Shulou(Shulou.com)11/24 Report--

Original title: ChatGPT, really fragrant! Google's top AI talent group defected OpenAI

According to The Information, OpenAI has poached at least a dozen Google AI employees in recent months, and these engineers have played a crucial role in ChatGPT research.

Now no one can deny that OpenAI has become a global fire with ChatGPT.

All AI practitioners and investors want to know how a small, seven-year-old startup can beat Google.

In fact, one of OpenAI's "big killers" is to dig people from Google.

Now that ChatGPT has become the top stream, Google's talents have begun to take the initiative to jump to OpenAI.

Today, Jason Wei and Hyung Won Chuang, two former Google employees, seem to have made an appointment, announcing their departure from Google Brain, joining OpenAI, and tweeting each other.

Only Google wounded the world reached.

OpenAI's secret weapon: former Google engineer In fact, for ChatGPT, OpenAI started "digging" work very early.

A few weeks before the public launch of ChatGPT, OpenAI quietly poached at least five employees from Google to put the finishing touches so ChatGPT could be released in November.

In OpenAI's official blog, Barret Zoph, Liam Fedus, Luke Metz, Rapha Gontijo Lopes, the names of former Google employees, are listed as major contributors to ChatGPT.

Top: Jacob Devlin, Liam Fedus and Shane Gu; bottom: Rapha Gontijo Lopes, Samuel Schoenholz and Barret Zoph Last month, OpenAI hired another researcher from Google to develop machine learning models for search engines. OpenAI then unsurprisingly applied this technology to its own ChatGPT.

More recently, as we saw at the beginning, Google's main AI team, Google Brain researchers, have jumped ship to OpenAI. Google Brain lost at least four of its core members.

Bard press conference a big fool, employees have time to complain about not careful, fool things, Google brain core employees have "defected", really should be that sentence: when people are unlucky, drink cold water are broken teeth.

But in fact, Google can come to this step, is inevitable.

After absorbing most of the world's machine learning talent, Google has been preempted by OpenAI.

Google can only rush to catch up with OpenAI and launch AI-centric products to the public.

Objectively speaking, Google's approach is deliberate and responsible. As an influential tech giant, Google is always on guard against high risks and fears that nascent technology will go wrong. (One reason, of course, is that chatbots will cost more to provide human-like answers than classic search.)

But business is like a battlefield. It doesn't give you the chance to hesitate.

Some of Google's early AI technologies have been unwittingly transformed by OpenAI into new revenue-generating services, including chatbots and AI that generates images and videos from text.

Google's talent is also slowly losing to such small start-ups.

Recently, two Google brain researchers said that many employees on the team think Google's current product plans are overly cautious and full of red tape. In this environment, engineers have to face the frustration that new technologies they have worked hard to develop have never been adopted.

Some employees are frustrated because they have been suggesting Chat features for years and haven't received feedback.

So they decided to leave and look for opportunities elsewhere, like OpenAI.

In fact, Google in the beginning of the establishment, it is also from the factory poaching people.

In the late 1990s, Digital Equipment, an established technology company at the time, faced problems commercializing academic research. Google took this opportunity to get the current AI handle-Jeff Dean.

Of course, Google is not sitting still now, and soon launched Bard, and ChatGPT positive hard.

The rest of the story is well known. Bard flipped the car right off the bat-giving the wrong answer to a question about the James Webb Space Telescope.

Google does predict accurately that AI chatbots can't avoid nonsense.

But since sooner or later to go to this step,"reputation risk" can only be ignored for the time being, why so cautious at the beginning, was OpenAI and Microsoft robbed first?

Google is so wrong.

Google's new and old employees are complaining that the company's technological innovation transformation of employees is too slow.

But Google's caution is based on antecedents.

Industry observers have long warned that AI can generate false images, misinformation, discriminatory comments, and therefore bad social impact.

For example, in 2015, image recognition AI in Google Photos labeled some black people as "gorillas."

Two years later, a team at Google Brain published that famous paper, proposing a stunning new method of machine learning, Transformer, and laying the foundation for OpenAI to create ChatGPT.

Oddly enough, while other companies have used Google's papers to create their own Transformer, AI chatbots, or text-generated image models, Google has never launched a similar product.

Four years after launching the Transformer revolution, Google claims to have created LaMDA, a Transformer-based language model that can understand and generate conversations with humans.

