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

Ultraman simply said, let AI start-ups shiver:

"Shell"OpenAI, doomed to die!

This is a speech by OpenAI CEO Ultraman at the latest YC alumni sharing meeting. For this reason, he added: especially don't spend too much energy on the UI interface.

It resonated with countless people. Many industry insiders said:

The reality is, I'm seeing a lot of that happening.

In addition to this sentence, Ultraman's speech can be described as frequent golden sentences, many of which are "dry goods details" and have also been extracted and forwarded by everyone-

These include progress on GPT-5 and 6, views on future trends in AGI, and a variety of hands-on start-ups.

For example,"AI girlfriend" is just a beautiful trap, do not try it easily.

ChatGPT was deliberately chosen as a machine-like name so that humans would not have feelings for it.🤣

After listening to the speech, netizens even described it as "a golden harvest."

So what exactly does Ultraman share in this talk? Let's take a look.

"Simple learn me die" At YC Alumni Reunion 2023, Ultraman is having some entrepreneurial advice on the body. In the summary of some enthusiastic netizens, it can be roughly divided into three points:

Companies that simply package OpenAI don't live long.

Optimistic AI medical consultant, AI personalized one-on-one counseling and other directions.

It is very unwise to raise money blindly.

The first point is for simple packaging, Ultraman for example, companies are solving small defects in the GPT model, especially in the interface.

First, OpenAI is already solving most of the work; second, these companies that compete with OpenAI only in cost and speed will not have a long-term and sustainable competitive advantage. Businesses must be delivering truly unique value.

Some netizens associate that some time ago Sequoia Capital proposed generative AI to enter the second stage_

The hype and quick displays are being replaced by truly valuable and complete product experiences.

All these views actually coincide.

Many industry insiders expressed their affirmation:

Don't put yourself on the chopping board and get killed by the next version of OpenAI.

(Yes, there is an official internal taste of killing fellow men)

But a different view has also been put forward: That's already too much of a statement, and it seems silly to me. It reveals a few things: Monopoly is good; first-mover advantage will win…but competition promotes progress, and ChatGPT does have a lot of shortcomings that can be improved.

If so, what direction is worth doing?

Second, Ultraman is optimistic about AI's empowerment in medicine and education_

AI medical consultants and AI personalized one-on-one tutoring. Both are of enormous social value.

For example, in education scenarios, OpenAI itself is very active.

OpenAI is preparing for the OpenAI Academy, which is expected to start by the end of 2023. It could be a free online teaching system accessible to everyone, where teachers can interact with GPT-5 and receive feedback and guidance during lessons.

Prior to this, OpenAI also demonstrated two educational scenarios in the GPT-4 customer case. GPT-4 incarnates AI learning assistants, both as virtual tutors for students and as courseware assistants for teachers.

Third, blind fundraising without a plan is unwise. Don't think money makes everything okay.

The summary netizen also said incidentally: Ultraman can do, does not mean you can also do.

What does GPT-5 look like? Another hot topic is OpenAI's next-generation large models, GPT-5 and GPT-6.

In this regard, Ultraman also spoiled the appearance of the next generation of large models in his speech, but not much:

GPT-5 and GPT-6 will have multimodal output capability, higher reliability and better personalized customization experience than existing GPT.

Specific multi-modal capabilities, some netizens predict that they can better complete tasks such as grounding, NLP and visual scene alignment, such as image annotation, visual question and answer, etc.

Combined with the previous Ultraman also hinted that GPT-5 will have new functions such as speech recognition, synthesis and emotion detection, GPT-5 should have the ability to align various modalities such as "text and text listening."

With the topic of GPT-5 and GPT-6, Ultraman also "laid a road" for the big model:

Moore's "scaling laws" for large models are now at work. Large models are becoming cheaper to train and cheaper to invoke GPT interfaces.

Scaling Laws is a law proposed by OpenAI in 2020, which simply means that as the model size, dataset size, and computational floating point number used for training increase, the performance of the model will improve.

Ultraman believes that with the rapid decline in energy and computing costs in the future, more powerful AI capabilities will emerge,"a lot of things that were previously unthinkable will be done."

But even if the big model has such a bright future, Ultraman says "there is still a long way to go from AGI"_

Both GPT-5 and GPT-6 are far inferior to AGI.

Even the standard of "human-like" alone, at this stage there is no AI Chatbot can do:

Even with the most cutting-edge AI technology behind it, I didn't feel like I was chatting with a "person."

Ultraman believes that true AGIs will be able to "reason on their own," that is, develop new knowledge over time. For example, AI that can write papers and do experiments based on existing knowledge of physics can reach the threshold of AGI.

However, the AI output should be treated with caution:

People forgive mistakes, but they don't forgive computers. The standards are different.

Ultraman also gives an idea of the AGI moment:

At that point, people may experience a degree of self-identity crisis, but not too bad and confusing. We've been through these moments many times, and technology will eventually become ubiquitous.

"An Undergrad Insists on Results" Finally, regarding ChatGPT itself, Ultraman also cue to two small "tidbits."

First, ChatGPT is called ChatGPT, there is no nickname like a person's name, because Ultraman does not want to go in the direction of "AI girlfriend" products.

Second, ChatGPT is not a project that OpenAI originally "invested in."

From beginning to end, OpenAI even only one undergraduate insisted on doing research in the direction of big models, others more or less turned to robotics, or game AI and other fields "publishing papers."

This undergraduate student who only works on large models, Alec Radford, has now reached 9.6w+ citations in Google Academic.

Alec Radford presented the Franklin W. Olin School of Engineering in 2016. Olin College of Engineering, then joined OpenAI for seven years.

It is the big language model route he insists on that has changed the direction of OpenAI and the entire AI field.

To this, Ultraman half ridiculed and half seriously said:

When hiring, remember to focus on people who "don't seem to be joining the company."

OpenAI didn't start out with big models, but we still recruited AI researchers and got more people to join after the breakthrough, which eventually led to ChatGPT.

Combined with OpenAI's "no industry" rules for recruiting talent, breaking convention seems to have become a new impetus for the company to gain inspiration and vitality.

Reference link:

[1]https://twitter.com/sama/status/1705751665293554120

[2]https://www.linkedin.com/posts/ibamasood_chatgpt-can-now-hear-see-and-speak-nervous-activity-7112087060068368384-0OXb/

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.