news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

text | wei linhua

editor｜wang yisu

as a hot ai company, open ai once again received a huge round of financing that would make its opponents jealous. on october 2, open ai announced that it had successfully completed us$6.6 billion in financing, which not only made up for its money-burning r&d losses, but also obtained financial support for the next round of competition.

although it has become a trillion-dollar unicorn, the anxiety of open ai still exists. huge r&d, personnel and other cost expenditures, as well as the transformation of profit-making companies, have become the whips that spur open ai to accelerate.

fortunately, while firmly turning to a for-profit company, open ai has already begun to make a big move towards super ai products.

walking on two legs

open ai plus products

within a week after the financing plan was revealed, open ai took frequent actions:

first, at the second developer conference on october 1, a number of practical api calling plans were presented; and one day later, it was officially announced that the 6.6 billion financing plan had been finalized, making it a trillion-level unicorn in one fell swoop; a new product called canvas was released without any notice and is open to plus members.

open ai, which has successfully received another round of financing, is eager to walk on two legs and seize the b-side and c-side to work together.

on october 1, at the second developer conference, open ai did not release a product. instead, it brought the benefits of tool applications to developers. it released four apis: real-time voice api, visual fine-tuning api, and prompts. caching api, model distillation api.

for subscribers, open ai also provides sincere feature updates.

in the early morning of october 4, open ai officially released canvas, an additional product function based on the gpt-4o model that can cooperate with ai to complete the writing and coding interface. it is currently open to chatgpt plus and team subscribers for testing.

in terms of interface design, canvas abandons the traditional model and uses a blank interface to carry the various needs of ai work. "my vision for the ultimate agi interface is a blank canvas."said karina nguyen, r&d member responsible for canvas.

different from the previous interactive form of multiple rounds of dialogue, canvas allows users to use [highlight] to mark the parts that need adjustment in the creation interface, and let ai complete local and detailed adjustments. if you are not satisfied, you can also use the [back] button to restore the previous version.

for different needs, open ai has also designed specific shortcut commands to help users improve efficiency. taking writing as an example, canvas has five built-in shortcut commands: suggest editing (providing revision suggestions), adjusting length (expanding/deleting content), changing reading level (from kindergarten to graduate level), and final polishing (checking grammar, expression, and consistency), add emoticons (insert emoji into text).

optimized editing function for highlighted parts source: open ai

sam altman himself seems to be particularly satisfied with the function of adding emoji expressions, and even posted a separate x asking "is adding emoticons the best feature of open ai ever?".

sam altman only gave the "yes" option for users to choose, which shows that he is satisfied with this feature. source: x

canvas is also committed to becoming a more user-friendly product that understands users better. in addition to providing new interaction methods and convenient command functions, open ai also enables intelligent recognition: when chatgpt detects a potentially helpful scene, canvas will automatically open. in addition, you can also jump to the corresponding interface by typing the command to start canvas in the dialog box.

it is worthy of recognition thatcompared with the previous series of open ai products, canvas has begun to truly touch the pain points of users in terms of product function design and interactive experience.

in terms of functional design, canvas provides shortcut functions that can improve work efficiency, some of which are well received by users. for example, "code review" for programming can use ai capabilities to help users automatically check for problems and complete modifications.

in terms of interactive experience, canvas stores text, coding, and other content that does not require memorizing contextual dialogue in a separate blank interface, and uses local selection to improve the flexibility of modification. coupled with the smooth switching of the shortcut icon bar on the right and the transition animation effect of content generation, open ai's product details are becoming richer.

behind the specific productization, perhaps chatgpt faces the problem of limited growth in paying users.

according to the new york times, as of august, chatgpt had 350 million monthly active users. but with a huge user base, its number of subscribers is only 10 million. it is very difficult to support the huge revenue target based on current paying users alone.

just like the problems faced by domestic pan-chat products such as doubao, wen xiaoyan, and kimi, chatgpt’s user questions in general scenarios are always limited, and it can only take away part of the cake of general search.

if you want to further increase your willingness to pay and customer unit price, you must find the right product in a dedicated scenario. this is also the reason why openai is focusing on both the b-side and c-side.

accelerate commercialization

productization is the key

open ai has been releasing updates one after another and is in the highlight stage of successful financing.

on october 2, according to the bbc, open ai has successfully completed the latest round of financing of us$6.6 billion, led by thrive capital (goldman sachs capital), with technology companies such as softbank, nvidia, and microsoft also participating. open ai is currently valued at us$157 billion, successfully becoming a trillion-dollar unicorn.

but under the spotlight, open ai’s goal of turning a profit and its disclosed operating income have also attracted much attention. subject to previous investment agreements, open ai is considering transforming into a for-profit company.

coerced by the contract, open ai’s money-making plan must be put on the agenda at an accelerated pace.according to the washington post, people familiar with the matter revealed that the terms of the new funding come with a stipulation that if openai does not complete the transformation within two years, investors can convert their shares in the company into debt with a 9% interest rate.

