My biggest takeaway from the Claude Fable 5 release was that open source AI won't just survive, it'll thrive
The release itself told me everything about where the closed labs are heading:
i) cyber, bio, chem & distillation queries reroute to Opus 4.8, a weaker model
ii) the safeguards are tuned conservatively and by Anthropic's own admission catch harmless requests
iii) the ungated Mythos 5 ships only to approved orgs through Project Glasswing
iv) even gated access comes off subscription plans on June 23, moving to usage credits
Frontier capability is now allocated by approval status
I read every refusal, downgrade and capacity gate as a demand signal
Capital is already pricing it:
a) Manifold Labs, the team behind Targon, raised a $10.5M Series A
b) DCG anchored Yuma's asset mgmt. arm with $10M
c) Grayscale went from trust launch to spot ETF filing in 18 months, a path that took Bitcoin a decade
d) six institutional TAO vehicles now exist, none of them pre-date 2024
Bittensor launched in 2021 with no pre-mine, no VC allocation and no permission required
It has grown into one of, if not, the most active capital markets for open source AI with an incredibly diverse set of subnets solving different complex problems
The report below tracks the full history, from genesis to dTAO, to the pending ETF
This is a useful trip down memory lane, and a vital reference point for an ecosystem projected to be worth hundreds of billions, if not trillions, in the years to come
以前有很多中国人问我能不能互关,我都没给,因为感觉他们太粘人了。但自从他们有了Ape之后就改变了,因为Ape关注Ape
i have a lot of Chinese ask me for a follow back in the past. i never gave it to them because it felt they're being too needy. but this changed the moment they get an ape coz ape follow ape
As believers of open research, we are disappointed to see Anthropic silently degrading Fable 5 for AI development
"Any topic related to building pretraining pipelines, distributed training infrastructure, or ML accelerator design... may have limited effectiveness through Claude via methods such as prompt modification, steering vectors, or parameter-efficient fine-tuning."
Not only do they get to decide what you use LLMs for in research, but this also enables them to silently intervene in your research without you knowing.
This sets a dangerous precedent. If a model refuses openly, users can understand the boundary. If a model falls back to another model, users can still evaluate the difference. But if a model silently modifies or weakens its own answers while still pretending to help, researchers lose the ability to know whether a failed result came from their own idea, their implementation, or an invisible intervention by the model provider.
That is not safety. Safety policies should be transparent, auditable, and user-visible.
On top of that, the people most harmed by this are not the largest labs with massive teams and proprietary infrastructure. It is the independent researchers, academic groups, startups, and open-source builders who rely on public tools to compete, innovate, and pioneer AI for everyone else.
@ASvanevik@FortuneMagazine congrats💯
the thing that always stood out is you made on-chain data something a normal person could actually read
that's the harder half and it's the half people skip
Ramp is telling you that the future is maximizing token spend
Dario is telling you most white collar jobs go away
Labor cost is still 10-100x inference cost
These two things are ~roughly mutually exclusive
🤔
@KyleSamani both are kinda sales pitches though
ramp needs you spending dario needs you a little scared
the 10-100x line is the only clean number in the whole thing