Original title: Owning vs. Renting Information
Photo by Lin Qiao
Photo by Peggy Block Beats

Editor: Mythos was shut down this week, and many AI entrepreneurs were reawakened to an issue that was overshadowed by cost discussions: When the core competencies of the product are built on external models and platforms, what is really owned by the company

OVER THE PAST FEW YEARS, OPEN SOURCE MODELS HAVE OFTEN BEEN DISCUSSED IN THE FRAMEWORK OF " CHEAPER FRONT-LINE MODEL ALTERNATIVES". HOWEVER, ACCORDING TO THIS PAPER, COST IS NOT THE MOST CRITICAL VARIABLE, BUT CONTROL. FOR AN AI COMPANY, CALLING ON THE FRONT-LINE MODEL API CAN JUMP-START PRODUCTS AND LOWER THE TECHNOLOGY THRESHOLD, BUT IT ALSO MEANS THAT CORE COMPETENCIES MAY BE SUBJECT TO MODEL SUPPLIERS ' RULES, PRICES, STRATEGY ADJUSTMENTS AND EVEN DOWNSIDE DECISIONS。

THE ARTICLE FURTHER ARGUES THAT "INTELLECTUAL" DOES NOT MEAN ABANDONING FRONT-LINE MODELS, BUT RATHER THAT ENTERPRISES SHOULD SINK THEIR OWN DATA, WORKFLOWS, FIELD KNOWLEDGE, ASSESSMENT CRITERIA AND MARGINAL CASES INTO MANAGEABLE MODEL SYSTEMS. THE COMPETITION FOR AI IN THE FUTURE WILL NOT NECESSARILY BE DOMINATED BY A SINGLE LARGEST MODEL, BUT WILL RESULT IN MULTIPLE "FRONTIERS": GENERIC FRONTIER MODELS, ENTERPRISE-SPECIFIC POST-TRAINING MODELS, VERTICAL SPECIALIZED MODELS AND ROUTING SYSTEMS CONSISTING OF MULTIPLE MODELS。

Mythos therefore shut down like a reminder that the real moat of the AI era is not just a model that can be used, but a way to turn intelligence into a company's own asset。

The following is the original text:

Mythos was shut down this week. Whether or not you agree with this decision, it's not really the point。

What really stings many people is that a company built on intelligence beyond its control is suddenly exposed to a set of decisions that it cannot influence. Many of the founders, when they saw this, asked themselves the same question: What part of my business is actually "rent"

OVER THE PAST FEW YEARS, DISCUSSIONS ON OPEN SOURCE MODELS HAVE LARGELY REVOLVED AROUND COSTS: CAN THEY REALLY ACCOMPLISH THEIR TASKS? IF YOU CAN, HOW MUCH CHEAPER IS IT COMPARED TO CALLING THE FRONT MODEL API

By now, we have had fairly clear answers. We have worked with companies such as @RampLabs, @cursor ai, @harvey and others, and basic pathways are similar: from a powerful open source model, post-training on what is really important to the company, and continuous rigorous evaluation to compare it with forward models。

The result was unexpected. On tasks of greatest interest to enterprises, a fine-tuned open source model tends to approach and even reach the quality of the front-line model at very low cost。

But what is really clear this week is that cost is never the most important issue。

The deeper problem is control. Who owns the intelligence on which your product depends

Many recent discussions have been summarized as the difference between renting and ownership. It's not perfect, but it's useful。

Lease intelligence

It's been working well until something went wrong. Apartments are ready to be packed, lights are on, pipes are working and maintenance is also in the hands of people. That's why most companies started off on this path。

FRONTLINE MODEL API IS AN EXCELLENT PRODUCT. THEY MAKE IT POSSIBLE FOR START-UPS TO BUILD SOMETHING THAT SEEMED INCREDIBLE A FEW YEARS AGO。

But renting also means restrictions. Landlords can raise rents, decide what changes you can make, and change the rules. Occasionally, for reasons not related to you, they can also tell you that you should move。

You didn't do anything wrong. You've just been operating on someone's territory。

And that's why Mythos' story resonates with so many people. When your core competencies are entirely dependent on the platform of others, you are exposed to a set of decisions that are beyond your control。

Most of the time, it doesn't matter. But sometimes it becomes so important in a moment。

Have intelligence

The lesson is not that companies should stop using front-line models. Far from it. The forward model laboratory has already developed extraordinary technology. Most products should be used. We use it ourselves。

In many senses, frontier models are becoming infrastructure. But infrastructure and ownership are different。

YOU CAN USE THE PUBLIC INFRASTRUCTURE WHILE STILL HAVING SOMETHING THAT TRULY ADDS VALUE TO YOUR BUSINESS. IN THE AI FIELD, SO-CALLED "OWNERSHIP" MEANS STARTING FROM A STATE-OF-THE-ART OPEN SOURCE MODEL AND SHAPING IT AROUND THE MOST UNIQUE PARTS OF YOUR COMPANY。

Your data。

You work stream。

Your field knowledge。

Your marginal case。

Your test。

Your definition of "good."。

Over time, this model will become less generic and more reflective of the work that your company really does every day. It is here that value was created。

You can imagine it as a house. It is easy to move furniture and to brush a wall. But if your future depends on the house itself, sooner or later you'll wish you had the ability to move the wall. So is intelligence。

When intelligence really belongs to you, no one can take the floor off your product silently。

And that's why we built Fireworks in this way。

We place training and reasoning in the same system, allowing companies to use the best open source models, to shape around the most important issues in their business and to deploy steadily to the production environment。

Not just consumer intelligence. It's smart。

There is no single front

THERE IS ALSO AN OPTIMISTIC REVELATION THIS WEEK: THE FUTURE OF AI DOES NOT DEPEND ON A MODEL WINNING。

There is no single front. There are many kinds of front lines。

The front model is a front。

A model of post-training based on multi-year corporate expertise is another front line。

A dedicated model that solves a narrow problem better than any model is another front。

A system that leads requests to multiple models and allows them to work together, moving beyond a single model in many tasks is also a front line。

THE MOST INTERESTING CHANGE IN THE AI FIELD IS NOT THAT A MODEL IS GETTING SMARTER, BUT THAT INTELLIGENCE IS BECOMING MORE AND MORE CUSTOMIZED。

Ultimately, the winning companies are not necessarily those with the largest models, but those that can turn intelligence into their own unique assets。

Looking ahead

Much of the week has been devoted to responding to news, and we have chosen to continue publishing products: @Kimi Moonshot K2.7 Code, @MiniMax AI M3, @Alibaba Qwen 3.7 Plus。

The future I look forward to is not a model that quietly devours everything it sees。

It's that many teams have their own part of the front line。

If mythos were to be shut down so you could start rethinking the trade-offs, we'd be happy to talk。

[ Chuckles ]Original Link]