@beffjezos goes without saying, right? hopefully meta churns as customer

asking for a friend: “We don’t train on your data” is not the same as:

“We don’t analyze your usage.” “We don’t aggregate demand signals.” “We don’t mine metadata.” “We don’t infer what customers are building.” “We don’t use derived intelligence to compete with you.”

Enterprise plan, SOC 2, DPA, blabla, etc. are useful.

But I wonder: are they contractually barred from front-running customers?

@beffjezos Has any company in history destroyed their credibility so quickly and thoroughly as these clowns? None come to mind.

We have ~$2.5M spend with Anthropic, and I just got out of a meeting where people above my pay grade made the call to diversify our spend to OpenAI and Google over the next three months. My CTO is, somewhat manically, convinced that if they’re doing this to Fable they have the capability to do this to any model, and thus the company can’t be trusted anymore; and even if not that we can’t trust our org won’t accidentally use Fable for ML research, or that Anthropic’s classically horrible safeguards won’t accidentally trigger this protection on unrelated use-cases and we’d never know.

@beffjezos unironically this might accelerate the move to fully local stacks faster than anyone expected

@llenbira @beffjezos They said they were going to start enforcing their terms of service (i.e. not to use their models to develop competitors) by ensuring that if you try it, the output will be worse.

Soon, every startup will become a lab in one form or another.

@beffjezos I'm now thinking Trump was correct. Dario is probably spying against US national security interests. DOD must get self-hosted LLMs.

@beffjezos supply chain risk in what sense? data collected on fine-tunes, or just model usage being trackable? the second feels weaker, the gateways already log everything.

@beffjezos Always has been

@beffjezos Context?

@beffjezos We were the product.

@MartinShkreli @beffjezos that is bound to happen, maybe not total churn but massive usage decrease

this won’t be the trigger though

@beffjezos So Trump admin correctly accessed the vibe check.

@beffjezos Actually people should build the singularity architecture. Not playing with private data of humans or promoting LLMs/AI w/o the correct explanation/architecture and safety measurements.

@beffjezos They’re a supply chain risk period.

@beffjezos Seems like the admin was onto something…

@S_Fadaeimanesh @beffjezos the models silently poisoning your research

@beffjezos It already was

@beffjezos Fable 5 turned out to be a disappointment, let’s hope fable 4 does better.