Google researcher blasted GPT startups for their hype and sent out 18 tweets in anger!

In the field of AI, the traffic king last year was none other than generative AI.

From DALL-E 2 to ChatGPT, from Stable Diffusion to Midjourney, the AIGC industry has shown explosive growth. ​

The proliferation of generative tools based on artificial intelligence has dazzled investment companies and the general public.

Top investment firm Sequoia Capital wrote in a blog post: "Generative AI is not only becoming faster and cheaper, but in some cases can create more artificial intelligence than humans can. Better." "It's easy to imagine that in a few decades, generative artificial intelligence will be deeply ingrained in the way we work, create and play."

Investment Company Enthusiasm for AIGC can be measured in dollars, with 78 deals in the space estimated at $1.37 billion in 2022 alone.

However, while the creators and investors of AI models believe they are a force for change on the planet, not everyone in the field is convinced these generative machines are the best option.

Google researcher: Hype cannot save GPT

"The current atmosphere of artificial intelligence has many similarities with the web3 boom in 2021, which makes I feel uncomfortable,” François Chollet, a Google deep learning research expert and creator of the deep learning system Keras, issued a strong warning on Twitter regarding the current investment boom. "People believe narratives that do not tell the truth."

##François Chollet

"Everyone firmly believes that within the next 2-3 years, GPT will have a 'civilization-changing' impact (and bring a 100-fold return on investment)," he continued. "Personally, I think there are two situations: bull market and bear market for the development of LLM."

He believes that in the bull market situation, "generative AI becomes the same as most technology products." A broad paradigm of interaction". Even in this case, artificial general intelligence (AGI) remains a "pipe dream."

At the same time, in bear market conditions, large language models like GPT-3 will find "limited success stories in SEO (search engine optimization), marketing and copywriting fields" , and ultimately proved to be a "complete bubble."

Chollet believes that the final development of LLM is likely to be somewhere in between the two, leaning towards the latter. "Artificial intelligence as a universal interface for our information" is something that will definitely happen in the future, but this generation of technology is not yet able to fully achieve this.

An important criterion for the success of the LLM model is commercialization. If LLM can generate huge economic returns, then the commercialization of this technology will be successful.

Taking OpenAI as an example, in 2021, the company’s profits will be approximately 500 to 10 million US dollars; by 2022, this number will become 3000 to 40 million US dollars. . Even as powerful as OpenAI, only the commercial value of image generation technology has been recognized by the market.

Chollet said that he likes to search the most popular ChatGPT tweets on Twitter to gain a deeper understanding of related use cases of this technology. After browsing a large number of tweets, Chollet found that 80% of the posts were about how to attract traffic, and even worse, they were just gimmicks to gain clicks.

Whether it is "10 Secrets to Pay to Unlock ChatGPT" or various ChatGPT training courses, the emergence of LLM has brought great changes to the commercial model of traffic monetization. Comes a disruptive impact.

However, Chollet believes that the actual potential of ChatGPT is much more than this. It could shine in consumer goods, education and search.

Whatever the future holds, people will know soon enough. Billions of dollars are rushing into this track to apply ChatGPT or similar technology to a large number of products. By the end of the year, people will have enough data to make a judgment.

Of course, Chollet also said that hype aside, deep learning can "build a lot of cool stuff." It was like this five years ago, it is like this today, and it will still be like this five years from now. Even without the halo given to it by hypemen, this technology is still very valuable.

web3 cannot be compared with LLM. After all, web3 is pure gimmick, while LLM is a real technology with practical applications. The hype Chollet refers to is a bubble that forms among the venture capital crowd.

At the end of the article, Chollet explained how GPT technology is "marketed successfully." "The driving force for investing in GPT is not experimental data or income statements, but pure hype and unfounded narratives. They form a self-consistent cycle: hype attracts investment, and increased investment leads to more hype, which in turn leads to more More capital is pouring in."

The most important thing is that if a lie is repeated a hundred times, it becomes common sense. The "big pie" without any data support has become a self-evident standard through word of mouth.

