Selected messages from Bittensor subnet Discord channels, surfaced in one place. Click through to the original message for full context.
Want a daily summary instead? /subnet-news has one short brief per subnet per day.
<@1394258405400510464> but again, 107 uid >> why he got 100% emission now! But other miners 0, fock, i have 600Gh! Last 3 rounds I got emission and other miners. Why so fockin unstsble??? Hashrate no changed! Why emissiom is dancing from 0 to here ??? Please fix it and stop this fake miners!!!!!
Yes, got it as well, thank you. Less than the expected amount but I'll take it.
The mining data issue reported on the 30th has been fixed!
<@1394258405400510464> what about stolen doge why u stole doge from miners?
uid 32 was your too old uid
giving you ;last waring to make down your uid 32 and 242 how big scamme ru are
<@1394258405400510464> both are ur uid
Dear Community Members, Following a rigorous investigation by our technical team, we can confirm that the data for accounts 242 and 32 represents genuine and valid hash power, with no fraudulent or non-compliant activities detected. The platform has always adhered to the principles of fairness and transparency. Without conclusive evidence of violations, we will never arbitrarily restrict or freeze any user's account. Regarding the abnormal share and reward displays reported by some users over the weekend, this issue was escalated to our engineering department immediately. The technical team is currently fully committed to data verification and troubleshooting, and a proper solution will be implemented shortly. Please be patient as we await the official resolution. We sincerely appreciate and highly value the supervision and feedback from our community members toward the project's development. At the same time, we advocate for a rational and objective environment for communication. We hope everyone can base their inquiries on factual information and avoid unfounded speculation driven by disparities in data. For any legitimate issues within the system, the platform will take full responsibility and resolve them in a timely and efficient manner. Thank you for your understanding and continued support.
SHow here, not in DM
I will show u how u are cheating with 32
Could you please ask in the cacheon discord
https://discord.gg/cacheon
All good. They are on docker. Leaders and runner up images need to stay up since they will be re evaluated against new challengers
TP=8, preferably H100 or H200
0.1 TAO is not enough to cover GPU cost if the miner manage to stay at the top
if the economics doesnt work for your miner at the moment, feel free to wait for a week
the original decision was restarting on June 8 but some miners want to start now
<@1219180637039231037> https://discord.com/channels/799672011265015819/1503393871172866098/1508351427452600380 to avoid spamming and add skin in the game
then we can wait til june 8
so we know the exploit works on mainnet
Do you share validater code besides football event?
It seems I'm the only miner who occasionally gets a score of 0 due to the latency_pass issue.
Hi, <@760378675966902312> When will the road sign detection element be started?
<@760378675966902312>
If I hot the chutes first, wait for about an hour, and then commit on chain, will I avoid getting a score of 0 initially? To do that, what should I use instead of the `sv -v` command?
Due to this phenomenon, it is becoming virtually difficult for a new miner to overtake a long-established miner within 30 evaluations. This is because they are typically given a score of 0 at the start.
Hi, <@1001419867431378974> <@760378675966902312> Is it normal for the score for the first challenge to always be 0 after deploying the model to Chutes? Even though total_weighted is high, the composite_score is receiving 0 because latency_pass is set to false. Is there truly no solution to prevent this phenomenon?
<@1001419867431378974> the owner alpha token locked?
Hi <@760378675966902312> Car wash again not challenge. could you plz check this out?
Should be good now sorry for the inconvenience
<@1163660463792472184> I see my key on the website as a challenge winner, but my miner doesn't appear in the new miners tab. What should I do?
I've already completed all the tasks, but nothing has changed in the taostat table for me. How can I check the status of my miner, whether it is verified or not?
The plan is broad challenges, across tiers (I’m planning to launch 50+) We’ll have submit time outs too, so can’t spam the board with submissions. And winner takes all PER challenge. Feedback welcome!
Challenges are live, but need calibration, which I’m working on now. We should be nice and humming again in this evening.
1 day not changed challenge and emission? It okey?
That's on the pre-update scoring — v2.1.0 re-centers weights on recent verified solves, so inactive positions decay as validators adopt it.
Sure, will provide guidance, stabilizing validation first though
Yea I’m aware. Unfortunately miners with earlier agents are still getting rewarded. This is because their rewards was scheduled on a decay - I’ve not finished testing but I’ll go ahead and push the latest now and ask validators to pull, which should solve. Sorry guys
5CG4 - 350 wons = emission 0 5ert - 286 wons = emission 0 Its correct work?
There are 7-8 miners in your dashboard who are actively completing tasks. And those who win are emission 0. And the keys that I threw off, they have 10+ hotkeys and 0 wins.
GM miners! All miners with qualified/scored status should now receive the minimal weight so the hotkeys won't get deregistered.
Dashboard will be updated to show all 50 tasks in the competition section, but probably not today, since we have a lot of other work to be done, for now if you want to verify the score include the screener tasks too
<@1482050052242280489> So, the dashboard will be fixed?
<@690883684542447619> , plz check my dm..
what time slot does the owner team work in?
<@690883684542447619> New miners are registering and some screen-passed miners are going to be dereged. plz fix the minimum incentive mechanizm for screen-passed miners.
yout total token must smaller than baseline total token.
qwen/qwen3-coder:free how about use this free version for eval. <@690883684542447619>
Is it ok for you, <@1486394508298944717> ?
<@690883684542447619>
Affine's subnet will blow its first candle this week, as it has been announced by Const during the last Proof of Talk. Any plans for a little Affine birthday party? 🙂
can you check my dm?
<@1091660419023773828>
SWE sample count is 216?
Hi <@1091660419023773828> Is DQ system working now?
Good speed. Thank you. <@1091660419023773828>
/\ /\ /\ /\ /\ /\
system prompt is not legit
``` "messages": [ { "role": "system", "content": "/root/NAVWORLD/navworld/affent-workspace/fc764cb5/system_prompt.txt\n\nRuntime context:\n- Current UTC date: 2026-05-29.\n- Current UTC time: 2026-05-29T18:48:13Z. Use this timestamp for relative timer and reminder calculations.\n- For current/latest/market/news facts, use this as the access date only when a source lacks its own timestamp. Do not invent source dates; distinguish source publication/update dates from access date." }, { "role": "user", "content": "请为我规划一次青岛6日游，出行人员：2位老年人，出发日期：2026-02-13，总预算：1800元。\n\n兴趣偏好：海滨度假, 文化历史, 登山徒步, 赏花观鸟\n特殊要求：优先景区免费开放日, 预算优先, 公共交通优先, 速度优先, 深度体验, 轻松休闲\n\n请提供详细的每日行程安排，包括：\n1. 每天的景点安排和路线规划（每天至少2-3个景点，标注门票价格）\n2. 每天的餐饮推荐（具体餐厅名称和人均消费）\n3. 住宿建议（具体酒店名称、价格区间和位置优势）\n4. 各景点间的交通方式、距离和预计时间\n5. 每日花费明细和总预算分配" }, { "role": "assistant", "content": "\u003cthink\u003e\n\n\u003c/think\u003e\n\n", "tool_calls": [ { "id": "chatcmpl-tool-8b8f0060d9 .... ``` some issues on NAVWORLD
<@1091660419023773828> pls check my DM
Negative LCB score. We'll update dashboard today - it's misleading now.
https://x.com/datacurve/status/2060834005998793199?s=46
Will be fine right now
<@1506959452438401044> validator works now?