But Google has been hesitant to publicly launch LaMDA because of concerns that the technology is not accurate enough.

A former Google Brain employee said they believe managers in Google's AI research division deprived some teams of computing resources to train AI models like OpenAI.

One current Google employee said that if anyone wanted to launch a new AI product, they had to overcome bureaucratic obstacles, so in the end, everyone gave up and stayed on the old track.

In contrast, Microsoft CEO Nadella and OpenAI CEO Sam Altman have reached an agreement to allow OpenAI to train AI in Microsoft's cloud for free.

Google is very cautious about all Transformer-driven technologies.

It is said that when OpenAI released its text-generating image model Dall-E 2 last year, Google already had two models with similar capabilities.

But when the images generated by Dall-E 2 began to go viral online, Google remained indifferent to its own models, unwilling to share them with the public, even with safeguards, because Google feared they might be misused, according to people familiar with the matter.

At the end of last year, Google employees still had to apply to use the company's image-generation technology.

Google modified the technology to not generate faces to prevent misuse. Internal users of the technology must also sign a disclaimer exempting Google from liability for content generated by the system.

It was revealed that some employees on the team that created Imagen, one of Google's two image-generating AI tools, had left.

Imagen co-author Mohammad Norouzi is said to have led a startup that recently raised funding at a valuation of more than $100 million, although the team has no clear idea what it wants to launch.

Within weeks of ChatGPT's launch, Google's leadership told employees at an all-hands meeting that while the company had similar technology, the size of Google's product (billions of users) meant it had to roll it out more cautiously than startups like OpenAI.

However, when Google saw ChatGPT flourish, it immediately issued a red code and introduced a "green channel" to shorten the process of assessing and mitigating potential hazards.

Unfortunately, it was too late.

One by one, they all went away. With ChatGPT's momentum, OpenAI continues to attract researchers at Google.

Jacob Devlin jumped ship to OpenAI in January. He helped Google create its famous machine learning model, Bidirectional Encoder Representatives from Transformers (BERT).

Shane Gu, David Dohan, Alexandre Passos and Samuel Schoenholz have also recently left Google to join OpenAI, according to social media profiles and people familiar with the matter.

Alexandre Passos is not only a senior software engineer at Google Brain, but has published numerous papers on machine learning and natural language processing.

"Scikit-learn: Machine learning in Python" has been quoted nearly 70,000 times.

However, the loss of a few people has little impact on Google's brain, which has 800 of the world's top talents.

In addition, it is also normal for star employees to leave large factories and set up their own doors. There are a number of AI start-ups that are the work of former Google Brain employees. For example, develop chatbot characters.

However, under pressure from rival OpenAI, Google's leadership is increasingly inclined to deliver new products faster.

Since the end of last year, as Google tries to further increase the speed of product launch, the demand for engineers in Google Brain has also increased, and the pace of work has accelerated significantly.

The result of a combination of factors is that Google, which has always been cautious, instantly wiped out hundreds of billions of dollars in market value due to an obvious factual error in Bard's presentation.

Google-the "Whampoa Military Academy" in history "Whampoa Military Academy in Silicon Valley", Google really deserves this title.

Over the past year or so, Google has seen a number of big names jump to more agile startups, such as OpenAI and Stable Diffusion.

Character.AI, Cohere, Adept, Inflection.AI, Inworld AI, and many others are startups built around large-scale language models, all of which are the work of Google's top AI researchers.

There are also search startups that use similar models to develop chat interfaces, such as Neeva, run by former Google executive Sridhar Ramaswamy.

Among them, Noam Shazeer, founder of Character.AI, and Aidan Gomez, co-founder of Cohere, are key figures in developing Transformer and other core machine learning architectures.

David Ha, a prominent research scientist, tweeted: "If Google doesn't perk up and start releasing its own AI products, it will go down in history as the Whampoa Military Academy that trained an entire generation of machine learning researchers and engineers."

The tycoon also left Google Brain in 2022 to join the star startup Stable Diffusion.

References:

https://www.theinformation.com/articles/openais-hidden-weapon-ex-google-engineers? rc=epv9gi

https://www.businessinsider.com/chatgpt-openai-google-ai-employees-hired-report-2023-2

This article comes from Weixin Official Accounts: Xinzhiyuan (ID: AI_era)

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.