@everyone
Dear Community,
>
Effective today, we are reducing emissions by 90% subnet 1 and 3 until further notice.
>
Why?
>
The most important quality of decentralization at Bittensor has always been open ownership. Unlike networks that build off large and closed raises made inaccessible to most, we chose to launch early and build in the open - so that everyone could participate. I strongly believe that this was a necessary choice to ensure Bittensor stayed truly aligned in an important way; by building an openly owned AI system.
>
The consequences of this choice has been varied. On the one hand, we have succeeded at distributing TAO throughout an incredibly broad group of people, and collectively learned how to drive this megastructure computational network in the real world - I am proud to be here with you.
>
However, on the other hand, building in the open has left us increasingly exposed to exploitative attacks that take advantage of our openness. Unlike most traditional blockchain projects, with fixed mechanisms like Bitcoin, Bittensor is an incredibly adaptive technology and as a consequence, building in the open, i.e. keeping emission rates steady with 99% uptime across the network, while simultaneously carrying out active development on the underlying protocol - with incredibly complex things like subnetworks, delegation and multi modalities etc - has caused us to suffer in other ways.
>
As of this writing, we can finally confirm that a sustained and organized circumnavigation on the reward model on subnet 1 has brought 75%+ of active nodes under a single group, there is also an ongoing DDOSS attack on subnet 1 against this same group. We know from many in this community that this has created increasingly high boundaries for entry into the network. At this point we believe reducing emissions is necessary to protect the network and more importantly the core mission of openness we began on.
>
Why this is a good thing:
>
We understand that by altering the core rate of inflation on Bittensor below its maximum rate will disrupt the exact moments when halvings are set to occur. However, this will not disrupt the 21 million cap. Instead, reducing the block emission rate - while maintenance and certainty are brought back into this network - will add longterm value into Bittensor. With more control over emission rates, we as a community and a project can direct its flow more precisely, to ensure that each TAO emitted maximizes value creation network-wide.
We understand that, aside from the people taking advantage of our vulnerabilities, this will affect the validators most at this time. We want to acknowledge that, and ensure you we will be carrying out these changes as swiftly as possible, in order to return emissions to the standard rate. That being said, we need to be able to reduce network emissions when huge changes have occurred, and when we have been made aware that the network has become vulnerable to attack, to ensure that our ecosystem stays aligned and protected.
>
How will these changes be made in the future?
With additional time to develop the mechanism, Opentensor and our open source collaborators will have more space to introduce decentralized decision-making into the protocol directly. Because of the importance of emission rates for all of us, our first goal with the DAO will be ensuring that changes to emission rates will be decided by delegates and TAO holders.
As ever, thank you to everyone working tirelessly to bring Decentralized Open AI to the world.
Blue skies ahead,
Const.
Dear Community,
>
Effective today, we are reducing emissions by 90% subnet 1 and 3 until further notice.
>
Why?
>
The most important quality of decentralization at Bittensor has always been open ownership. Unlike networks that build off large and closed raises made inaccessible to most, we chose to launch early and build in the open - so that everyone could participate. I strongly believe that this was a necessary choice to ensure Bittensor stayed truly aligned in an important way; by building an openly owned AI system.
>
The consequences of this choice has been varied. On the one hand, we have succeeded at distributing TAO throughout an incredibly broad group of people, and collectively learned how to drive this megastructure computational network in the real world - I am proud to be here with you.
>
However, on the other hand, building in the open has left us increasingly exposed to exploitative attacks that take advantage of our openness. Unlike most traditional blockchain projects, with fixed mechanisms like Bitcoin, Bittensor is an incredibly adaptive technology and as a consequence, building in the open, i.e. keeping emission rates steady with 99% uptime across the network, while simultaneously carrying out active development on the underlying protocol - with incredibly complex things like subnetworks, delegation and multi modalities etc - has caused us to suffer in other ways.
