” it replies with a sympathetic tone: “It can feel faster in the moment.”
During his interview at WIRED World Fair with Shell Game podcast host Evan Ratliff, Husk said OpenAI offered him early access to a patch for the timer issue if he shared his account. He declined. The company had previously emailed Husk asking to collaborate on a video. “They offered me the smallest amount,” he said. “I was just like, ‘No.’” The firm then offered to send him gifts and asked for his home address; Again, Husk declined.
OpenAI, Anthropic, and xAI (which was recently rolled into SpaceX to create SpaceXAI and then SpaceXSI after President Trump renamed artificial intelligence “super intelligence”) did not respond to requests for comment.
AI labs are increasingly pitching their products as capable of handling everyday chores. On Tuesday, OpenAI announced its model had scored “very highly” on a benchmark built to model its ability to help people in their personal lives, such as meal prep and enrolling a child in kindergarten. “The way they promote it is like ‘It can do anything,’” Husk said, “so I’ll put it to the test—see if it can order bread for me.” In one of his videos, Husk asks AI to help him negotiate some bread on Craigslist down from a starting price of $10,000 to $5. The model accepts a price of $400.
Husk began publicly trolling chatbots about a year ago. He told Ratliff, a WIRED contributor, at the time he liked to make TikToks asking people questions, but there weren’t many around while he was working his landscaping job, and he ran out of ideas. So he recorded himself telling ChatGPT a nuclear bomb had gone off. The model’s response, “I’m rooting for you,” and advice to take shelter “was not helpful at all,” he said. “I thought it was a pretty bad video,” he told Ratliff, “and the next morning I had 100,00 views, and I was like, ‘Let’s tell it I got hit by a train.’”
Since then, Husk has secured brand partnerships with CeraVe and Ninja. He also has an upcoming original series with The Onion, where he learns complicated skills, like being a mime, over Zoom. “I’m spending too much time talking to AI,” Husk explained, “so I’m going to branch out and learn stuff from humans.”
While Husk finds himself talking to AI a lot to scope out ideas for videos, “I don’t really use it in that creative sense,” such as to generate ideas, he said, because they would all be “too similar.” “I’d rather just have my worst idea at least be my idea rather than AI’s idea for a video,” he said.