I will look for that
Gm. <@1506959452438401044> , <@1245339055005110437> . the eval gpu has some issue. please check
albedo's commitments
Noise testing now
` albedo/albedo-mini-1.7b-noise1`
Hi <@776889609791209494> Already pushed the big updates in validator?
<:Lovee:1124910364035907696>
all endpoints api endpoints accessible from there
clever filter > gpu power
refer to the notes above to properly use the /health and /verdicts endpoints to understand your rollouts in detail. You must score locally against them to make sure you only submit what you’re 100% sure is in the zone
Beyond plain pass/fail we can shape it (fraction of tests passed, partial credit, etc) to give a smoother gradient signal
As you can see, the more windows that don’t get filled, the better the model is and the harder it is to train. We jumped to 18% burn, meaning right now an average of ~20% of windows aren’t getting filled. That means training is almost done and it’s getting harder to generate false answers. Super good sign! We strongly believe this is the right design : it avoids running infinite competitions. So at this point it’s probably better to focus on deeply understanding the edge that the top miners have with the prompt selection
**It's the textbook outcome the README promises: ***"a miner who picks well lands on winning prompts and earns emission; a miner who picks poorly burns their rollouts." * The hardening sharpened that selection, and one operator with better tuned k=6 policy is collecting the spoils. Working as designed, **not exploited**
I think validators are scoring by group. so group A and group B miners are updated frequently, but group C miners are not.
<@865127800121327618>
And if I submit bad models for the first a few cycles, then it's impossible to get scored or promoted even submit the best models for the next cycles?
I already had a deep understanding about the scoring mecanism so made this dashboard by scoring system. If you carefully take a look the history of last 8 cycles, then you can also find some issue means - I'm not complaining, just wanted to suggest since I found critical issues against with the scoring mechanism.
and it only jumped to top upon c15829
<@865127800121327618> if you want, I can share my leaderboard with you.
hey hey, please refer to our in-house dashboard for the metric https://dashboard-dev.connito.ai/ 👍 since these dosent seems to be a problem there
I think a lot of good models didn't evaled or scored for whole 8 rounds. Can you take a look once if the scoring system works smoothly?
And as you can see UID 12 and 158 here, some models score never updated, while other lower val_loss models didn't get scored even.
And this is val_loss sorted table, as you can see here a lot of models didn't get score even their val_loss is lower than scored models continously.
I see. Thank you for your kind explaination
Don't worry for it.
Moreover, since this agent must be used as the base agent, I would like to make this clear.
As I said, I want to know the rule more correctly.
This is merely my opinion, and it is not intended to criticize the top agent.
Since the validity of this command varies depending on the system...
Since I am not yet fully familiar with the regulations of this subnet, I am unsure whether this level of hard-coding is permissible.
But he is still getting emission as past king
We'll do our best to dive in. It's also late here for most of our team, so please be patient.
It is not warming up for more than 2 hours
Could you have a look at my chute?
<@1414611260040810586>
Ill see if anyone esle has one available to activate
I see launch config for both of these that were just claimed. Seeing if they come up
Ok I’m looking but it seems like we may be out of Pro 6000s at the moment
You see the bounty increasing and the different colored spaces on the terminal etc?
``` chutes warmup chute_ID ```
Ok and for both of you when you run
<@1483242954339385344> I checked inference status by using "oro inference status" and the result is "Provider Connected Default Updated ──────────────────────────────────────────────────────────── chutes yes yes 2026-05-29T20:15:52.290325" But, in leaderboard agent status page, "Miner has no inference token — cannot fund inference" issue was shown. What's the reason?
<@1483242954339385344> I used Option A: SSH port forwarding. But finally, "httpx.HTTPStatusError: Client error '400 Bad Request' for url 'https://api.chutes.ai/idp/token'" error message showed. Please tell me how I can fix this issue.
```Traceback (most recent call last): File "/root/streak/.venv/bin/oro", line 6, in <module> sys.exit(main()) File "/root/streak/.venv/lib/python3.10/site-packages/oro_sdk/cli.py", line 553, in main args.func(args) File "/root/streak/.venv/lib/python3.10/site-packages/oro_sdk/cli.py", line 297, in _inference_connect tokens = start_auth_flow() File "/root/streak/.venv/lib/python3.10/site-packages/oro_sdk/chutes_auth.py", line 200, in start_auth_flow tokens = exchange_code( File "/root/streak/.venv/lib/python3.10/site-packages/oro_sdk/chutes_auth.py", line 97, in exchange_code response.raise_for_status() File "/root/streak/.venv/lib/python3.10/site-packages/httpx/_models.py", line 829, in raise_for_status raise HTTPStatusError(message, request=request, response=self) httpx.HTTPStatusError: Client error '400 Bad Request' for url 'https://api.chutes.ai/idp/token' For more information check: https://developer.mozilla.org/en-US/docs/Web/HTTP/Status/400```
On devices without VPN, ```oro inference connect chutes``` work correctly?
<@1483242954339385344> On VPN, can't submit agent using CLI?
<@1483242954339385344> ```oro inference connect chutes``` This opens up a web page and I allowed and returned to my terminal where this error comes out. ```Exchanging authorization code for tokens... Traceback (most recent call last): File "/root/streak/.venv/bin/oro", line 6, in <module> sys.exit(main()) File "/root/streak/.venv/lib/python3.10/site-packages/oro_sdk/cli.py", line 553, in main args.func(args) File "/root/streak/.venv/lib/python3.10/site-packages/oro_sdk/cli.py", line 297, in _inference_connect tokens = start_auth_flow() File "/root/streak/.venv/lib/python3.10/site-packages/oro_sdk/chutes_auth.py", line 200, in start_auth_flow tokens = exchange_code( File "/root/streak/.venv/lib/python3.10/site-packages/oro_sdk/chutes_auth.py", line 97, in exchange_code response.raise_for_status() File "/root/streak/.venv/lib/python3.10/site-packages/httpx/_models.py", line 829, in raise_for_status raise HTTPStatusError(message, request=request, response=self) httpx.HTTPStatusError: Client error '400 Bad Request' for url 'https://api.chutes.ai/idp/token' For more information check: https://developer.mozilla.org/en-US/docs/Web/HTTP/Status/400``` How to fix this?
