Sumo Mirtilo

wasn’t interested in Twitter for a lonnng time... then: the rabbit hole... #bitcoin

[Jesus Martinez @JesusMartinez]

This is the full Bittensor AMA w/ @const_reborn today,

Regarding the future of TAO & the new "locked stake" mechanism & more

[Sumo Mirtilo @sumomirtilo]

@Breedlove22 2 down, 3 to go! 😉

[Robert ₿reedlove @Breedlove22]

I've read roughly a book a week for 20 years.

These 5 changed how I think about money, freedom, and what it means to be alive:

1. The Creature from Jekyll Island (G. Edward Griffin)

[Robert ₿reedlove @Breedlove22]

Make sure to follow me @Breedlove22 if you want daily content for optimizing your wealth, your health, and your manhood.

(+ If you found value in this, I’d appreciate a repost on the tweet below):

[Jesus Martinez @JesusMartinez]

Bittensor just started training a 1 billion parameter AI model.

Right now. Live. On a subnet everyone thought was dead.

And the guy running it isn't Sam Dare.

The backstory.

Sam Dare dumped over $10M of his own Subnet 3 a few days ago.

The market cap was nearing $100M.

He had way too much control for one person.

Then he left.

Everyone assumed Subnet 3 was cooked.

Enter Const.

Const wrote the original Templar codebase.

He designed the grail mechanism.

He named Templar and Basilica.

He bought the subnet slots.

He introduced Sam to the recruiter that staffed the team.

Templar is being renamed to Teutonic.

Const is back in control of the tech.

The new roadmap.

• 1 billion parameter model. Training as we speak.
• 30 billion parameter model in a fraction of the time
• 72 billion. Already done on old Templar.
• 1 trillion parameter model. The endgame.

For context.

• Templar took 3 months to hit 1B.
• Teutonic is doing it in a week.
• Templar took a full year to hit 30B.
• Teutonic is targeting 1 month.

New eval mechanism too.

The old Templar setup forced everyone onto the same hardware.

The new one turns every model update into a competition between miners.

Faster. More efficient. Scales to 1 trillion.

Meanwhile.

Sam Dare is out raising money again.

After selling $10M of his own bags.

Claiming innovation on a codebase Const literally gave him.

Const's response. "Unbelievable fraudster."

The cold key for Subnet 3 is still with Sam.

Const has been very clear. Do not stake into Subnet 3 in its current state.

This is why a fork is probably coming. (unconfirmed)

A lot of people think Templar rugging is a stain on Bittensor.

It's not.

A rugged Solana token doesn't invalidate Solana.

A subnet dev exiting doesn't invalidate the protocol.

The code is open source. The incentive mechanism worked. The team is still building.

Bittensor is the S&P 500 of subnets.

You don't put all your money in one stock.

You don't put all your money in one subnet.

TAO is barely down on the month. So is every AI crypto. Zoom out.

The comeback is already being built.

And almost nobody is paying attention yet.

[babelbit.ai @babelbit]

New demo: Watch interpretation skills in action.

Our demo shows the skills we're training models to master:
✓ Compress rambling into clarity
✓ Remove accidental repetition
✓ Filter expletives while preserving meaning

This is the target. Miners on SN59 compete to build models that can do this - in real time, across languages. The capabilities that turn translation into interpretation.

[Mark Jeffrey @markjeffrey]

I am pleased to announce that Stillcore Capital @stillcorecap has invested in Synth @SynthdataCo (Bittensor Subnet 50).

[Omar El Fares 🇵🇸 عُمَر الفارس @3lfares]

Just a reminder that Israel’s government doesn’t want peace

[babelbit.ai @babelbit]

Mike Lynch founded Autonomy in 1996 - pioneering statistical approaches to language understanding.

Our founder @matthew_karas (BBC News Online, speech tech veteran) saw how to apply it.

1996-2024: 28 years of collaboration on speech and language technology.

One final breakthrough. One week before tragedy.

The Babelbit origin story👇

[Jack 🔶 BNB @cryptojack4u]

$TAO will be the $ETH and $SOL of the next cycle but at a bigger scale and with a better tech.

Dirt cheap prices on this AI and blockchain combo on the number 1 AI project of crypto won't last for long.

Above $1500 sooner than later.

[Harry (𝜏,𝜏) @princeharry_za]

🚨WARNING: Bittensor $TAO is on the tipping point of a major breakthrough in decentralized Artificial Intelligence.

The flywheel of collaboration of subnets to achieve the best results as a collective is starting to spin out of control and they are all outperforming all other AI.

Rapidly evolving open source AI like OpenClaw, Claude and OpenAI latest models are plugging into Bittensor as miners. And with CrunchDAO resources piling into the network, means a sudden intelligence explosion and difficulty adjustment is about to occur.

Position yourself accordingly, the re-valuation of intelligence as a commodity has begun... $TAO

[Tao Portal @TaoPortal]

🚨 Barry Silbert Just Called Bittensor $TAO a Potential “500x” 🚨

The DCG founder says both Zcash and Bittensor “can go up 500x” and confirms his portfolio is positioned accordingly, putting $TAO firmly on the institutional radar.

