The "Burning Token" competition within Meta has reached new heights due to Alexander the Great.
Under the strong push of this Chief AI Officer, the company burns up to two trillion tokens every day, equivalent to going through the entire Wikipedia over 40 times.
The top ranked super user spent 281 billion tokens in a month, averaging 9.36 billion tokens per day.
Previously, Zuckerberg issued a hard target to completely rewrite the code repository with AI, attempting to allow AI to fully take over the underlying code modification work.
So, 85000 employees embarked on an economic pursuit to grab vanity titles such as "Token Legend", directly turning the scale of computing power consumption into a new badge for measuring workplace status.
However, Alexander King's daily two trillion won't go to waste, and Meta's first model led by him will soon be revealed in a "semi open source" form.
Meta's "Token Competition"
Within Meta, there is a ranking called Claudeomics, which records in real-time the token consumption of 85000 employees. Employees are frantically scrolling through the ranking in order to compete for medals such as Token Legend or Session Immortal.
In this' Token Competition ', Alexander the Great demonstrated strong dominance as the new AI leader.
Under his strong push, Meta achieved a terrifying daily throughput of 2 trillion tokens, pushing the scale of computing power consumption to an unprecedented height.
In this frenzy known as Tokenmaxxing, consuming more tokens is seen as a symbol of greater productivity.
Employees gradually assume that token throughput is the hard indicator for measuring workplace value, and even regard this extremely expensive consumption record as a symbol of identity.
In order to maintain a stable ranking on the leaderboard, some engineers have even started directing AI agents to conduct redundant research in the background for several hours, even running empty.
For example, some investment professionals have stated that they have many friends at Meta, and they say that Meta employees have specifically built robots that run in a loop to consume tokens.
He criticized this approach as extremely foolish, no different from measuring output by lines of code.
What's even funnier is that even if this simple idling doesn't actually produce any output for the company's business, as long as it can generate more impressive consumption data, it can attract attention in this competition involving all employees.
Zuckerberg strongly promotes AI
The trigger for this entire storm was actually a request made by Zuckerberg in an internal memo.
He requested the engineering team to rewrite all existing code repositories and thoroughly clean up those old accounts that AI cannot understand, so that AI can take over the underlying code modification work without any obstacles.
In order to cope with this seemingly impossible code refactoring task, engineers began to extensively use fully automated tools such as MyClaw and Manus.
Even if it's just dealing with some of the most inconspicuous scripts in daily life, we need to fully involve AI tools and try to use this saturated automated office to resist the heavy pressure of refactoring.
Faced with this kind of resource waste that almost burns out computing power, Chief Technology Officer Andrew Bosworth not only did not intend to extinguish the fire, but also showed considerable generosity.
He clearly supports this unlimited resource investment in public, believing that as long as money can be used to make a leap in productivity, this deal is a sure win at the company level.
Huang Renxun has previously publicly expressed similar views, believing that a top engineer with an annual salary of $500000 would be a "serious professional dereliction of duty" if he did not burn a token worth around $250000 every year.
Now it seems that Meta's team has really taken Old Huang's words to heart and is even implementing them in an almost crazy way to the end.
However, many netizens do not agree with this viewpoint, stating that if there is no actual value generated, the number of tokens is meaningless.
The criticism from netizens is not unreasonable, but on Meta's end, it cannot be said that there is no output at all - the new model developed by Alexander Wang is about to go online.
Meta is not closing its source again?
Prior to this, there were widespread rumors in the industry that Meta may completely abandon its open-source path.
At that time, many industry analyses pointed out that Meta may shift to a fully closed source model in order to protect its core technological assets.
The emergence of such market expectations reflects doubts from the outside world about whether Meta can continue to maintain an open strategy in the aggressive AI competition.
However, the first new model family developed under the leadership of King Alexander ultimately decided to return to the open-source path.
According to the currently determined plan, Meta has decided to provide the versions of these new models to the outside world through an open source license.
While opening up model call permissions to the developer community, Meta still maintains confidentiality regarding the most core technical details involved.
However, this time Meta also gained some self-awareness about its new model -
They knew that these models would not outperform OpenAI or Anthropic in all aspects, so they chose to bet on specific areas of consumer advantage to maintain competitiveness in an increasingly competitive market.
Specifically, Meta plans to deeply embed the various capabilities of these new models into the business processes of WhatsApp and Instagram.
Through this saturated social penetration, Meta attempts to build its own AI experience in daily communication and content creation.
Some netizens believe that with Meta's three major super social apps, this strategy of "controlling distribution channels" is a very wise choice.
img