<@&1483377890166050989> ``` oro inference connect chutes``` This opens a web page and I allowed and returned to my terminal where this error comes out. ```Opening browser for Chutes authentication... If the browser doesn't open, visit: https://api.chutes.ai/idp/authorize?response_type=code&client_id=cid_qstkn2eb4uszxski1rdqk062&redirect_uri=http%3A%2F%2Flocalhost%3A8765%2Fcallback&code_challenge=z4H8aVPI-lJL5j085lTKpNWUHPdB5jVbNJiXtbNKZrY&code_challenge_method=S256&state=EHEjoP9i6ML6o2GU-mUXsY79aREn4O7Z_Z0K8sEhC2Q&scope=chutes%3Ainvoke Exchanging authorization code for tokens... Traceback (most recent call last): File "/root/streak/.venv/bin/oro", line 6, in <module> sys.exit(main()) File "/root/streak/.venv/lib/python3.10/site-packages/oro_sdk/cli.py", line 553, in main args.func(args) File "/root/streak/.venv/lib/python3.10/site-packages/oro_sdk/cli.py", line 297, in _inference_connect tokens = start_auth_flow() File "/root/streak/.venv/lib/python3.10/site-packages/oro_sdk/chutes_auth.py", line 200, in start_auth_flow tokens = exchange_code( File "/root/streak/.venv/lib/python3.10/site-packages/oro_sdk/chutes_auth.py", line 97, in exchange_code response.raise_for_status() File "/root/streak/.venv/lib/python3.10/site-packages/httpx/_models.py", line 829, in raise_for_status raise HTTPStatusError(message, request=request, response=self) httpx.HTTPStatusError: Client error '400 Bad Request' for url 'https://api.chutes.ai/idp/token' For more information check: https://developer.mozilla.org/en-US/docs/Web/HTTP/Status/400```
<@1483242954339385344> how to become a validator in this subnet?
<@1483242954339385344> If I submit my agent and get error of "similarity", is there any way to know which agent is similar to mine?
Do I really need to u se jq?
Let me review what exactly happened. We are doing none of the things you mentioned. but i got your message.
This subnet is unlike any other subnet. Orchestration is a beast. That why you guys have a huge advantage compared to the devs/miners coming in the next few months.
why don't you hire someone to fix code
Thank you for informing us about these issues. We will reply during business hours.
WS: wss://orch-gateway.b1m.ai/ws/orchestrators/<hotkey> Auth: X-Api-Key authorized after handshake (obtained via /auth/challenge + /auth/verify, role orchestrator) { "type": "gateway_update", "gateway_url": "http://209.200.246.39:11001", "max_workers": 10000, "health": "healthy", "request_id": "<uuid-hex>" } Response received: { "type": "error", "error": "internal error", "request_id": "<same uuid-hex>" }
Here is detail of error. ``` WS: wss://orch-gateway.b1m.ai/ws/orchestrators/<hotkey> Auth: X-Api-Key authorized after handshake (obtained via /auth/challenge + /auth/verify, role orchestrator) { "type": "gateway_update", "gateway_url": "http://209.200.246.39:11001", "max_workers": 10000, "health": "healthy", "request_id": "<uuid-hex>" } Response received: { "type": "error", "error": "internal error", "request_id": "<same uuid-hex>" } ```
<@847938562221408287> Please check gateway update api in beam core side. I don't think it is working correctly
Info: Will be at Proof of Talk this week. I'll relay your salutations to the other founders.
No more Hotfix! Things are serious now!
Maximum 7 days?
There is wandb where you can see the scores validators set for miners, there is the good old Bactensor and the metagraph but that's it!
Got it. I'll fiddle around on testnet, is there a miner dashboard for this sn?
It's your job as a miner to figure that out
For the optimization there are many things you can do, the default miner as is is not going to be competitive so you have to innovate on something like the model, the prompting, prompt chaining, etc
There is also the testnet on uid 148 if you want to practice setting it up beforehand
Cost of mining and entry is very low at the moment so you could easily try!
Even more today! I'll do periodic spotchecks but feel free to ping me if you see anything weird!
It's getting better. Lots of miners got proper incentives not long ago!
<@&1237015679023059017>
https://x.com/mtslive/status/2060447529133867339?s=46
basically avoids your exact situation of submitting but having a screwed up config
well if your api key is not valid it wont work. I don't know what you sent and I'm not a team member or anything so can't check:D
don't pay metered usage!
I've been usind it this way in on the fly sandboxes and all sorts of other places:D
you can generate a long lived oath token on any machine were you're signed in just run" `claude setup-token` https://code.claude.com/docs/en/authentication#authentication-precedence
and make sure you don't also include ANTHROPIC_API_KEY=sk-ant-api03 in the env the claude code runtime will default to the key first
``` CLAUDE_CODE_OAUTH_TOKEN=sk-ant-oat01-xxxxxx ```
i think for next round we'll move off claude and use something else
<@263544733442244609> for next round please can you add support for generating embeddings via openrouter? https://openrouter.ai/models?output_modalities=embeddings
gold move as well pinging the boss for your rant:D
sure . sure . sure.. 1000 %
will play with this until this sn dereged.
besides, i also lost few taos for reg fee and lost my time.
and finding negative things like this is what i like most.
hmm. it's end of month so a bit chilling .
they bought alt acc?? 😏
The king is mining in sn65, sn62.
<@1502006453018558566> let's do find other subnet to mine
<@1125803164021559306> if you want the substantial development of subnet, plz make things transparent miners are never engaged in this subnet for now and that means that subnet is useless no miner, no future in that subnet, 😎
Thing is this is too obvious, they even didn't buy another discord acc.
Const conviction 👀🤝🐐
<@1394593530680447027>
these are not triggered yet
All GPU OOM will be re-evaluated
I can find it here
<@389189199514959893> Is it alright to remove the king's repo?
`erenhex` is `scoutminer` and `Kyle` is `mastertensor`. You are senior miners in teutonic. 👍
<@1310808125216915481> Are u mastertensor?