>
As of this writing, we can finally confirm that a sustained and organized circumnavigation on the reward model on subnet 1 has brought 75%+ of active nodes under a single group, there is also an ongoing DDOSS attack on subnet 1 against this same group. We know from many in this community that this has created increasingly high boundaries for entry into the network. At this point we believe reducing emissions is necessary to protect the network and more importantly the core mission of openness we began on.
>
Why this is a good thing:
>
We understand that by altering the core rate of inflation on Bittensor below its maximum rate will disrupt the exact moments when halvings are set to occur. However, this will not disrupt the 21 million cap. Instead, reducing the block emission rate - while maintenance and certainty are brought back into this network - will add longterm value into Bittensor. With more control over emission rates, we as a community and a project can direct its flow more precisely, to ensure that each TAO emitted maximizes value creation network-wide.
We understand that, aside from the people taking advantage of our vulnerabilities, this will affect the validators most at this time. We want to acknowledge that, and ensure you we will be carrying out these changes as swiftly as possible, in order to return emissions to the standard rate. That being said, we need to be able to reduce network emissions when huge changes have occurred, and when we have been made aware that the network has become vulnerable to attack, to ensure that our ecosystem stays aligned and protected.
>
How will these changes be made in the future?
With additional time to develop the mechanism, Opentensor and our open source collaborators will have more space to introduce decentralized decision-making into the protocol directly. Because of the importance of emission rates for all of us, our first goal with the DAO will be ensuring that changes to emission rates will be decided by delegates and TAO holders.
As ever, thank you to everyone working tirelessly to bring Decentralized Open AI to the world.
Blue skies ahead,
Const.
🔥6👍3❤1
@everyone
Dear Community,
We want to state for the record we now know that the “cabal” on subnet 1 was created by a few individuals who directed their energy and intelligence at beating the mechanism. The manner in which that was done was not perfectly aligned with the proper functioning of Bittensor. However, the mechanism is the mechanism, and these miners simply applied themselves to beat the incentive structure which was badly designed by myself and the foundation.
The blame rests with the mechanism designers, the cost here is on our shoulders, and the emission reduction was done to give us more time to explore more long lasting solutions to the exploit found on subnet 1. I do not believe there is considerable blame to be shared with any other group. It is our sole role to build the systems that are resistant to base level collusion, and thus align with the production of intelligence. We cannot and do not expect the self-interested market participants to do that for us.
Let's remember that, Bittensor is an engine that uses human ingenuity as well as compute to drive innovation. The drive from miners to beat the mechanism is the fuel we use to run this entire network, for all of us. I would like to note that the people involved in this directed exploit have come forward to me personally. They have shared everything they have with me, and have even offered to reduce their position in subnet 1 to zero if required. It is truly heart warming to see this.
With love and respect for this amazing group of people,
Happy Mining,
Const + Opentensor
Dear Community,
We want to state for the record we now know that the “cabal” on subnet 1 was created by a few individuals who directed their energy and intelligence at beating the mechanism. The manner in which that was done was not perfectly aligned with the proper functioning of Bittensor. However, the mechanism is the mechanism, and these miners simply applied themselves to beat the incentive structure which was badly designed by myself and the foundation.
The blame rests with the mechanism designers, the cost here is on our shoulders, and the emission reduction was done to give us more time to explore more long lasting solutions to the exploit found on subnet 1. I do not believe there is considerable blame to be shared with any other group. It is our sole role to build the systems that are resistant to base level collusion, and thus align with the production of intelligence. We cannot and do not expect the self-interested market participants to do that for us.
Let's remember that, Bittensor is an engine that uses human ingenuity as well as compute to drive innovation. The drive from miners to beat the mechanism is the fuel we use to run this entire network, for all of us. I would like to note that the people involved in this directed exploit have come forward to me personally. They have shared everything they have with me, and have even offered to reduce their position in subnet 1 to zero if required. It is truly heart warming to see this.
