news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

editor: editorial department hyz

[new intelligence guide] just now, foreign media revealed: sora is in a research dilemma! the reason behind it is security risks and cooperation with hollywood. in addition, a recent goldman sachs report misinterpreted the decline in chatgpt traffic, which directly triggered panic in the market and caused ai stocks to be sold off.

sora is in deep research trouble?

after sora was released in february this year, it is still a "futures" so far. why is it not open yet?

just recently, foreign media cnbc revealed that the reason why openai's sora model was delayed for so long was due to "dialogues with policymakers."

as early as the end of february, bloomberg reported that openai hopes to avoid security risks as much as possible considering this year's us election.

another reason was that sora was not technically ready for release at the time.

the reasons behind the delay in fulfilling futures contracts are complex.

since sora came out, there has been a huge explosion of successors, but it has never really been fully released.

from time to time, the official openai account releases a video generated by artists using sora.

in the latest video, a singaporean artist named niceaunties used sora to try a project on the theme of "aging, beauty, freedom and fun".

niceaunties named it "auntie's eggs".

another new video is a fashion show created using sora by david sheldrick, a korean-born artist living in london.

after watching the video, many netizens expressed their amazement. some even described it as the birth of an art movement.

for many people, openai sora is just a tool in the hands of artists.

it is obvious that ai video models such as runway gen-3 alpha, dream machine, and domestic popular video models are not only close to sora's strength, but can also be used for free.

even midjourney is preparing to officially announce its ai video model.

why is openai so slow to release its work?

foreign media said that sora may be in a dilemma in its research. on the one hand, it must discuss security risks with the government in depth; on the other hand, it urgently needs the cooperation of hollywood and artists.

some netizens speculated that sora must undergo censorship before being released.

on the other hand, the potential legal issues openai faces in training models using youtube data is also a factor.

in march, openai chief technology officer mira murati was interviewed by bloomberg and was vague about sora's use of youtube data for training.

she simply said, "we use publicly available data and data that has been licensed."

later, youtube ceo neal mohan warned against this possibility in an interview.

murati has said that it is certain that sora will be released this year, but first the cost of production needs to be reduced. currently, sora is much more expensive than other existing ai systems.

although not open to the public, artists have been given the opportunity to try it out.

since the release of sora, hollywood artists who have been able to obtain internal testing qualifications have already shown the direction that openai wants to focus on in the future.

in march, openai released the first batch of videos generated by seven artists using sora.

for example, the inspirational short film "balloon man" that impressed many people.

then in may, writer and director paul trillo released the first official music video featuring sora.

to make this 4-minute video, he spent 6 weeks, generating a total of 700 clips, and finally selected 55 of them as the finished film.

not only is openai targeting hollywood, it is also "marketing" sora to film studios, media executives, and brokerage companies, hoping that filmmakers will use it in the production of their works.

ashton kutcher, one of the directors testing the sora model, expects it will have a profound impact on the film industry.

according to recent reports, openai is working with the new york city museum of art to provide artists with access to sora for their strada gallery exhibition.

it is reported that the strada nuova: new road art exhibition will last for three weeks.

many artists, researchers, scholars, and creative people at the intersection of physical and digital art will be attending the exhibition.

curator paul hill said that they began working with openai about six months ago to provide artists with accessible ai tools.

these include not only the sora video generator, but also the voice engine speech generator, the dall-e 3 image generator, chatgpt, as well as educational resources and artist grants.

minne atairu, an interdisciplinary artist, has been exploring the application of ai tools in art for the past four years.

she uses ai to generate 2d/3d images, as well as videos and other works, focusing on understudied areas in black historical archives.

for the exhibition, atairu used sora to generate the regina gloriana video, inspired by supernatural horror films made in nigeria in the 1990s.

sora developers believe that the current level of the model is equivalent to the early gpt-1. just as gpt-1 proved that language models can scale, sora may gradually acquire new and unexpected capabilities like the gpt series.

therefore, further scaling of video generation ai models may bring about revolutionary new applications.

however, it is worth noting that gpt-1 appeared very early, and it was not until gpt-3 that the road to commercial planning was truly started.

goldman sachs just retracted its "ai bubble" statement, but misinterpreted "chatgpt crash" again

in addition to sora, which has been postponed for an indefinite period of time, openai has had other bad news recently.

a report from goldman sachs misinterpreted the decline in chatgpt traffic, which directly exacerbated people's negative sentiment towards ai stocks and caused them to start selling!

recently, goldman sachs analyst peter oppenheimer released a report.

a chart in the report shows that chatgpt's traffic has plummeted in recent months.

shortly afterwards, the financial times reprinted this chart.

now, there is new material for the "ai bubble" story, which is very sensitive in the financial market.

everyone’s negative sentiment towards ai stocks has once again been triggered!

however, this time, the financial circle was really too shocked.

if you look carefully at the report of this goldman sachs analyst, you will find a big bug.