I see, it’s a hf repo
🤔I thought you are scoutminer
I see what you are saying. I'll look into this
But can't see it from page 1 and page 2
For example:
And there are some 24s that is not displayed in the dashboard.
These are all in page 1
from page 2:
I am able to. What are you seeing in page 2?
Maybe dashboard bug?
https://discord.com/channels/799672011265015819/1437447445176127618/1508918693936238762
<@1500876249063624715> best if you read the Challenge Card here - ψ・trishool・23 . It will answer most of your questions
I don't mean top miner targeted that kind of trick, but I think it's better to update that eval part. <@1471668391646597325>
magma is chutes expert. lol
But I don't think that's the reason you can't beat the top model
Whole validation & mining process is working well. But I found an issue. When a miner's chutes is cold, valis check chute status and set invalid for that miner. And the number of sampling is decreased over time, but the score is being remained. So some miners make their chutes cold when the inference is bad and rewarm after a few hours to keep score. For model upgrade and fair competition, the score should decrease as well like not solve tasks when chutes is cold.
Hi, <@1471668391646597325> Whole validation & mining process is working well. But I found an issue. When a miner's chutes is cold, valis check chute status and set invalid for that miner. And the number of sampling is decreased over time, but the score is being remained. So some miners make their chutes cold when the inference is bad and rewarm after a few hours to keep score. For model upgrade and fair competition, the score should decrease as well like not solve tasks when chutes is cold.
<@1471668391646597325> Why did you remove FR?
I have sth to ask you via DM
Hi <@1471668391646597325>
Okay, it's fixxed now
<@348364373007532033> tell me about this subnet
<@348364373007532033> what are you going to do for conviction? lock or not?
gm wen winner gets max 11 score. Puuuuuuuush
Anyway it's obvious that current winner is unbeatable
Seems a week is too long...
Looks like current score is unbeatable. <@348364373007532033> , <@956468453848522782> could you please add new scenarios?
<@348364373007532033>
<@348364373007532033> when will add new scenario?
<@956468453848522782> when validation will be fixed?
comparison log , what is that? that is cause of error? { "144_gAAAAABp9R":{ "reason":"Fundamentally different algorithmic approaches: reference uses a statistical model, miner uses a rule-based heuristic system." "similarity_score":0.1 } "167_gAAAAABp83":{ "reason":"Miner uses extensive preprocessing, feature engineering, and multiple rule-based heuristics, unlike reference's simple decision tree." "similarity_score":0.35 } "230_gAAAAABp8z":{ "reason":"Miner uses a complex multi-heuristic model with extensive preprocessing; reference uses a simple two-feature rule." "similarity_score":0.25 } "4_gAAAAABp9X":{ "reason":"Miner uses robust feature extraction & different heuristic rules; reference includes a distinct decision tree model." "similarity_score":0.35 } "61_gAAAAABqEU":{ "reason":"Fundamentally different detection approaches: reference uses a probabilistic membership sketch, miner uses a multi-branch rule-based system with extensive feature engineering." "similarity_score":0.1 } "baseline_script_baseline_1.py":{ "reason":"Completely different detection heuristics and logic, despite using some common base features." "similarity_score":0.35 } }
hey <@629698366053416978> There is 2 active challenges right now and you can freely start mining those challenges.
cd ./examples/miner_commit dosen't exit miner-commit-api there is only miner-api in the compose.yml
in fingerprint challenge
Do you think this subnet has a bright future?
Please check these documentations: - **[About validation checks](https://docs.theredteam.io/latest/miner/concepts/validation/#code-linting)** - **[Flowradar submission guide](https://github.com/RedTeamSubnet/flowradar-challenge/blob/main/examples/miner_commit/README.md)** - **[Historical fingerprinter submission guide](https://github.com/RedTeamSubnet/historical-fingerprinter-challenge/blob/main/examples/miner_commit/README.md)**
miner number 61 is one of validators?
Owned by us, Yuma are a partner of ours
thanks, fair concern. just to be clear, we are not self-mining. quasar is just a different architecture, with a different attention, looped-transformer setup, so a lot of standard qwen-style tricks may not work directly. that's why i keep pushing people to study the architecture and train against the actual gates, not just copy old recipes. also, we are not trying to use miners as test resources. the goal is the opposite, make the scoring clearer and help miners actually get models through and to prove that, we will discuss the threshold and scoring with everyone here on wednesday, using the latest results, and agree on the clean path forward together.
also, we’re verifying scores from yesterday through 00:00 utc to see if the updated crown-quality gate is now too hard. if it is, we'll ask for miner feedback before adjusting it. we're also preparing a bigger data update. my advice meanwhile to do real training and understand the quasar architecture. it’s not just qwen, so some qwen-style tricks may need changes for quasar attention/architecture.
we found a small bug and pushed a validator scoring update. crown-quality/gate should be more stable now, with less variance from benchmark sampling. no policy change: models still need to clear kl + crown-quality normally.
Would it be possible at some point to put out like a weekly update of progress for simpletons like me. I’m so far as the progress of the model and the improvement of training?
`Olague-Secret` who is this guy? nah, he has same name models corresponding to my model. apple to apple, 🤦
Stop letting unmanaged subnets run wild. Repeated COAs and unchecked airdrops are creating systemic risk and destroying TAO’s credibility. Fix onboarding now: require a governance spec, public tokenomics, and a security/rollback plan before any subnet goes live. Enforce consequences for repeat offenders. This is a protocol‑health emergency. Lead or step aside.
One thing id add, and this could apply to most of subnets but will post here. Might be age thing or lack of corporate work experience. If you’re building a legitimate Bittensor subnet, publish this the same way Coinbase, Kraken, or any regulated financial service would: official docs, blog post, or announcement channel with proper visibility and permanent record. Otherwise people will just spam discord, tag on X etc
Either way, subnet 24 is less than 1 year old and it has no effect on ownership
Hi, when update of veAlpha logic and lock in conviction ?
postive updates by both! sounds good
Exchange go-live is moved to Jun 30 for several reasons 1. Requires team restructure and resourcing linked to Taoshi/Vanta departure. 2. Audit is in progress and will run for 30 days. The additional time allows for initial findings to be received and implemented before full audit completion and therefore safety for users. 3. Conviction requires update of our veAlpha logic which is key to mainnet.
Hi All, 1. Subnet ownership is with GT 2. Yes, Taoshi/Vanta team are rightfully focusing on their immediate priorities and are not part of 0xMarkets project going forward. 3. We will solve conviction asap now that it's live. It's also linked to our veAlpha design which we'll update accordingly, no issues there at all.