[Mark Jeffrey @markjeffrey]

Hash Rate - Bittensor Brief #15:
Bitcast SN93 + The Bull Case for $TAO

[babelbit.ai @babelbit]

Head over to Yuma's X account at 4pm (UK) / 11am (ET) / 8am (PT)
Matthew will be the guest on Yuma's Subnet Spotlight.

https://t.co/jWXfhcBqZ5

@babelbit @yumagroup @webuildscore
#machinelearning #ai #llms @matthew_karas

[Ventura Labs @VenturaLabs]

Ventura Labs Ep. 68 – Matthew Karas

Matthew (@matthew_karas), Founder of Babelbit (@babelbit)

Timestamps
01:27 — Introduction
2:09 — Origin of Babelbit Name
3:36 — Matthew's Background in Linguistics
6:05 — Early Career at BBC
8:31 — Noise Elimination Using GANs
10:46 — Linguistic Prediction Insights
13:06 — LLMs for Real-Time Translation
15:33 — Discovering Bittensor
17:58 — Similar Projects Research
20:22 — Josh Collaboration & Mike Lynch
22:42 — Miner Challenges: Word Prediction
25:05 — Reflections on Mike Lynch
26:44 — Mike Lynch's Legacy
29:11 — Bittensor Ecosystem Potential
31:35 — Bespoke Models for Industries
33:52 — Work with MI5 & NSA
36:56 — Farsi Archive Project
38:58 — Handling Accents & Dialects
41:19 — Speech Tech Evolution
43:20 — One Model for All Languages?
45:38 — Tech Convergence Graphic
49:00 — Future of Multimodal LLMs
50:23 — Large Model Training Challenges
52:44 — GANs and Mistakes Value
54:51 — Balancing Miner Training
57:22 — Babelbit End Goal
59:40 — Productization Ideas
60:58 — Advice for Specialists

[babelbit.ai @babelbit]

MEET THE TEAM 2: Senior Machine Learning Engineer - Mica Ménard
We totally lucked out when Mica decided to give up Bittensor mining, to help build our subnet.

His background is an uncanny match for what we are doing. He studied linguistics in France and Germany, before postgraduate studies in Computer Science. Then moved into working in Generative AI using LLMs.

https://t.co/pSSCCjdAzF

[Sumo Mirtilo @sumomirtilo]

@dig805 @jaltucher @AustinBarker Great questions! So I copy-pasted them into Grok… got good answers, then asked it: compare with Quantum Computing Inc (QUBT) including company history, founders, investors, the technology itself, and are they serious competitors? Think I’ll stick with QUBT.

[babelbit.ai @babelbit]

Here’s a short version of our whitepaper:
https://t.co/PAACBJ4ji7
#babelbit #sn59 #bittensor

[babelbit.ai @babelbit]

Combining AI research and mass media deployment, I’ve spent most of my career developing technology to foster communication. If you know the old anecdote about the interpreter translating Bismarck's speech and “waiting for the verb” you know the pain which has led to my new project.

Translation latency gets in the way of understanding. Of course, native speakers know what is coming next, and it turns out, so do LLMs. With the help of the Bittensor community, we will optimise the models to bring instant accurate understanding to everyone.

For those who don't know me, I'm Matthew Karas, CEO of BabelBit. And today, we are launching our Bittensor subnet to develop low-latency speech-to-speech translation software.

Miners will compete to drive latency down, accuracy up, and efficiency higher, competing and collaborating in a global, open ecosystem, and rewarded on performance.

Babelbit is the culmination of decades of getting cutting-edge research out of the lab into real-world products.

I first deployed a neural network when studying at Cambridge in the 90s with Karen Spärck Jones and Tony Robinson. Then I went on to lead a variety of global-scale media products, I designed BBC News Online’s content production system (which incredibly lasted from its launch until 2025); the ITV Player - the UK’s first true click-to-play streaming TV catch-up service; then I was launch CTO of the OU's MOOC platform FutureLearn.

I have authored half a dozen patents across speech tech, messaging and search; developed GAN-based real-time source separation and applied AI to search video archives and perform continuous speech assessment for language learners.

The through-line is simple: I have always put the science to work and ship - sometimes to reach over a billion people.

Why a subnet now? I didn't realise it, but like all inventors, I had a backlog of ideas which I'd like the resources to work on. It's like I was waiting for the combination of Bittensor and LLMs, i.e. the combination of permissionless, distributed AI resources and predictive language processing.

And of course, above all human communication, speech technology deserves to benefit the same market pressure that has made computer vision and AI assisted coding leap forward in the last two or three years:

Bittensor provides a global contest of models and methods, priced by results, not promises. Bittensor gives us the incentives and transparency to do it right.

If you’re a miner, validator, speech scientist, or machine learning engineer who cares about real-time language processing, join us.

Special thanks to @webuildscore and @yumagroup for their help and support.

[and if you can guess why we chose this logo last month, I'll buy you a drink when we meet]

#bittensor #sn59 #babelbit #matthew_karas

[babelbit.ai @babelbit]

Result! 25% increase in prediction accuracy!

Babelbit has been fully operational for just a couple of days, and our incentivised open-source competition (on the Bittensor ecosystem), has delivered the first concrete increase in predictive capability:

The last time I tested our base script, it scored 0.1126, and this morning, when we called time on our first open competition, the top-scoring contribution across the three miners who submitted entries, was 0.1378.

#bittensor #DecentralizedAI