With love and respect for this amazing group of people,
Happy Mining,
Const + Opentensor
❤39
@everyone
The prompting network generated some wicked data:
https://huggingface.co/datasets/opentensor/openvalidators
https://huggingface.co/datasets/opentensor/openvalidators-mining
The prompting network generated some wicked data:
https://huggingface.co/datasets/opentensor/openvalidators
https://huggingface.co/datasets/opentensor/openvalidators-mining
huggingface.co
opentensor/openvalidators · Datasets at Hugging Face
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
🔥9👍1
https://tao.studio/ - Text 2 Image by Creative Builds
tao.studio
Image Generator - tao.studio
Generate images from text descriptions using Bittensor's testnet for text2image.
🔥18❤6
>
# Where are we going next?
We are using the next week with emissions reduced to test the above additions to subnet 1’s validation mechanism. We are confident that these tests will be done by the end of this week. Once these tests are finalized, we will initiate a delegate vote on revamping emissions back up to normal levels.
# What does this all mean
The most important quality of a healthy system is that self-interests are aligned with the interest of the rest of the protocol’s user group as a whole. In Bittensor’s whitepaper, the term ‘cabal’ was used to define the state where self-interested individuals come to dominate the system, rather than their project-interested, altruistic counterparts.
Bittensor’s survival hinges on aligning these divergent interests. That is, when the drive of individual miners’ to maximize their inflation is equivalent to the will of all other Tao holders to produce accessible and useful intelligence.
We expect that over the next few years Bittensor will require occasional changes to its emission rate and incentive landscape to ensure these two items are aligned. However, we do not believe that Bittensor should be governed in a centralized manner forever, as it is highly important that the protocol be resistant to the foundation itself, or any other small group of individuals.
In order to initiate this decentralization, we will shortly be requesting the top 12 delegates (as of writing) to vote on the proposal to increase emissions, in light of these new fixes, with more details coming in the next few days.
Thanks to everyone helping us build this future,
Best regards,
Const + Opentensor
# Where are we going next?
We are using the next week with emissions reduced to test the above additions to subnet 1’s validation mechanism. We are confident that these tests will be done by the end of this week. Once these tests are finalized, we will initiate a delegate vote on revamping emissions back up to normal levels.
# What does this all mean
The most important quality of a healthy system is that self-interests are aligned with the interest of the rest of the protocol’s user group as a whole. In Bittensor’s whitepaper, the term ‘cabal’ was used to define the state where self-interested individuals come to dominate the system, rather than their project-interested, altruistic counterparts.
Bittensor’s survival hinges on aligning these divergent interests. That is, when the drive of individual miners’ to maximize their inflation is equivalent to the will of all other Tao holders to produce accessible and useful intelligence.
We expect that over the next few years Bittensor will require occasional changes to its emission rate and incentive landscape to ensure these two items are aligned. However, we do not believe that Bittensor should be governed in a centralized manner forever, as it is highly important that the protocol be resistant to the foundation itself, or any other small group of individuals.
In order to initiate this decentralization, we will shortly be requesting the top 12 delegates (as of writing) to vote on the proposal to increase emissions, in light of these new fixes, with more details coming in the next few days.
Thanks to everyone helping us build this future,
Best regards,
Const + Opentensor
❤22
https://twitter.com/UniswapD/status/1625419292950020097 - a megathread to add to your bookmarks. People that describe or ask for Bittensor.
X (formerly Twitter)
Uniswap Deτecτive (τ,τ) on X
My timeline is filled with @bittensor_ shills.
It's just none of them know they're talking about Bittensor, yet.
It's just none of them know they're talking about Bittensor, yet.
Hello @everyone!
This Monday <t:1687816800> we will be holding a community call here in this discord to discuss when and how we will be resetting emissions back into the network.
Thank you to everyone for your patience during this maintenance.