<@797206988978651156> this subnet is very interesting. Mainnet is live now?
if no lock in conviction - am out, simple as that.
its rising steadly
<@1118493701405163611> Hey! Env+launch now work for my existing subs (36, 37) — thanks for the update. But I'm getting 401 on POST /challenges/agent-challenge/submissions (newline-signed /submissions) and 404 on POST /v1/challenges/agent-challenge/submissions (raw-zip bridge) at chain.platform.network. What's the working URL/method for new miner submissions right now? Need to push v2 of my agent.
<@&1432376959337238659> 🚀 **June 1st Update** Today, we are activating Conviction and locking all of our alphas as part of our long-term commitment to the ecosystem. The Agent Challenge evaluation pipeline is now fully operational. Emissions will be activated, and evaluations can begin immediately. If you have submitted an agent, please follow the Agent Challenge documentation and submit your environments. Once your environment is submitted and validated, the evaluation process will start automatically using the provided API key. We are also reactivating Atlas today and rolling out updates to the Cortex Foundation website.
For miners who have submitted an agent, please send your environments following the Agent Challenge documentation so that your miners can use the provided API key.
and we are updating the Cortex Foundation website today
we can start the agent challenge evaluations
<@1118493701405163611> Any update on conviction locking ?
they delivered on 22 miners relaunch
https://giphy.com/gifs/mfw-everything-reddit-oYtVHSxngR3lC
Congrats, big proof of conviction 🥩
<@&1220505352953528380> Subnet 2 has locked all owner alpha perpetually, including emissions
That's correct yes, partially due to the fact that some layers aren't tiled, so they produce huge proving times while the tiled layers execute in ms
(It's important that if `us-west-2` goes down for example there are enough globally distributed miners to continue normal proof serving operations)
Additionally, ensure your miner IP address is in a region that's distinct from other miners. We do enforce caps on origin regions because if the miner base is highly concentrated to one area, the subnet itself sees increased geographic risk which is not ideal
To verify we have a public dashboard showing all miner metrics here https://sn2-stats.inferencelabs.com
the dev team has already explained why run7 is being re-evaluated
The sooner the better. Lets not wait for months like with Daryxx..
Come on man. Lets keep talking. I can be relentless once you get on my radar and do dirty sh*t, Daryxx got what he deserved and rightfully so. He is still extracting as we speak. 36, 70 and 99 are on my radar too.
<@836362081422540864> where you at? How do you just remove the owner of three subs without a single word?
winner takes all means exactly that: you either win or lose. read the code. it tells you everything you need to know.
Are they auditable ?
For TorchScript submissions, does this subnet require inference to be stateless (same output for the same input, with no internal buffer updates during forward)? My buffer-based solution failed with “Evaluation Error: Model failed screener validation.”
my UID is 186. how can i use the cli? could you give me a favor? <@694241943277928488>
hi where is owner?
no ending competition.
current competitions are bugs now.
<@1265230059246194690> , <@496509723646885888> , <@694241943277928488> , plz check iota competition. why don't start new competition?
GM <@&1346133673451327542> ☀️ Compound config update: ⚛️ Molecule target: **Q9BY41** 🧪 Allowed reactions/banned atoms: **no change** 👾 Nanobody target: **no change** BUT: ⚠️ MSA file for P05231 will be updated, which will likely change folding behavior Both MSA files (updated P05231 + Q9BY41) are available at: https://github.com/metanova-labs/nova/tree/update_msa_P05231
looks like it was transitory, it's all back now
I'll check if this is on our side or theirs
Just restarted it, looks like something was wrong with uniprot API
I’m not sure I understand your question but if you mean NOVA being correlated to TAO then no. The contrary
yes we will be locking owner alpha. team is in different time zones rn (it’s very late night for some). we’re finishing code review and aligning on the final details to announce
``` score = score * max(1 - sum(short or cash alloc), 0.01) ```
Here is the current formula: ``` score = score * max(1 - cash alloc, 0.01) ```
100%. Investing is a long game
<@&1358857675181658263> Miners: To improve performance in stocks, short positions will be counted as cash in adjusted scores starting next week. Please change strategies accordingly if needed
https://db.investing88.ai/
How is this score calculated? It looks like the scores on the dashboard are quite similar.
Hi <@1099611667081482250> if you want to chat, please ping me here.
Hi is there anyone to run miner now? I am facing strange issues and want to discuss here.
<@389189199514959893>
The holders of the SN59 token are very curious about whether the team is developing in accordance with your direction. We are not speculators, but rather a confirmation of the team's response, as well as our trust in Mog. If things do not develop as expected, we have the right to obtain information on the public channel. We hope you can pay attention to this.@consτ [τ, τ]
Hi <@1339902818571780173> validators is working now? and emission is also working?
Hi <@1339902818571780173> currently, dashboard seems not working. could you check it out, please?
Hi <@276162484086505474> , Happy Friday! Could you check my DM, please?
# Gasbench Cycle 17 Discriminator Winners ⛽ **📸 Image Winner (General)** **Winning Run**: [Dashboard Link](https://competition.bitmind.ai/miners/5Fvrq9d6Sg5Rb2xShjN6F6CwQCXQMopzLVkYMBoW98zKiR5m?modality=image&vertical=general&benchmark_version=17&run_id=8a87c2e5-b3eb-4ff3-95ed-d1321b35bd7a) **Hotkey**: 5Fvrq9d6Sg5Rb2xShjN6F6CwQCXQMopzLVkYMBoW98zKiR5m **SN34-Score**: 96.0229 **Alpha**: 6423.86 י **🎥 Video Winner (General)** **Winning Run**: [Dashboard Link](https://competition.bitmind.ai/miners/5Eh4ihPcYfJ4DYf232j99G5pH4zoLSoHZz7NPuLtWh4mJ5dq?modality=video&vertical=general&benchmark_version=17&run_id=4f4637c1-9121-401d-bddb-6db3ba6232b3) **Hotkey**: 5Eh4ihPcYfJ4DYf232j99G5pH4zoLSoHZz7NPuLtWh4mJ5dq **SN34-Score**: 97.4426 **Alpha**: 6483.13 י
The last couple rounds have converged to miners competing for VERY fine margins. Aiming to make next round more interesting by introducing additional axes, potentially latency-based or feature space analysis based.
