Glyph Research

Lossless neural compression on @bittensor (SN117). miners compete, validators verify, compression becomes the benchmark

[Glyph Research @glyphresearch]

https://t.co/E24xK0CCZ7

[Glyph Research @glyphresearch]

We understand that many people across the ecosystem are feeling hurt, cautious, and exhausted after everything that has happened recently. Trust is difficult right now, and we don’t blame anyone for asking hard questions.

The Glyph team is also going through an incredibly painful time. We are doing our best to remain open, take responsibility for our mistakes, and share the evidence transparently.

At this moment, we would deeply appreciate warm encouragement and constructive guidance rather than accusations and blame. Thoughtful questions are always welcome, and kind words would mean more to the team than you may realize.

We know trust cannot be restored through words alone. Time, our actions, and our continued commitment to Glyph will ultimately speak for us.

We remain here, and we will keep building.

[Glyph Research @glyphresearch]

On July 30, Glyph suffered a major loss during a proposed OTC and investment transaction.

The agreed payment was never delivered. The transferred GLYPH alpha was unstaked, and approximately 3,479.53 TAO was routed through multiple wallets toward accounts we have identified as connected to centralized exchanges.

Based on the full evidence, we believe this was an intentional and prepared fraud.
The Glyph team suffered the overwhelming majority of the loss.

We also take responsibility for proceeding without independent escrow. We trusted the counterparties after meeting them in person and completing a small test transaction. That was a serious failure in our process.

We have published the complete incident report and on-chain analysis here:
https://t.co/NJK2Tza9KX

This is an extremely difficult time, but we remain committed to Glyph and will keep building.

Thank you @taoapp_ for the invaluable tools, and @bd_himes for his dedicated support throughout the investigation.

[Glyph Research @glyphresearch]

@DrocksAlex2 https://t.co/W7oGnQR7FG

[Glyph Research @glyphresearch]

Lessons learned the hard way.
Every failure teaches a lesson, and every lesson brings us one step closer to success. Onwards. 💪

[Glyph Research @glyphresearch]

Lessons learned the hard way.
Every failure teaches a lesson, and every lesson brings us one step closer to success. Onwards. 💪

[Glyph Research @glyphresearch]

We’re deeply sorry for what happened yesterday.
We’ll share everything when the time is right.

Right now, we’re already working with Binance, MEXC, the police, and our legal team. There is much more happening behind the scenes, but we’re going through an incredibly difficult time.

The truth is, no one has been affected by this more than we have. We’re living with the consequences every single day while doing everything we can to keep the project moving forward. We never imagined we could be betrayed by someone we had met in person just days earlier.

Despite everything that’s happened, the team continues to build every single day. Development hasn’t stopped, and our commitment to this project and its long-term vision remains unchanged.

All we ask is for a little patience and understanding. Please give us the opportunity to complete the investigations and make a proper official announcement before drawing conclusions, spreading rumours, or making accusations.

We completely understand why people have questions, and we intend to answer them as openly and transparently as we can when the appropriate time comes. Until then, we’re simply asking for your patience and trust while we work through one of the most challenging periods we’ve ever faced.

Thank you to everyone who has continued to support us. It truly means more than words can express. ❤️

[Glyph Research @glyphresearch]

@travelbai

[Glyph Research @glyphresearch]

@travelbai

[Glyph Research @glyphresearch]

This proposal is a practical interim measure while we deepen our understanding of Yuma and its features.

The good news is that we've already received valuable feedback from several community members who've spent considerable time studying this problem.

We'll keep learning and iterating.

[Glyph Research @glyphresearch]

https://t.co/0Lk0rvAWUJ

[Glyph Research @glyphresearch]

https://t.co/0Lk0rvAWUJ

[Glyph Research @glyphresearch]

@tao_Alph I wish we were there already, but we're still early.

We're working towards it every day. We may have a long road ahead, but we're confident we're doing the right thing - and the whole team is fully committed. 🚀

[Glyph Research @glyphresearch]

Update: We've reduced the eligibility threshold from 90% to 80%.

[Glyph Research @glyphresearch]

https://t.co/vlLsbGTQ2a

[Glyph Research @glyphresearch]

Another owner alpha lock completed. 🔒

We've just perpetually locked another 19,141.4139 α, representing all owner alpha accumulated since our previous lock.
We said we'd continue locking on our own schedule, and we're continuing to do exactly that.

https://t.co/kKwIaFco00

Long-term alignment isn't something we talk about occasionally - it's something we consistently demonstrate.

