Chaos Labs (@chaoslabs) / X

Pinned

Mar 13, 2025

1/ Introducing Chaos AI—The World’s First AI-Powered Crypto Researcher. Built on years of proprietary data from securing trillions in trading volume, Chaos AI transforms fragmented market data into institutional-grade financial intelligence. Get Early Access:

00:00

The enterprise AI playbook is changing .@datadoghq's recent developer survey found that 70% of enterprises now use three or more LLMs across multiple tools, making multi-model AI the enterprise default. Against this new reality, enterprises need two capabilities: → Routing:

Replying to @chaoslabs

2/ What this means. AI workloads span a wide range of reasoning requirements. However, inference at most enterprises still follows a one-size-fits-all approach, with the same model handling requests that require fundamentally different levels of reasoning. Every request sent to

1/ Are you spending dollars on AI requests that could cost cents? @cursor_ai’s recent developer report found that identical AI workflows can vary in cost by nearly 9× across leading models, making model selection one of the largest drivers of inference costs.

Collapse the noise; retain the work Every enterprise AI workflow generates multiple execution events, including planning steps, tool calls, code revisions, and human review. Together, these events produce a validated outcome that captures how the problem was solved. We call this

A Framework for Mapping AI Spend to Business Outcomes

Most organizations evaluate AI through cost and usage. A million tokens spent on a billion-dollar risk decision and a million tokens summarizing a Slack thread cost exactly the same on the invoice....

Insane that this works, i'm a bit curious about how much quality gets hit and definitely going to try it once my limit is reset tomorrow. Aside from compression losses, two more things come to mind though: 1. It makes not much sense that it actually works as a saving strategy,

Michigan TypeScript

@MiTypeScript

Jul 3

~60% Fable cost cut by transparently turning the code into an image and having the model OCR it. WILD idea. also hilarious. github.com/teamchong/pxpi…

We are getting to the point where this thesis gets one more validation every day. Yesterday, Together AI announced an investment from Aramco Ventures. Together AI is essentially a bet from Aramco that intelligence becomes commoditized, but that the infrastructure for producing

The Commodification of Intelligence

Two years ago, enterprise AI procurement was relatively straightforward. Frontier capability was concentrated among a handful of model providers, so most organizations standardized on a single vendor...

Pay for inference, own the intelligence OpenAI’s recent reductions in inference costs will likely reduce the cost of deploying AI across most enterprise workflows. However, inference costs are only one part of the economics of AI deployment. Equally important is the infra for

0xGeeGee

@0xGeeGee

Jul 2

We're definitely heading in the direction of commodified intelligence... OpenAI found a way to decrease inference cost, because that's what we have to do, due to supply chain constraints (every part of it is becoming incredibly expensive) and competition pressure from open

Over the last few months, companies have started reacting to uncontrolled AI spending. The temptation is to answer the “tokenmaxxing” phase by simply turning the clock back, with less AI: lower token...

Seems some "interesting" choices were made around Sonnet 5. Initial testing points out to it being extremely token inefficient, especially at max effort, almost reaching Fable and passing Opus 4.8 (max) by quite some, despite achieving LESS than both (and than GPT 5.5 xhigh) on

Fundamentally, AI has to (and will) undergo a "financial rationalization". Just like you wouldn't want to leave your capital working at 0.1% APY when you can get 4% at the same risk, you wouldn't want to obtain some meager results by burning millions of dollars in tokens when you

A Framework for Mapping AI Spend to Business Outcomes

Most organizations evaluate AI through cost and usage. A million tokens spent on a billion-dollar risk decision and a million tokens summarizing a Slack thread cost exactly the same on the invoice....

You can only optimize what you can measure. Token Yield was a good theoretical start, but measuring useful work proved easier said than done. Sharing more on our semantic routing research and creating the base for attributing agentic work to outcomes and spend.

A Framework for Mapping AI Spend to Business Outcomes

Most organizations evaluate AI through cost and usage. A million tokens spent on a billion-dollar risk decision and a million tokens summarizing a Slack thread cost exactly the same on the invoice....