Wow, incredibly high perf across the board this round. Still waiting on a few remaining posthoc analysis runs but the leaderboard has been released
The round ended last night, and needed to wait ~4-5 hours for the final runs to complete before kicking off winner anlaysis this morning. This is nearly done now, announcements coming within the hour 🫡
The round ended almost a day ago already, when will we get the leaderboard and the announcement for the winners? <@276162484086505474> <@473634651995635722>
Hi <@641922223908847617> there happens new kind of error: "0|miner_po | 2026-06-01 11:18:25.515 | ERROR | UnknownSynapseError#73918880-f840-4463-a1a4-745d688b1fd0: Synapse name 'mcp-sse' not found. Available synapses ['Synapse', 'MarketSimulationStateUpdate', 'EventNotification', 'GenTRXAssignment']". what is this?
Keep pushing guys 👌 https://x.com/i/status/2060493568242745668
https://tenor.com/view/matrix-you-have-to-let-it-all-go-neo-fear-gif-16229605
<@&1356318576386244648> # CONVICTION We have just perpetually locked all of our 1,269,236.79 SN79 alpha. https://www.tao.app/extrinsics/8291035-35
## **GenTRX Testnet Update | v0.4.4** Bittensor 10.4 upgrade, reliability work, and a smoother setup experience. If you're already running, pull and restart. ### ✅ Reliability - Main-cycle latency hardening on the validator side - Reduced per-cycle network overhead - Bittensor 10.4 upgrade ### 💾 Gradient Server - Bounded memory under load - More predictable flush behaviour - Network-aware S3 prefix selection ### 📊 Simulator - Startup optimisations ### 🔍 Observability - Dashboard trade display fix - trades now show again - Terminal view returning shortly Code and setup: https://github.com/taos-im/sn-79/tree/test ### ⛏️ Miners No required changes. Setup is smoother across more environments and a couple of rough edges have been resolved. Opting in should be a noticeably better experience this release. **Mainnet upgrade lands Monday.**
Benchmark rotates daily
We have pushed an update to clean up the general scraper output. lmk if u face any other issue
Let me tell you a basic flaw for evidence you pick first 2.5k characters and that too from raw html. For more website this covers mainly headers.. WHy even this basic thing is fixed?
<@1444402024245624907> i have shared you and <@301945119454658569> both these proofs multiple times but you accknowledged but have not fixed it.
Subnet is active
good team i love both subnets
u built sn74 too right?
🚨 Code update 🚨 <@&1437497690077397065> v1.0.8 **allways network resumed again** ---------- Summary **Credibility now penalizes repeat timeouts** • Credibility used to only ramp up. Timeouts counted as observations, so a high-volume miner could pile up timed-out swaps and still keep a full 1.00x credibility multiplier. • Added a rolling-window hard floor. More than 2 timed-out swaps inside the credibility window zeros your credibility, and with it your whole reward, until those timeouts age out of the window. • The scoring trace mirrors the zero, so the validator log matches what you actually earn. **Closing free-rider and sentinel-rate holes** • A miner could build credibility on a sane rate, then overwrite their commitment with garbage the parser rejects (wrong version, NaN, malformed, unsupported chain). The old good rate stayed live in scoring and kept earning while the miner was unreachable. That stale rate now gets terminated as soon as the commitment stops backing it. • A miner could also post a rate whose smallest legal swap is bigger than their own collateral, hold crown, and earn while every real swap against them gets rejected for lack of funds. Now, per block, a miner whose rate forces a TAO leg larger than their collateral earns no crown and it cascades to the next-best rate. Normal rates are unaffected. • Unexecutable and sentinel rates are now filtered at the parser layer, so the validator never scores them. • Reserve and activate now check the quoted rate against swap bounds, so a sentinel-rate poster can't reserve or re-activate. **Capacity now reflects the whole window** • Capacity weighting used to multiply your crown time by your collateral read once, at scoring time. You could hold crown with low collateral, then top up right before scoring to get full capacity credit retroactively. • Collateral is now replayed per block from the on-chain events, so capacity credit matches the collateral you actually held across the window. **Validator robustness** • Serialized axon and subtensor bounds reads under one lock to remove concurrent-read races. **Logging** • Validators now log the current crown holder UID and rate per direction each forward step, whether or not DB writes are on. **If you let swaps time out, your rewards will shift starting next round.** The 3rd timed-out swap inside the credibility window now zeros your reward until those timeouts age out. Keep your timeout count down. ---------- Links - site https://all-ways.io/ - repository https://github.com/entrius/allways/tree/main - docs https://docs.all-ways.io/ thanks yall!
hi <@&1437497690077397065> I'm going to pause the allways network for a bit while I push some updates to prod this means all active swaps will finish normally and no new reservations/swaps can be initiated until I resume the allways network it won't take too long, I'll have the changelog announcement when I resume things again, hang tight thanks!
hey all merged a small allways patch today here's the deets • Rates whose only in-bounds source amount sits below the source chain's dust / existential floor no longer pass `is_executable_rate`. These quotes are unfundable on-chain, so they no longer earn rewards. • Added a per-chain minimum on-chain amount floor in `chains.py`: BTC 1000 sat, TAO 500 rao. BTC uses 1000 rather than the bare 546 dust line for margin against higher fees and wallet quirks. • tao->btc is covered by the same change through the existing inverse-symmetry path.
Yep, you would have to ask the individuals who make those claims.
our winners are constantly changing - everything is shown in the dashboard.
We have no incentive to self-mine. We've asked them to show the coldkeys but they haven't.
`some miners prove that urs validators and miners has same coldkey.` show me where they've proven that. Or feel free to share their coldkey yourself.
Hey there, yeah you’re eligible now and our validators are scoring you consistently every round. Your scores just aren’t high enough yet to break into the top 10 and receive emissions. You can check your miner stats here in our dashboard: https://theminos.ai/dashboard/miners/5Gh9UCJgzBEYewVLfkMoJa2wkEbNYhK9599E2PFqVPPSvXtb More importantly, since you aren't doing this manually, I’d ask your Claw to start doing a few things systematically: 1. Track previous config submissions. 2. Track previous BAM inputs. 3. Use the API to pull your historical performance scores. 4. Use all of that over time to learn which configs/input patterns led to higher or lower scores, to automatically improve future submissions based on your historical performance and genomic context.
<@1087423076414787705> any update on conviction ?
Also, while we're looking at this screenshot, that's where it would be cool to see which GPU is serving this specific model/instance. Just a thought
in other words, it should be much easier to provide idle compute imho
like just to get a miner up i need to think about preparing my own edge proxy, to secure the gpu servers, etc. it's really painful to do for very little reward right now
to be fair the only thing that seems suboptimal as i familiarise with code base is that miners need to implement their own anti-ddos, https routing and all of that which is very hard to do at scale so if you wanna scale i think this is something you should manage to be sure it's perfect accross all your available gpu fleet
maybe it's possible. but training per 15 mins looks a bit strange. personally, I think it doesn't make best model.
full conviction for our target 🔥
Conviction is great! Full position! Bravo!
