Sam Altman, CEO of OpenAI, made an announcement that raised expectations in regards to the ‘OpenAI o1’ that was released the day before. He said that o1 is a giant breakthrough in the event of artificial intelligence (AI) technology, however the experience for general users will likely be limited.
Altman CEO said on X (Twitter) on the thirteenth (local time), “o1 is probably the most capable and aligned model we have ever built,” but “it still has flaws and limitations.” He also explained, “The impression I got once I first used it was stronger than after spending numerous time using it.”
Also, to the query, “Is it true that AGI was born today?” he simply answered “No.”
This appears to be as a result of the excessive expectations that emerged locally immediately after the discharge of o1.
“The AI community has been excited for the reason that launch of o1,” Kian Katanforusi, Workear’s CEO and an adjunct lecturer at Stanford University, told TechCrunch. “But for those who take OpenAI’s modeling techniques and mix them with o1’s reinforcement learning algorithms for training, you may technically create step-by-step pondering and step back and see the massive picture of what the AI model is trying to resolve.”
Rohan Pandey, a research engineer at Reworked, also said that “the hype surrounding OpenAI is so great that even the corporate can’t stop it.” He said he hopes o1’s inference ability will surpass GPT-4’s, but that it doesn’t feel as revolutionary as GPT-4’s emergence.
As they indicate, the response on social media is generally certainly one of anticipation. Altman’s tweet stating that it just isn’t AGI was met with comments similar to “Then it’s baby AGI” and “Anthropic will likely be incomparable now.”
CEO Altman’s remarks also appear to caution against excessive expectations.
To begin with, this model tends to think too deeply, which implies that the prevailing ‘GPT-4o’ is more suitable for many queries. One X user said that when he asked o1 to write down sentences in keeping with a selected pattern, it took 92 seconds to think.
One other burden is that the price is 4 times that of GPT-4o. As well as, o1 is just a ‘preview’ without multimodal functions.
So OpenAI also explained, “For applications that require image input, function calls, or consistently fast response times, GPT-4o and GPT-4o mini will proceed to be the best selection,” “Nonetheless, if you need to develop applications that require deep inference and might accommodate longer response times, the o1 model could possibly be an awesome selection.”
Meanwhile, some developers are quickly testing to seek out tasks which are optimized for o1.
In line with VentureBeat, o1 has been shown to excel in ▲planning and writing white papers ▲organizing infrastructure and writing risk assessment reports ▲quickly creating apps and games ▲writing requests for proposals, etc.
There may be also a response that OpenAI is powerful in coding, mathematics, and logical pondering, which it touts as its strengths.
Reporter Im Dae-jun ydj@aitimes.com
