ChatGPT의 짧은 역사
- 2025-10-25
패권: 누가 AI 전쟁의 승자가 될 것인가 14장 중 일부 내용을 발췌/요약.
Chapter 14. A Vague Sense of Doom
ChatGPT 출시:
The idea of generating fully formed content was what made Altman’s next move even more sensational. GPT-1 had been more like an autocomplete tool that continued what a human started typing. But GPT-3 and its latest upgrade, GPT-3.5, created brand-new prose, just like how DALL-E 2 made images from scratch.
As the world gawked at DALL-E 2, rumors swirled that rival Anthropic was working on a chatbot, sparking the competitive juices at OpenAI. In early November 2022, OpenAI managers told staff that they were going to launch a chatbot of their own in just a few weeks, that was built on GPT-3.5. …
This woudn’t be a product launch, OpenAI’s leadership assured staff, but a “low-key research preview.” Still, some employees said they weren’t comfortable releasing the tool so quickly. They didn’t know how the public might misuse a language model that was so fluent and capable. …
Altman pushed to launch. … In a way, OpenAI was doing the world a favor and getting it ready for OpenAI’s more powerful, upcoming model, GPT-4. … On its website, OpenAI called this its “iterative deployment” philosophy, releasing products into the wild to better study their safety and impact. …
On November 30, 2022, OpenAI published a blog post announcing a public demo of ChatGPT.
대중들의 폭발적 반응:
Most of the public hadn’t heard of OpenAI, never mind GPT-3. And no one, including researchers at OpenAI, knew what would happen when they let anyone test its capabilities. …
Within the next few hours, their reviews started popping up on Twitter. … It was hard to find a single negative appraisal of ChatGPT. …
Over the next few days, people flooded Twitter with screenshots of ChatGPT’s poems, raps, sitcom scenes, and emails. The more offbeat, the better. …
After two months, ChatGPT had attracted thirty million registered users, making it one of the fastest-growing online services in history. By early 2024, around on hundres million people were using ChatGPT weekly.
Claude와 GPT-4:
On March 14, 2023, the very same day that Anthropic had finally released its own chatbot called Claude, OpenAI launched its upgrade, GPT-4.
내부의 우려:
But a divide was also growing between OpenAI employees who were focused on product development and those focused on safety, who were struggling to monitor the soaring incoming traffic on ChatGPT for abusive queries. Believing they were taking significant steps toward AGI, Ilya Sutskever began working more closely with the company’s safety team.
구글의 반응:
Inside Google, executives recognized that more and more people might just go to ChatGPT for information about health issues or product advice - among the most lucrative search engine terms to sell ads against - instead of Google. …
More than 90 percent of online searches around the world happened on Google. But now, for the first time, Google’s more-than-twenty-year dominance as gatekeeper to the web was on shaky ground. …
Google had for years been taking a cautious, almost fearful approach to new technology. It “didn’t move” unless something was a billion-dollar business, and it certainly didn’t want to mess with its own ads business that made nearly $260 billion a year. …
But now Google executives didn’t have much choice. …
DeepMind와 Google Brain의 병합, 그리고 Gemini:
OpenAI’s big splash also called into question DeepMind’s work over the past thirteen years. And it rattled Hassabis. Weeks after ChatGPT’s release, he told staff in an all-hands meeting that DeepMind shouldn’t become the “Bell Labs of AI,” a place that invented everything but saw its ideas commercialized by others, a former employee remembers. …
Now he had to follow in Altman’s footsteps. Google executives told DeepMind to start working on a series of LLMs that would be even better than LaMDA. They called the new system Gemini, and DeepMind imbued it with the strategic planning techniques that AlphaGo had developed.
To help move things along more quickly, Pichai made another drastic move. He merged the two rival AI divisions, DeepMind and Google Brain, and called them Google DeepMind. …
To the surprise of many, Pichai picked Hassabis to run the combined unit. Jeff Dean, Google’s most revered engineer who oversaw AI research at the rest of the company, had seemed like the more likely candidate. …
When Hassabis announced the merger with Google Brain to DeepMind staffers, he told them in an email that the units were joining forces because AGI had the potential to “drive one of the greatest social, economic and scientific transformations in history.”
In reality, they were merging to help a panicked Google beat a business rival, just as OpenAI’s mission to benefit humanity (without “financial pressure”) had shifted toward serving the interests of Microsoft.