Analyst Zephyr says Fable 5's early NSA access program has damaged trust among non-US enterprises

Thanks for the additional context. What I meant was this likely leads to the government becoming more involved in the rollout of these models. Disagree on this event changing the trajectory of the ai strategies of enterprises - We were already trending toward a multi-model approach & temporary restrictions to the latest frontier models while still allowing unfettered access to previous frontier models should not hugely disruptive.

That said, I'm often wrong, so we shall see.

@zephyr_z9 Any biz workflow just NEEDS a stable end point.

Intelligence is like the fourth or fifth most important thing to a corporation integrating AI.

Waaaaaaay down the list after limited liability, stability, cost, confidentiality, <intelligence / capability are somewhere down here>

The rest of the world is mostly made of püssies Everyone Developed who wants to pay pensions to their boomers and does not already have a competitive AI industry will bend the knee

@Object_Zero_ Mostly agree But, AI is a bit different cuz more intelligence unlocks more workflows The rapid growth in Anthropic's ARR & higher agentic adoption happened after Opus 4.5 release (more intelligence was the main prerequisite here)

@zephyr_z9 Do the top orgs even have other options though?

If Mythos 2.0 is the best cyber security entity in existence - won't their systems be unsecure if they use inferior models?

@zephyr_z9 Any business will always be dependent on the whims of the government in charge of the jurisdiction in which it operates. I do agree that making yourself dependent on a proprietary model is foolish though.

I think you’re over indexing on mostly a non event. Net net, the most valuable workflows will always get in line for intelligence at the frontier.

Foreign states/corps may attempt to invest in sovereignty in model development, but will mostly fail.

This will be yet another industry that has to implement and navigate compliance flows and work with the USG on something important to national security.

@aaronimity @zephyr_z9 but the governments behaviour will differ based on the context. for example they can say "ban all ai for foreigners" and no one bats an eye. but if they said "deport all foreign healthcare workers" then there would be protests.

@rargulati @zephyr_z9 No sane none us entity will depend on the whims of usg. The fact that you guys think usg can get away with stuff like this tells a whole lot

@zephyr_z9 Investment implications?

@zephyr_z9 @Object_Zero_ It’s a lot of hair on fire. You are basically saying that just because somebody swapped out your Bugatti for a Porsche 911 Turbo you know cant drive to the grocery store.

@zephyr_z9 It’s not that simple they can’t, they must be using it or they get eaten by the US competitor that use it, they will migrate the operation headquarters to the US 😉

@zephyr_z9 @grok is this bullish $NOW ?

@zephyr_z9 Interesting. I tend to agree, and this was overdue in this sense.

FWIW it was already clear that Anthropic is very aggressive and gets into their customer’s markets, so it makes a lot of sense to me to not fully depend on them anyway. At least for small-medium companies.

@zephyr_z9 I've sold to enterprises. Using these public APIs is transitory. The costs can balloon to insane amounts + the vast majority are going to on prem AI with a datalakehouse + local LLM strategy.

Not to mention the majority of business use cases do not need mythos level reasoning.

@zephyr_z9 Probably not a good idea to scream doom from the mountaintops for two years while crying for government oversight and subversively doing everything you can to amass regulatory capture.

Don't be surprised when you call the cops, tell them there's danger, and they start shooting.

@zephyr_z9 It's crazy we're still debating if enterprises should own their model layer. USG uncertainty makes open weights the only sane default. I treat every revocable API as supply chain risk now.

@zephyr_z9 or just don't piss off usg and they won't clap back

@zephyr_z9 No it doesn’t.

For example: you do not use a 1% more intelligent model that is 50,000x more expensive.

The idea that intelligence is the main metric is just flat wrong.

A business will always want to use the most profitable tool for the task.

@Object_Zero_ @zephyr_z9 Very accurate. The decision matrix for work flows can be very large and dynamic making it hard to automate with traditional programming, but not really challenging; because people have to make them now.