Home > Industry Information > Exposure Of Openai "Strawberry" Was Released This Fall, Formerly Known As Mysterious Q*Model
Openai's strawberry is going to be cooked?
WisdomIt was reported on August 28 that today, foreign media The Information quoted people familiar with the matter and said that OpenAI will launch a new model codenamed strawberry (previously called Q*) this fall.
The strawberry project is a long -standing mysterious Q*model. It is rumored to be one of the key reasons for the previously dramatic OpenAI coup.It shows the mathematical questions that have not been seen and have more complicated programming, and researchers that make AI security at that time were shocked.
There are two most important uses in this project: First, improvementThe existing products such as ChatGPT are to help Openai develop the next -generation flagship model Orion.These all point the determination of Openai to maintain their leadership in the big model and get more income opportunities.However, recently, OpenAI has also been exposed to show the strawberry project that has been showed to the US national security officials. This may be a measure to increase the transparency of the product in the context of frequent resignation of this founding of the founding of the founding of the founding of the founding of the founding of the founding startup.Earlier this month, Openai CEO SAM When ALTMAN secretly pokes to show 5 strawberries in the garden of its own garden, some netizens jumped the prophet based on the mature time of the strawberry: GPT-5 may be confirmed to be released within 4-6 weeks in the next 4-6 weeks.This is exactly the time to release the time that may be released in autumn.
Maybe in the near future, we can seeThe strawberry project was directly integrated into ChatGPT.
01. Stabilizing the status of the big model, the strawberry can help the new flagship model to reduce the hallucination
The strawberry project is an important technical milestone in OpenAI. It can solve mathematical problems that have never been seen before. After training, it can solve subjective problems involved in programming, answer product marketing strategies, and solve complex word puzzle games.
I have never surfacedStrawberry models have become more obvious in the past time.
Apart fromAltman showed 5 strawberries, as well as OPENAI researchers Trevor Creech posted a document to have dinner in Openai, with strawberries in the plate.
Despite the currentOpenAI's model is still the first choice for enterprises and AI applications, but models from other companies such as Google, XAI, Anthropic and Meta are quickly catching up on many rankings.
therefore,The prospect of OpenAI depends on whether it can eventually launch a new flagship model code -named Orion.The model aims to improve its flagship model GPT-4 launched at the beginning of last year, which was launched at the beginning of last year.
andThe release of the strawberry project is crucial to the training of Orion -it can generate high -quality training data for Orion to reduce hallucinations.
The strawberry model can help Openai overcome the limitations of obtaining high -quality data, thereby using real world data such as text or images extracted from the Internet to train new models.
Smart startupAlex Graveley, former chief executive of Minion AI and former chief architect of GitHub Copilot, believes that using strawberry models to generate higher -quality training data can help OpenAI reduce the number of errors in its model, which is the so -called illusion.The reason why this model can do this is because there are less ambiguity in the training data, so it guess less times.
An obvious idea isThe reasonable reasoning capabilities of the strawberry model are incorporated into the CHATGPT.This may mean that although users get more accurate answers, the speed will slow down.
Therefore, this may not be suitable forUsers and other users such as Searchgpt search engines want to obtain tools that responded immediately, but it is very suitable for use cases that are not sensitive to time, such as repairing non -critical codes in Github.
Is that in the near future,ChatGPT users can choose to open or turn off the strawberry model according to the time sensitivity of the request.
02.OPENAI income is urgent?It has been displayed to government officials this summer
Faced with a large technology company with strong funding strength and the rival of creative startups who attract gold,Openai needs to open more income opportunities.
Although compared to a year agoOpenAI business has grown rapidly. At present, its API and ChatGPT subscription revenue that it sells to enterprises has increased by about two times to $ 283 million a month, but the monthly loss of OpenAI may be higher.The Information's analysis of the internal financial data that had not been disclosed and participating in the business based on the previously disclosed, OpenAI may lose up to $ 5 billion this year.
People familiar with the matter revealed thatAltman hopes to raise more funds for the company and find ways to reduce losses.As part of a commercial cooperation with Microsoft, OPENAI has raised about $ 13 billion from Microsoft since 2019, and the cooperation will last until 2030.However, the cooperation clauses may change, including how OpenAI pays Microsoft to rent a cloud server to develop its model, which is currently the cost of OpenAI.
In view of the existingDialogue AI and other dialogues such as ChatGPT are not good at in mathematical -intensive fields such as aerospace and structural engineering. The model of solving difficult mathematical problems may be a potential and favorable application.At the same time, the improvement of mathematical reasoning can also help the model better reasoning and dialogue, such as customer service requests.
Google and some startups are also developing reasoning technology.Last month, GoogleDeepMind's model defeated most human contestants in the International Mathematical Olympic Contest; the latest model of Anthropic can write more complicated code, answer the ability of charts and graphics;There are also some creators to improve their reasoning capabilities by decomposing problems into smaller steps, but this method is slow and higher in cost……
therefore,The release of strawberry models may bring new ideas to reasoning technology.
Altman has revealed that we feel that we have enough data to develop the next model.We have conducted various experiments, including generating synthetic data.He may refer to the Orion model at the time.
A person who knows the matter directly said,OpenAI showed the ability of strawberry models to national security officials this summer.
This demonstration may beOpenAI has worked hard to improve part of the transparency to the US policy maker. In recent months, one of the founders of OpenAI co -founders, John Schulman, resign from several security departments such as John Schulman, which has also paid more attention to the technical security of OpenAI.
03. Conclusion: OpenAi seek to open up more income opportunities
roll outThe strawberry model is part of the endless battle in the big model industry. It must lead the strong competitors of other funds and stabilize its domineering position in the field of big models.In addition, this technology also affects future products, such as intelligent bodies that solve multiple steps of tasks.
Big models such as Openai also hope to open up more income opportunities in this field to support high model training costs.
Industry Information