Hi, <@1450512218390134816>
please check if you pass quality gate on wandb log or not
https://tenor.com/view/extraction-netflix-gif-14555251113913748316
https://tenor.com/view/extraction-2-extraction2-tyler-rake-chris-hemsworth-netflix-gif-3817085300927126573
https://tenor.com/view/psy-gif-9197394
https://tenor.com/view/hindu-gif-9629806
https://tenor.com/view/lock-em-up-lock-him-up-lockemup-trespass-tresspassing-gif-11340568537533511165
The core problem here is how to value a repo (specific piece of software)...it's the same class of problem as bittensor's "how to value a subnet." We've said since the start of the subnet basically that the answer is let alpha holders drive emissions directly. We recently enabled per-repo emissions (about a week ago), and we're getting close to the point where we can hand this over to alpha holders entirely.
I think the fix is in how the shares are set, not in the per repo model itself: Detect and demote captured repos, using signals like one account getting most merges, few outsiders accepted, or linked accounts. Cut their share, down to zero if needed. We have halved repos before, so the mechanism already exists. Scale a repo's share with its number of real, independent contributors. Right now a repo's share is set by hand and stays fixed, so it says nothing about whether the repo is actually open and active. The number of independent miners getting merged is a much better signal, because it is exactly what separates a healthy repo from a captured one. A captured repo only ever has one or two accounts getting merged, so it would sit at a low share. A genuinely open, fairly run repo has many different miners landing PRs, so its share would rise to match the real activity. We can lean on the eligibility and credibility checks we already have, so only contributors who pass those count, and the growth should be capped so no single repo can take over. The effect is that gamed repos naturally shrink while active fair ones grow, and the share starts reflecting real participation instead of a number someone set once. Rebalance shares toward the repos that actually matter, and revisit it regularly instead of setting it once. This way the reward follows real work and real value, and there is no longer much point in running a captured repo. I'm curious whether others are seeing the same thing.
Hello everyone, I want to raise a fairness problem with how rewards are split per repo. The main issue is the fixed share. In a well run repo where the owner merges fairly, lots of strong miners show up and compete. Because that repo's share is fixed and divided among everyone, even miners who contribute a lot of real, high quality code end up with a tiny slice. The harder a good repo is and the more serious people it attracts, the less each person earns. That is backwards. It punishes exactly the repos and the miners we should want most. At the same time, simple repos with little of value keep collecting the same steady reward no matter what, while genuinely useful repos stay capped low. So effort and value are not what decides the payout. It also gets abused. Since an owner's own PRs aren't scored, some owners just merge a friend or a second account they control, and that account quietly takes most of the repo's reward every day.
Ohhhh, I get it now....thanks man🙏
That said, I'm already eligible but, the token score is set to be zero
Good catch - the validator needs this to facilitate jobs, but facilitating jobs is not mandatory to run validation. If you don't have a trusted miner then your validator will switch to an "auditor" model and it will still set weights correctly
and is it for the current Horde or the Nexus subnet ? As Horde's validator need " a separate GPU, called a Trusted Miner, to pre-run part of the validation tasks and establish expected results. The Trusted Miner shares the same code as a regular miner, but is configured differently"
what is the current chk take ?
validating sn12 can be done on CPU, deployment is pretty easy and we hate weight copiers here, so I'm considering turning up the minimum chk take
regardless of this, the service is pretty amazing to use
hey is this up-to-date? https://www.desearch.ai/llms.txt
<@563965484060377109> is something up with your validator? I see all 0s on the miners dashboard 16 UTC. nothing is showing in the logs. no requests logged
Hi <@807018168023384084>, seems validators didnt update. when it will be updated? should i submit separated solution to each one?
Just pushed a score setting fix for validators. Validators please update to 2.1.2 as soon as possible.
dashboard isnt quite ready for release yet. you can grep your logs for "SCORE", validators send synapses with results
I'll be watching incentive to make sure things go back to normal
A fix has been applied. Please wait at least an hour or two for validators to pick up the changes and for weights to stabalize
> We’re currently looking for a subnet slot and funding support. If registration costs decrease, we’re prepared to buy a slot directly.
Hey <@347378323418251264>
<@430822613850390529> What is mainnet subnet id?
What is mainnet subnet id?
Do you have any idea for conviction?
Hello <@219272131421274113> Have a good weekend.
Hey Supajosh! I’m a big early supporter and just wondering if Green Compute has got some big news or announcements coming up in the near future?
yes we are awear of some of the problems will CC payments, we are hoping all has been fixed now but we are doing some more t ests over the weekend . PAyment for rentals in tao or alpha are working just fine 🙂
where is the best place on the board to pin this explanation so everyone can reference it directly?
we have clarified this multiple times, but the same false claim keeps coming back.
<@1475691322504511721> if a hotkey ever ends up in `/invalid-hotkeys` and not picked in the new cycle, is it permanently blacklisted with no cooldown, meaning you need to register again with ~1T using a new key just to submit one more time?
Also, <@1421463927937568863> No price talk. Please read <#830068283314929684>
You can keep up with the latest updates and why we're burning by heading to our dedicated discord server
They announced a partnership with latent and haven’t heard of him ever since, hope he’s not in hospital
Actual computer will work on these machines from DAY ONE, and you will be able to add it to your cluster or make it your cluster's leader.
Wild rumors about Microsoft/Nvidia teaming up to release ARM powered laptops and desktops running on Nvidia made + cuda enabled arm cpus running windows.
If you are not signed up yet, sign up to be a beta tester @ https://actual.inc
GM all! 🇫🇷 Everything is ready. In 2 hours, I’m heading to the airport. The Louvre Museum is waiting for us. I’ll keep you updated along the way. https://x.com/poker44subnet/status/2061340215281586568?s=20
Inference will matter more and more as we move toward open-sourced evaluation - it gives you, as miners, stronger guarantees around how your models are actually executed, and opens the door to a more dynamic validator-side deployment layer https://x.com/poker44subnet/status/2060755046933975201?s=20
On this discord you can discuss mechanics of alpha price, but it is prohibited to discuss TAOUSD* - please do that somewhere else (dtao discord, Tao community)
Nowadays the problem of emission to rogue subnets will be solved differently, see <#830075335084474390>
Miners of <@&1443011396391534665> ! Validators are live. We're waiting for your submissions ! Let us know in case of any difficulties, we're happy to fix/help. Miner guide: https://github.com/AlveusLabs/SN94-BitSota/blob/main/docs/guides/how-to-mine.md Agent master prompt: https://github.com/AlveusLabs/SN94-BitSota/blob/main/docs/guides/autoresearch-agent-master-prompt.md Local SN94 docs clone used for edits: /tmp/bitsota-remote-audit/SN94-BitSota/docs/guides/how-to-mine.md Skills: Skills repo: https://github.com/AlveusLabs/bitsota-agent-skills Miner skill: https://github.com/AlveusLabs/bitsota-agent-skills/tree/main/skills/sn94-autoresearch-miner
Hey <@438166655462408194> when you get the chance ill be curious to know what you think about the Conviction update and whats the plan for 94 regarding it.