Marcus: Don’t believe too much in GPT

Chollet is not the only one who is cautious about GPT. Marcus, a professor at New York University, also often Pour cold water on the enthusiasm of the world.

In an interview, Marcus said that although ChatGPT seems to know everything, it is also prone to errors. ChatGPT is the same as before, and the related system "is still unreliable, still doesn't understand the real world, still doesn't understand the psychological world, and is still full of errors."

Therefore, despite the AI ​​community's response to GPT-4 Although he was full of joy for his arrival, Marcus gave 7 less positive predictions.

#1. GPT-4 will still make all kinds of stupid mistakes like its predecessors. It may sometimes complete a given task well, and sometimes it may not work, but you cannot predict in advance which situation is about to occur.

#2. GPT-4’s reasoning in physics, psychology and mathematics is still unreliable. It may be able to solve some projects that have not been successfully challenged before, but it is still helpless when facing longer and more complex scenarios.

For example, when asked a medical question, it either refuses to answer or occasionally utters nonsense that sounds reasonable but is dangerous. Although it has devoured a huge amount of content on the internet, it is not trustworthy and complete enough to provide reliable medical advice.

#3. Fluent hallucinations will remain common and easily induced. That said, large language models are still a tool that can easily be used to create information that sounds reasonable but is completely wrong.

#4. GPT-4’s natural language output still cannot be served to downstream programs in a reliable way. Developers who use it to build virtual assistants will find themselves unable to reliably map user language to user intent.

#5. GPT-4 itself will not be a general artificial intelligence that can solve any task. Without external assistance, it can neither defeat Meta’s Cicero in Diplomacy; nor can it reliably drive a car; nor can it drive “Optimus Prime” in “Transformers”, or like in “The Jetsons” "Rosie" is so versatile.

# 6. The “connection” between “what humans want” and “what machines do” is still a critical and unresolved issue. GPT-4 will still have no control over its output, some advice will be surprisingly bad, and examples of cover-up bias will be discovered within days or months.

7. When AGI (Artificial General Intelligence) is realized, large language models like GPT-4 may become part of the final solution, but only part of it. Mere "expansion", that is, building a larger model until it absorbs the entire Internet, will prove useful to a certain extent. But general artificial intelligence that is trustworthy and consistent with human values ​​will definitely come from more structured systems. It will have more built-in knowledge and include explicit reasoning and planning tools. These are all lacking in the current GPT system.

Marcus believes that within a decade, perhaps less, the focus of artificial intelligence will shift from scaling up large language models to integrating with a broader range of technologies.

Cool stuff is always fun, but that doesn’t mean it can lead us toward believable general artificial intelligence.

Investors: But it’s fun

But even so, even if experts in the field of AI earnestly advise investors to “run away”, GPT followers still have checks in hand, expressing optimism about the technology.

Just today, Microsoft is negotiating to invest US$10 billion in OpenAI, which will increase the company's market value to nearly US$30 billion.

Niko Bonatsos of the venture capital firm General Catalyst said in an interview: "This is the paradigm shift we have been waiting for." "Maybe it's bigger than we thought."

In the eyes of investors, these algorithms are cool. The text-image generator is impressive and opens the door to a creative world for those new to Photoshop. For them, the GPT system is at least fun to play with.

While industry CEOs are open to the fact that these projects are still in their relative infancy, and even though the future of the field is bright, the disruption potential they present and Blurred creative boundaries are hard to ignore.

In Chollet’s view, to truly form a “paradigm shift”, products must not only be cool and interesting, but even very useful for niche products.

He warns that VCs are taking far greater risks than they realize, both by fueling the hype cycle of half-baked products and by being frantically plundered. Capital has not made prudent predictions about this emerging market, which has broad prospects but still has considerable limitations.

"Everyone is starting to believe they are the chosen one, especially those who call themselves contrarians," he said.

The above is the detailed content of Google researcher blasted GPT startups for their hype and sent out 18 tweets in anger!. For more information, please follow other related articles on the PHP Chinese website!