information disclosed by the media shows that open ai’s operating revenue this year is expected to reach us$3.7 billion. in order to achieve the ambitious goal of annual revenue of 100 billion u.s. dollars in five years, open ai needs to find more valuable payment points, whether in c-side subscriptions, which account for the majority, or in b-side service companies and individual developers. to attract users to pay for it.

since last year, sam altman’s product strategy has begun to take shape.

in november last year, at the first developer conference held by open ai, it not only announced the large model gpt-4-turbo, but also significantly reduced the price of tokens while ensuring performance, with an input equivalent to 1/3 of gpt-4. and 1/2 output tokens were launched with discounts, and also used this occasion to showcase their new product-gpt store (gpt store). in the gpt store, without writing code, anyone can create a customized version of chatgpt by setting prompt words based on model capabilities and their own needs.

gpt store source: open ai

in july this year, open ai officially released the ai ​​search product search gpt, but only opened a test quota limited to 10,000 users. "we believe that today's search still has a lot of room for improvement." sam altman commented on x on the day searchgpt was released.

while testing ai products, open ai is also recruiting experienced product developers to find suitable talents for the productization of open ai.

in june this year, kevin weil officially announced that he would join open ai as chief product officer. officially, kevin will lead a product team. previously, he served as vice president of product for facebook novi, vice president of product for instagram, and vice president of product for twitter, and has extensive experience in building products.

however, unlike open ai's absolute advantage in developing large multi-modal models, open ai is always slow in developing new products from gpts, ai search to canvas.

so far, the ai ​​products released by open ai have always found benchmarks. for example, gpt store is not a new feature. as early as the beginning of last year, a number of platforms that integrate the capabilities of large models such as open ai, google, and anthropic and provide customized agent functions have already been launched. for example, the "overseas version of zhihu" created by the president of quora the ai ​​robot replaces poe.

at a time when the discussion of "can ai search kill traditional search engines" is becoming increasingly heated, open ai's product releases also lag behind traditional giants and entrepreneurs in the same track, and in terms of product form and interactive experience, searchgpt has no provide more eye-catching optimizations.

the canvas released this month is not an innovative product. it is regarded as a benchmark for the artifacts and programming software cursor released by anthorpic in june.

product artifacts for code visualization source: anthorpic

open ai’s ambition to seize the market for ai application products is clear, but its moves are not surprising.

real product or fake demand?

although there are constant actions, it is difficult to break down the products released by open ai in the past two years and leave an explosive communication effect similar to chatgpt in the mass market.

according to foreign media reports, the release of chatgpt is a bold attempt that was not expected by the open ai board of directors. chatgpt set a record of over one million users in 5 days and over 100 million users in 60 days, allowing sam altman to taste the wealth brought by the flagship product.

since then, while open ai has released the latest models, it has also been observing the implementation of products based on large models.open ai does not want to miss out on these tracks that have the potential to meet market demand and create new gold-absorbing beasts, as evidenced by gpt store and ai search.

however, after chatgpt, it is difficult for open ai to satisfy users again with its answers.

in nearly a year, gpt store has failed to prove its potential to become the “next app store” and has even struggled to retain the few early adopters.

similar to the app store, the gpt store's ecosystem relies on users and developers. however, the design of gpt store is not developer-friendly and performs poorly in aspects such as traffic drainage and plagiarism control. for example, it allows multiple users to create gpts with the same name, and there are no restrictions on gpts that are extremely similar. without a way to make profits and gain effective exposure, developers' creative enthusiasm will inevitably be suppressed.

what's worse is that gpt store cannot attract users to stop. according to the information, a developer studied more than 30,000 gpts and found that the vast majority of them only have 1-2 users per day, and only 5% of gpts can attract 150-500 users.

there is no news yet on the large-scale testing of ai search. like sora, the video model launched at the beginning of the year, it has become a blank check that open ai has not yet cashed.

in terms of product form, the search function implementation of open ai's searchgpt still focuses on the "answer engine" method of ai-generated content, which answers users' questions by integrating massive internet information. even though the barriers to ai search products are not high, open ai has not come up with more differentiated new products.

searchgpt interactive interface display image source: open ai

even if it goes online, can searchgpt really steal the limelight from other ai search products as expected? there is still a question mark over its effectiveness.

at present, from a revenue perspective, open ai's cash cow still comes from subscription services and cooperation with enterprises and individual developers.these products have neither brought the expected traffic drainage effect to open ai, nor can they support a beautiful new growth curve in terms of revenue.

however, in the next few years, open ai is destined to expand chatgpt’s services more and more.in documents disclosed by the new york times, open ai plans to increase the subscription price of chatgpt from us$20 to us$22 this year. in five years, the price will double to us$44.

fortunately, the release of canvas allows people to see the progress of open ai in product design and interaction. as for how many users will be willing to pay for it, it still needs to be tested by the market.

how to make users willing to pay for expensive pricing? canvas is only a small part of value-added services. finding more payment points is a question that open ai needs to answer with action in the next few years.