looking more closely, the analyst cited data from similarweb but did not take into account openai’s recent domain change from “chat.openai.com” to “chatgpt.com.”

with this domain name change, the traffic of "chat.openai.com" will naturally decrease.

in fact, if you look at the data from similarweb itself, you will find that this is not the case at all.

similarweb’s monthly ai service shows that chatgpt achieved 66.2% year-on-year growth and remains the most popular generative ai application to date.

competitors like claude and perplexity are still catching up, but have yet to catch up with chatgpt.

chatgpt’s global user base is growing rapidly: desktop and mobile visits surged from november 2022 to july 2024, peaking at nearly 2.5 billion visits in june and july

other indicators also show strong demand for openai's services.

openai recently announced that the company has reached 200 million weekly active users, double what it had in november 2022.

not only that, the use of openai models through apis is also increasing.

microsoft applications that use openai models, such as github copilot, are also growing rapidly.

it can be said that if we were to take stock of the challenges facing openai, the departure of security researchers and high costs would definitely be among them.

but to say that there is insufficient demand is absolutely untrue.

a recent analysis suggests that openai could generate $3.5 to $4.5 billion in revenue this year.

for a company that only began serious commercial operations at the end of 2022, this is impressive data.

however, the cost behind this is as high as 8.5 billion us dollars.

"too much investment, too little return" sequel is coming

recently, goldman sachs also emphasized: we are not against generative ai!

in june, goldman sachs released a report titled "generative ai: too much investment, too little benefit?"

this may give people the misunderstanding that goldman sachs is pessimistic about ai.

in fact, they are just asking questions - note that there is a question mark after the sentence.

to clarify, goldman sachs analyst peter oppenheimer and his team recently released a sequel, explaining in more depth their understanding of the so-called "rational prosperity of technology."

in the report, they analyzed it like this:

the technology sector has generated 32% of global equity returns and 40% of u.s. stock returns since 2010. this reflects stronger fundamentals rather than irrational exuberance.

global technology industry earnings per share have grown by about 400%, while all other industries combined, even considering the peak before the financial crisis, have only achieved growth of about 25%.

it seems that goldman sachs does not think that ai is a bubble, but there is still a hint of warning in it.

any good story carries with it other dangers.

it can amplify interest so much that it monopolizes investor attention at the expense of other opportunities and lead to unrealistic expectations of future profits, which can easily lead to a sharp downgrade of the company.

disruptive technologies almost always go through a boom-bust-and-boom process.

after the initial bubble bursts, the opportunistic followers are eliminated and the big companies will do something truly useful.

goldman sachs says the same risks and opportunities emerge in every cycle -

historically, investors have tended to overfocus on innovators; they have underestimated new companies that use other people’s capital to enter an industry and develop new products and services, and they have also underestimated opportunities in non-tech industries to use technology to generate higher returns.

the survivors plowed the land on the bones of the dead. in goldman sachs' view, the infrastructure left by the initial wave of investment and capital expenditures paved the way for new products and services.

the bubble in artificial intelligence is not exactly the same as in other fields.

because most of the leading companies are "winners" that emerged from previous bubbles.

these technology giants make profits through advertising, which means there will be a steady stream of large capital expenditures.

goldman sachs believes that despite the huge investment, there is still great uncertainty in the development of ai, which means that other new players will also have the opportunity to gain a foothold in the development of ai in the future.

the current winners’ moats in ai are quite significant and valuations do not look like bubbles, but the number of new patents in the field is growing rapidly, indicating that new competitors will emerge and costs will fall.

super-large companies have strong capabilities to invest in ai models, but lower-cost open source alternatives are emerging rapidly. huggingface, the world's largest open source platform, already has about 650,000 models. this shows that large-scale investment growth and competitive models are emerging in the field of ai, just as happened in previous waves of technology.

just as competition is often underestimated, the returns on capital expenditures on innovation are often overestimated because the marginal cost of technology falls and capacity increases over time, goldman added.

for example, in the early days of the internet, telecom companies were considered the winners.

compared to the more speculative, unprofitable internet companies, they seemed like a “safer” way to make money.

but the ultimate winners will be those companies that can "free ride" to leverage spending and production capacity to build business models.

many of these companies did not emerge until the advent of smartphones and apps in 2006, which subsequently gave rise to large-scale platform companies, shared mobility, and social media.

in short, generative ai is not a bubble.

however, in goldman sachs' view, today's ai winners will no longer be asset-light industries.

ai is driving a major capital spending boom that has reversed a 15-year period of high returns, and current valuations suggest that this will continue.

moreover, there is currently no evidence that pioneers like chatgpt have the ability to achieve sustained commercial success.

furthermore, technology is inherently deflationary.

this may help evade antitrust scrutiny, but it is not conducive to maintaining profit margins.

therefore, goldman sachs' advice to investors is that investments should be more diversified and they should not just buy more nvidia shares.