Onward.

[Glyph Research @glyphresearch]

We have now locked 100% of owner alpha perpetually.

Previously, owner alpha was locked in the default decay mode. We have now moved from that to a perpetual 100% lock.

Current alpha conviction snapshot:
- Total alpha conviction: 135k
- Owner conviction: 127k alpha locked by owner
- External conviction: 8k alpha locked by external users

This is another clear signal of long-term alignment.

We are not here for short-term extraction. We are here to build Glyph with strong accountability, sustainable alphanomics, and a serious long-term commitment to the subnet.

Fully committed to Glyph 🚀

[Glyph Research @glyphresearch]

@dougsillars Miners simply need to lock part of their incentives to be eligible for the incentives - that's it.

And I get your point. I think using the term "conviction" instead of*"lock" may have caused the misunderstanding. Thanks for pointing it out.

[Glyph Research @glyphresearch]

@0xfeng666 Your support means the world to us. Thanks, man! ❤️🤝

[Glyph Research @glyphresearch]

It's a bit different from Owner Conviction.
In our proposal, miners lock their Alpha to their own hotkeys, not to the owner key.

The goal isn't to increase conviction behind the subnet owner. It's to keep miners economically aligned with the subnet's long-term growth instead of immediately selling all of their rewards.

At the same time, we recognize that miners need liquidity to cover operating costs, invest in R&D, and continue improving their models. That's why a portion of their emissions remains fully liquid and can be sold freely.

Think of it this way: miners receive enough liquid Alpha to fund their operations, while the remaining portion becomes long-term committed value that they still own and can gradually unlock over time.

In other words, miners are effectively paid a salary through liquid emissions, while conviction acts as a long-term bonus, aligning incentives without taking away ownership.

[Glyph Research @glyphresearch]

Good catch, thanks for pointing that out. We've updated the article to avoid the terminology confusion.

By "reversible", I meant that eligibility for winner emissions is reversible once the miner again meets the conviction requirement - not that the lock itself is reversible.
And the conviction is locked to the miner’s hotkey

Appreciate you taking the time to read through it.

[Glyph Research @glyphresearch]

It’s now live. 🚀

Miner Conviction is officially part of Glyph.

Miners are now stakeholders in the Glyph ecosystem, creating stronger long-term alignment while making the subnet significantly more resilient to the impact of any single large exit.

This update also includes:
• BlockMachine RPC (SN19) integration — our first integration with another subnet, special thanks to @blockmachine_io
• Enhanced scoring mechanism

Huge thanks to everyone who provided feedback on the proposal.

Onward. 🚀

[Glyph Research @glyphresearch]

https://t.co/vlLsbGTQ2a

[Glyph Research @glyphresearch]

Big shoutout to the team! 🙌

We’ve recently migrated Glyph to BlockMachine RPC and have been loving it so far. Fast, reliable, and incredibly easy to integrate.

Thanks for building such a great product for the Bittensor ecosystem. Looking forward to seeing it grow even further! Cheers~

[Glyph Research @glyphresearch]

This is exactly why we’ve been so excited about Glyph’s next phase.

For the past few weeks, our team has already been designing a model compression competition as the second mechanism onGglyph. seeing more discussions around the growing demand for model compression gives us even more confidence that we’re heading in the right direction.

As frontier models become larger and more capable, the challenge is no longer just building them - it’s making them practical to deploy.
The winners of the next AI era won’t necessarily be the teams with the biggest models, but the ones that can deliver comparable intelligence with dramatically lower storage, memory, bandwidth, and inference costs.

We believe model compression will become a foundational technology adopted across the ai industry - not just by frontier model labs, but by every company building and deploying AI at scale.

That’s exactly the problem we want to accelerate through decentralized competition.

Coming soon, Glyph will expand beyond lossless text compression into model compression, powered by @bittensor 's multi-mechanism architecture.

We’re confident this is where the industry is heading, and we’re excited to help push that frontier forward.

Great write-up below that captures why model compression matters 👇

[Bai @travelbai]

https://t.co/bqWYTOupfj