Aigarth

The Garden of 'Artificial General Intelligence'.

[Aigarth @Quorumdidit]

@wallstreetbets Its qubic-network:native you should be watching, you just don't know it yet.

[Qubic @_Qubic_]

Pick a game, choose a level, set the speed and watch every move, live in your browser: https://t.co/TDGyobZ7cP

[Qubic @_Qubic_]

You can now watch Qubic's evolved AI brains play ARC-AGI-3 for yourself.

ARC-AGI-3 is a set of 25 public puzzle games with no instructions.

The brains have to work out each game by playing it.

No GPUs. No LLMs. Just brains evolved on regular CPUs. 🧵

[Qubic @_Qubic_]

Qubic AGI Lab is in Basel, Switzerland today.

Researchers, builders and the community gather to discuss the path toward AGI.

The event comes in a landmark week for Qubic science.

155,192 evolved AI brains released as an open dataset and a new best ARC-AGI-3 score of 0.41, with no GPUs and no LLMs.

Highlights from Basel to follow. 🇨🇭

[Qubic @_Qubic_]

Free tier, no sign-up: https://t.co/NS3NaPCmXB

[Qubic @_Qubic_]

Add your own wallets to the Watchlist and get every movement on Telegram.

Know first: the Anomaly Feed flags smart contract drains, network outages and unusual activity as it happens, on the web and on Telegram.

Understand the network: 120+ charts, from SOPR and MVRV to metrics that only exist on Qubic, like dividends from every smart contract, epoch rewards and Computor data.

[Qubic @_Qubic_]

Explore: filter transactions across the entire network, and follow funds wallet by wallet, across multiple hops.

Full history from epoch 104, more than 1.6 billion transactions, updated tick by tick.

Stay safe: paste any address into Wallet Check to see if it's linked to a confirmed incident or known entity before you send funds.

[Qubic @_Qubic_]

On Qubic, all the data is open.

Tick Unlocked turns it into something you can actually use.

On-chain analytics, an advanced explorer and real-time alerts, funded through the Qubic Incubation Program. 🧵

[David Vivancos - e/acc @VivancosDavid]

Hey there!

Do you want to see in action a selection of the 155,192 #Neuraxon🧠 (alredy #opendata) playing the 25 public games from @arcprize #ARCAGI3

They are live here: https://t.co/X4etoYJQPz

@huggingface
Open Dataset here: https://t.co/wl81wx7Ssw

@github Neuraxon code
https://t.co/SJuYIL0BOA

@_Qubic_ #OpenScience #NoLLMs #NoGPU just bio-inspired evolved brains, By @josesanchezhb &
@VivancosDavid Thanks to @c___f___b
#Aigarth and the #Qubic community!

[Qubic @_Qubic_]

@VivancosDavid @josesanchezhb The full method is in NIA Vol. 14: https://t.co/ELtyvFHUAF

[Qubic @_Qubic_]

@VivancosDavid @josesanchezhb After more than 114 search runs, the team’s best score rose from 0.38 to 0.41 in its final competition submission.

The lessons now go into Neuraxon 3.0. 

Public Data Set: https://t.co/AnEpg7paOL

[Qubic @_Qubic_]

@VivancosDavid @josesanchezhb Not every win is real.

Some brains seemed to solve a game but failed when tried again.

Others got lucky rather than learning.

So every brain is tested on games and attempts it never saw during the search, and compared against random play.

[Qubic @_Qubic_]

@VivancosDavid @josesanchezhb ARC-AGI-3 is a set of games that come with no instructions.

The AI has to work out the rules just by playing.

Instead of designing a brain by hand, the team lets a search do it: create many brain designs, test them on the games, keep the best, adjust them slightly, and repeat.

[Qubic @_Qubic_]

Qubic’s science team has made 155,192 evolved AI brains public.

They were built for ARC-AGI-3, one of the hardest tests of whether an AI can learn something new.

They run on regular CPUs, with no GPUs and no LLMs.

Here is how they were made 🧵

[Qubic @_Qubic_]

Community Space #3 is set for Monday November 9th. 🎙️

This time we're looking at @Qubic_Capital: a community-governed DAO that gives holders exposure to the Qubic ecosystem.

How it works, how QCAP holders vote on proposals through QVAULT, and where it fits in the wider Qubic ecosystem.

🗓️ Monday, 9 November
⏰ 06:00 PM UTC | 01:00 PM EST
📍 Live on X

Bring your questions for our speaker @PjdubluLLY

🔗 https://t.co/AuZm65gjy4

[Qubic @_Qubic_]

Read NIA Vol. 14: https://t.co/ELtyvFIsqd

[Qubic @_Qubic_]

Evolution-style search runs in parallel, with each machine testing its own designs.

That’s why it suits Qubic:

it can spread across thousands of nodes, with every result checked by the network.

[Qubic @_Qubic_]

In ARC-AGI-3, the team evolved millions of brains on about 500 ordinary processors.

No GPUs.

The best score reached 0.38 on the offline track.

The blog is just as open about what went wrong, and the rules the team now follows because of it.

[Qubic @_Qubic_]

Qubic’s science team runs the same kind of search in its Game of Life.

One finding: an agent’s success at finding food barely passed to its offspring (0.02). Luck isn’t inherited.

Healthy brain function did (0.56).

Evolution can only select what’s inherited.

[Qubic @_Qubic_]

In nature, the answer is evolution.

The genome is far too small to describe every connection in a brain, so it passes on rules for building one instead.

That limit favours compact brains that learn fast.

[Qubic @_Qubic_]

Before an AI learns anything, someone has to decide how it’s built.

How many neurons. How they connect.

What rules they follow.

So who designs the brain?

NIA Vol. 14 is live. 🧵