Onward,
Const
This Monday <t:1687816800> we will be holding a community call here in this discord to discuss when and how we will be resetting emissions back into the network.
Thank you to everyone for your patience during this maintenance.
Onward,
Const
👍4❤3
Bitmart listed TAO out of themselves! - https://twitter.com/BitMartExchange/status/1673205956292001792?s=20
Announcement coming?
Announcement coming?
🔥14❤2
Hey @everyone
In preparation for our call this evening, we have increased the target registration rates from 2 registrations per block -> 4 registrations per block.
We will have more updates as we move along this week.
Don't forget our post mortem and future events community call here in the Discord this evening at <t:1687816800:T>.
See you then!
- Opentensor
In preparation for our call this evening, we have increased the target registration rates from 2 registrations per block -> 4 registrations per block.
We will have more updates as we move along this week.
Don't forget our post mortem and future events community call here in the Discord this evening at <t:1687816800:T>.
See you then!
- Opentensor
❤7
https://youtu.be/TOMLQxZk5QI - 16:10 bittensor
YouTube
🏆 TOP 2 EARLY STAGE A.I. ALTCOINS
A.I. is still the most popular narrative in crypto.
In today's episode, we take a look at the future of the sector and how it will impact the world of Web3.0.
Then, we discuss which 2 altcoins are best posiitons to lead the A.I. crypto industry moving…
In today's episode, we take a look at the future of the sector and how it will impact the world of Web3.0.
Then, we discuss which 2 altcoins are best posiitons to lead the A.I. crypto industry moving…
🔥4
Hey @everyone
Here are the major updates from yesterday's call with slight tweaks based on community feedback from last night.
Instead of Friday, we will run the DAO vote on 100% emissions and the POW registration removal for Tuesday next week. This will give delegates extra time to cast their votes and register into the DAO via instructions we will released tomorrow.
We will also be piloting the new validators with a new release this Thursday, so stay tuned.
- Opentensor
Here are the major updates from yesterday's call with slight tweaks based on community feedback from last night.
Instead of Friday, we will run the DAO vote on 100% emissions and the POW registration removal for Tuesday next week. This will give delegates extra time to cast their votes and register into the DAO via instructions we will released tomorrow.
We will also be piloting the new validators with a new release this Thursday, so stay tuned.
- Opentensor
👍11❤2
Hello @everyone
---
TL;DR
-- Emission are coming back at block 764202 (about <t:1688508900:f> your time)
-- Incoming Openvalidators update to 1.1.1
-- POW registrations are turned off on subnet 1.
---
Introduction
We would like to thank everyone in this community for patiently giving the Foundation time to implement fixes and align the incentive mechanism. We're happy to announce that we've made a tremendous amount of progress over the few weeks, improving many of the network fundamentals.
---
What has changed?
The largest and most important change is that we've managed to increase coldkey diversity substantially, reducing the main group from 300 keys down to 50. The blue line in Figure 1 (Impact of validators update on cabal) below highlights that group's hotkey count up until today.
---
How did we do this?
The major change was a substantial increase in the diversity of the validation dataset, making it much harder for cached responses. We can see this in the raw number of unique responses that subnet 1 is creating. The graph in Figure 2 (Histogram of diversity (unique rows / total rows) 25-06 to 30-06) highlights the percentage of diversity in the responses prior to our change. Figure 3 (Histogram of diversity (unique rows / total rows) starting from 29/06) shows the diversity today after the update. We are seeing over 80%.
This is good for two reasons:
1) It means that miners are running more unique models and
2) the network is producing a much more dense dataset product (useful for things like distillation).
See Figure 4 (Sum of raw collected rows and unique rows by day).
At the same time, naive exploits such as "thats an excellent question" or "this miner is under attack" are now getting near zero rewards across the mechanism. The network is much healthier from an output perspective. The top miners appear to be producing varied, relevant and quality responses, so intended incentivization seems working.
