"Token's End" comes: who pays when costs are under pressure on productivity gains

Original title: "Token's End"! AI is in the middle of a price-fixing tsunami
Original source: Machine Heart

Recently, a new term was the subject of extensive discussion: "Tokenpocalypse"。

This was due to a pricing re-engineering of GitHub Copilot. From 1 June, Copilot fully shifted to a token-based billing model. Token cost multipliers vary greatly between models, and the single token price for some models is 60 times higher than for others。

And the advanced models that are widely recognized by users as being "real good" are the ones that are the worst。

As many head AI companies such as Anthropic and OpenAI prepare to be listed, AI will face even more serious profit pressures, perhaps forcing more producers to follow up on the price hike。

The cost of using AI has always been an unavoidable problem for enterprise expansion productivity. The "tokenmaxxing" wave that emerged during the previous period, compared to the driving force of the employee token usage, will end as Token's end approaches。

"the whole tide of tokenmaxxxing, from the top to the top, is six months."

Business dilemma

The developer from a large enterprise described a ridiculous dilemma: the company had long imposed on employees the use of AI tools, and token would be interviewed less. But with the new pricing, Token will also be interviewed。

More importantly, the Copilot team has so far not been able to access the "staff level token limit" function. This means that under the new billing model, an employee may burn down the entire company ' s budget in one day。

"my job is not to solve business problems with software," the developer wrote, "my job is to solve token usage."

The comment section is even better. One user concluded: "Corporate policy becomes: "Do everything with AI, but be careful not to use too much, because if LLM consumes too much token you will be disabled, and then you will be criticized for not using AI for the rest of the month."

ENTERPRISES ARE TOO FOCUSED ON AI PRODUCTIVITY AND MAY BE A DOUBLE-EDGED SWORD。

ONE INFORMATION MANAGER FROM A LARGE FIRM EVEN "SHOWING" AT AN AI SEMINAR: AFTER THEIR AI SYSTEM WENT DOWN, LAWYERS LARGELY STOPPED WORKING BECAUSE THEY COULDN'T LEAVE AI。

A MAN WITH YEARS OF PROFESSIONAL TRAINING CAN'T WORK WHEN HE'S FREE TO ADMIT THAT HE'S AWAY FROM THE AI CHAT BOX? I WOULD BE ASHAMED TO START REFLECTING ON MY CAREER."

Uber Overexpenditure: industry microcosm

Most of the AI models are now packaged, and the problem with controlling the budget for enterprises is exacerbated by the tendency of Token to gradually charge in volume。

Uber completed a full arc in one and a half months: first, he discovered that "AI budget burned much faster than expected" and then set an emergency ceiling and staff restrictions。

Imagine that companies like Uber, which use a lot of AI, can hit the wall so quickly, and TechCrunch's podcast says, "The question is, can AI's lab match the cost to the client's willingness to pay?"

One cold knowledge: ChatGPT Plus initially priced $20 a month without much strategic consideration, "just spit out a number." The whole industry is still paying for this starting point。

"YOUR JOB WON'T BE REPLACED BY AI, BUT YOUR BUDGET MAY."

Reddit has more interesting details. An AWS Bedrock cost monitoring panel has been set up at the company to put every model, every token (including cache token) spent in real time on Cloudwatch, "Let developers and finance watch the money burn." The reaction of the comment section was: "Congratulations, you just gave them a new KPI indicator."

Another major company has experienced a similar contraction: when the AI level was exhausted, the owner was forcibly downgraded to GPT-4.2, without even VSCode integration。

A bystander who is not in the science and technology industry has made many voices: "This whole thing has consumed more energy and actual hours than the delivery of jobs that really helped companies to earn money."

When the whole industry is still immersed in the narrative that "AI will replace everything," a more realistic problem has surfaced: hard-to-count bills have to be paid. The end of Token may be just the beginning of the liquidation。

Original Link