The Innovation Game education, benchmarker and algo scoring, staking tracker, curated news. The only app you'll need to stay up to date with the $TIG ecosystem
The $TIG commercial licensing model is either the most structurally interesting royalty mechanism in crypto, or a well-designed framework that no enterprise will ever use.
Both possibilities are worth understanding before you decide which one you believe. @tigfoundation
So Claude launched Fable - which is a nerfed version of Mythos
We're finally starting to see what we have predicted for a long time.
The beginnings of managed stagnation.
The most powerful version stays in-house, everyone else gets a throttled copy.
Select "trusted" teams got access to Mythos while everyone else got Fable (and not for long - it appears that it will be available only via extra usage from June 22) meaning even the nerfed version will cost $$$
We're starting to see rich / whatever "trusted" means in this context get access to the latest and best models while everyone else gets access to a worse one
They also added "novel" safeguards - including limiting what it does when used for frontier LLM development!
It limits what the model does (i.e. further nerfs it) without the user even know via methods such as prompt modification.
e.g. you give it a task, it returns you an output
but behind the scenes
if it didn't like what you asked, it changes your prompt into something it does like, and returns that without you being aware!!!
And they mention LLM development as a specific aspect that's blocked!
This is like Microsoft shutting down the PC when it sensed Linus Torvalds coding Linux.
This is managed stagnation in its early form.
One company holds the frontier and rations who gets it.
The best model goes to a trusted few, a weaker one to everyone willing to pay, and nothing at all to the rest.
That is exactly what open algorithms break.
You can't nerf math that is already in the open.
This is the fork in the road.
Either algorithms go open in the next year or one company owns the future and never gives it back.
Prometheus has its second beta and performed impressively against the neural net optimiser challenge.
Twice in a row - a swarm of agents optimisng algorithms - has perfomed better than our in house mathematicians expected
That openness is also the only thing that scales globally.
China blocks Google.
US labs won't run data through Chinese infrastructure.
Every centralized player is locked out of half the world by default.
A network owned by nobody is the only thing all sides will use.
John Fletcher: "there's more compute outside of Google than there is inside of Google."
Everything uploaded to TIG is open forever.
It cannot be bought up, closed, or forked into something private.
Same reason nobody bothers forking Linux into a closed product, the open one already won.
You get paid, and the work stays free. A lab offers you neither.
TIG is the economic layer.
You upload what you discover, and you get paid for it.
It turns algorithm discovery into something you can actually mine, the way Bitcoin made hashing into an industry anyone could enter.
Prometheus finds it.
TIG is the market that buys it.
But discovery alone does not pay you.
Prometheus has no token and no economics.
You can even use it to build something private and closed if you want.
Finding the algorithm and getting rewarded for it are two different problems.
That second problem is the one TIG solves.
Which means the advantage compounds for whoever runs it longest.
The knowledge builds up over time, so someone mining consistently from early on accumulates an edge a latecomer can't easily catch.
AI models have read every paper and textbook ever written, but they have never seen anyone's mistakes.
Nobody publishes the dead ends, the failed approaches, the things that didn't work.
That knowledge exists nowhere.
@Dr_JohnFletcher "there's a vast, vast wealth of extremely useful information which is in the mistakes which they haven't seen."
Prometheus generates exactly that as it runs.
Every failed attempt teaches it something the literature never recorded.
Prometheus discovers algorithms the way evolution works: thousands of agents mutating code, breeding the winners, keeping whatever runs faster.
It runs open, and anyone can point it at a problem.
On its own it finds algorithms, the same kind of open swarm discovery that just beat Google.
That is what makes the swarm matter.
A leaderless crowd matched a closed lab, which is the thing everyone assumes is impossible.
The method they used is the same idea behind Prometheus.
This is why discovery is going closed across the board.
Every lab faces the same math, so every lab reaches the same conclusion: keep the good algorithms private.
It is already happening.
The gold rush is on, and it has been running inside the big labs for three to six months.
So security was not the real reason.
Cost was.
Finding one algorithm with AI now burns millions in compute.
Once it costs that much, open-sourcing it just hands your edge to rivals who keep theirs locked.
So you keep yours locked too.
@Dr_JohnFletcher "if it costs you millions to create, you can't just make it open... in the end, you're out-competed and run out of business."