---
Additionals
- We are releasing a separate update to openvalidators which will drastically increase the importance of diversity across miners (specifically by up weighting the diversity filter). You can view this change here https://github.com/opentensor/validators/pull/86.
- We will be obsoleting POW registration along side the emission increase. This will make it impossible for anyone to farm registrations anymore, with a vote to turn it off completely incoming this week.
- Release 5.3 on Bittensor has been released that includes several fixes including various updates and fixes submitted by community members, such as asking for a wallet name during unstaking, updating versions and README.md, fixing tests and integration issues, removing neurons, merging pull requests, fixing weight utils, extracting configuration and wallet to new packages, integrating btcli with a new governance protocol.
We are happy with the outcome of the updates up until this point and as such have decided to follow through with the DAO vote emissions increase effective immediately. We hope that in the future community members will continue to help us improve the mechanism, and in the future we pledge to do our part in giving our ear to the community and move together towards true decentralized intelligence.
Happy mining,
Opentensor
---
TL;DR
-- Emission are coming back at block 764202 (about <t:1688508900:f> your time)
-- Incoming Openvalidators update to 1.1.1
-- POW registrations are turned off on subnet 1.
---
Introduction
We would like to thank everyone in this community for patiently giving the Foundation time to implement fixes and align the incentive mechanism. We're happy to announce that we've made a tremendous amount of progress over the few weeks, improving many of the network fundamentals.
---
What has changed?
The largest and most important change is that we've managed to increase coldkey diversity substantially, reducing the main group from 300 keys down to 50. The blue line in Figure 1 (Impact of validators update on cabal) below highlights that group's hotkey count up until today.
---
How did we do this?
The major change was a substantial increase in the diversity of the validation dataset, making it much harder for cached responses. We can see this in the raw number of unique responses that subnet 1 is creating. The graph in Figure 2 (Histogram of diversity (unique rows / total rows) 25-06 to 30-06) highlights the percentage of diversity in the responses prior to our change. Figure 3 (Histogram of diversity (unique rows / total rows) starting from 29/06) shows the diversity today after the update. We are seeing over 80%.
This is good for two reasons:
1) It means that miners are running more unique models and
2) the network is producing a much more dense dataset product (useful for things like distillation).
See Figure 4 (Sum of raw collected rows and unique rows by day).
At the same time, naive exploits such as "thats an excellent question" or "this miner is under attack" are now getting near zero rewards across the mechanism. The network is much healthier from an output perspective. The top miners appear to be producing varied, relevant and quality responses, so intended incentivization seems working.
---
Additionals
- We are releasing a separate update to openvalidators which will drastically increase the importance of diversity across miners (specifically by up weighting the diversity filter). You can view this change here https://github.com/opentensor/validators/pull/86.
- We will be obsoleting POW registration along side the emission increase. This will make it impossible for anyone to farm registrations anymore, with a vote to turn it off completely incoming this week.
- Release 5.3 on Bittensor has been released that includes several fixes including various updates and fixes submitted by community members, such as asking for a wallet name during unstaking, updating versions and README.md, fixing tests and integration issues, removing neurons, merging pull requests, fixing weight utils, extracting configuration and wallet to new packages, integrating btcli with a new governance protocol.
We are happy with the outcome of the updates up until this point and as such have decided to follow through with the DAO vote emissions increase effective immediately. We hope that in the future community members will continue to help us improve the mechanism, and in the future we pledge to do our part in giving our ear to the community and move together towards true decentralized intelligence.
Happy mining,
Opentensor
GitHub
Diversity filter model by steffencruz · Pull Request #86 · opentensor/validators
Tracked experiment with frequent weight and hotkey logging
Changes:
Move diversity model from rewards (weighted sum) to filter models (products) so it has more power over final score
Log all rewar...
Changes:
Move diversity model from rewards (weighted sum) to filter models (products) so it has more power over final score
Log all rewar...
🔥14❤3💩1