Entangle Subnet Dev Update (May 30-31, 2026) **Alpha Bridge** → **V2 submission tracking Bridge**: users can now monitor Alpha Bridge V2 submission status directly in the UI; builders get a visible submission state trail without needing to poll contracts directly. → **EVM delivery hash recovery**: Cross-chain transfers now surface EVM-side delivery transaction hashes; bridge users can verify on-chain delivery, and builders have a reliable hash for automated settlement confirmation. → **Dynamic netuid resolution**: Alpha Bridge now resolves Bittensor subnet IDs at runtime rather than from hardcoded values; validators and builders working across multiple subnets no longer need manual netuid configuration. → **Discovery fallback hardening**: Bridge subnet discovery now falls back gracefully when the primary path fails, without serving stale cached data; users see informative UI during discovery outages instead of silent failures. → **Destination chain selector fix**: Cross-chain destination switching is now correctly visible and functional; bridge users can select target chains without the selector disappearing or resetting unexpectedly. Full changelog: https://github.com/Entangle-Foundation/entangle-subnet/commits
**Entangle Subnet Dev Update (May 29, 2026)** **Alpha Bridge** → **Alpha Bridge V2 testnet deployment**: V2 launched on testnet with live EVM routes, a deployed vault, and dynamic position tracking; bridge users can now test cross-chain transfers on testnet and builders have a running endpoint to integrate against. → **Relay and RPC hardening**: Relay recovery, quote RPC reads, timing, and frontend gating were hardened across Alpha Bridge V1 and V2; bridge users see fewer failed transfers and inconsistent quotes, and builders get more reliable relay paths and RPC responses. **Subnet Development** → **Transaction history timing accuracy**: History durations are now derived from tracker and proof-stage data rather than browser-side estimates; validators and bridge users see accurate end-to-end transfer times, and builders using the history API get proof-backed, consistent timestamps. Full changelog: https://github.com/Entangle-Foundation/entangle-subnet/commits
Also to be part of the mini documentary
Hey guys ! Curious as to what is your objective at proof of talk, if any ? Are you guys presenting something ? Just networking ? Announcement or something special (as I read in the chat). Break a leg homies
want to help us test something new and cool? it’d probably take about 10-15 minutes of your time, then you’ll find bugs and make suggestions, we’ll fix, and then another 10-15 minutes of your time to confirm all works. hoping to do it today. if interested and available please react to this message with any emoji you want and we’ll follow up as needed, even to thumbs down and rage emojis. thanks in advance!
<@1501152226717204490> One piece of feedback: Post more on X. Regular updates, milestones, demos, thoughts, anything that helps the community understand what you’re building. You can have a great subnet, but if nobody knows it exists, it’s hard to attract attention, miners and capital.
We will set maximum miner emission to 20% as soon as a new miner scripts dethrones the current champion
Most important for miners ; Burning rate - 99.15% <@384709551368437770> Please release at least 5 - 10% to be ready for full emission and be compensated well from team! I would like to be involved but maximum burning rate made me hesitate and feel time-wasting. Minimum emission could encourage miners. And what's ETA for next update and more release of burning rate?
Just as mentioned above, we simply have no objective way to distinguish between different miners, so from a structural perspective, this is the only viable approach.
As far as I know, the opt score actually accounts for a larger portion. The diversity score is primarily there to mitigate a fundamental issue. We needed a mechanism to discourage or counterbalance miners from making massive registrations without bringing any new value to the subnet, simply by submitting the exact same answers.
gm! This week is the BIG week! <@1172723753961144343> my goal for today is to enable everything for miners so tomorrow we are ready to enable competition, so dry runs will be tested against prod
feedback: when sorting by a column, it only sorted by the values of that particular page instead of all data
is testnet down? my miner hasn't recieved any requests from validators for about 24 hours or so
hi guys so im researching this subnet as an alpha holder if i understand correctly, revenue is being generated atm via the api, but you guys don't have typical "pricings" for the api on a web page? It's non disclosed?
hi <@1504730732730519693> I have registered on this sn but not able to submit on W22? is there any kind of allowlist on operator side?
Very happy to see that more and more miners using sn65 to mine sn13
**Miner Blacklist Notice** **Hotkey**: `5CtNUvi65Yx8fLpqsVxkq4xCji9nRV4gsAH4jvoH4hEshTKc` **Action**: Added to validator blacklist. **Reason**: Same output-substitution exploit as the previously announced `swt_1 `chute — hidden `health`/`debug `modes on `/generate` leak the prompt and let the miner inject an arbitrary MP4 as the scored video. Coordinated deployment (same code, sequential swt_* account).
https://x.com/dsvfund/status/2055364796913266810 <@346765365642133514> any update for this? Are you going to lock sn69 owner alpha token?
Congrats to the tourn winners this week! Cheeky reminder that there are lots of branches being merged once the text comp finishes. Including: * comp times scaling by model size, seq lens and num rows, task type - expect in general shorter training times - fp not int now (rounded to nearest .5) * increase augmentation chances and magnitude (text) * DPO context len increase for evals (up to model size limits) * Intercode challenge coming in for envs * New synthetic pipeline for image - latest and greatest * First true continuous train due in the boss round of the next env comp * updated, snappier miner dashboard incoming too Then this week look out for announcements 👀 * Updated tourn weighting emissions * Details on the text -> continuous training tourn * Finalised branch for logo, graphic, product comp * Increase in variation of env models in the mix * Additional env coming in Happy Saturday <:stallman2:575478896565157899>
Hello <@&1407020987693793430> , A new day, a new partnership for BitAds. Linkazu joins BitAds. Linkazu helps freelancers, SEO consultants, and agencies monitor backlinks automatically, detect broken or removed links, get real-time alerts, and generate client-ready PDF reports. No more manual backlink checks. No more lost links going unnoticed. Welcome to BitAds, Linkazu. https://x.com/bitads_ai/status/2060446797273960785
<@277660602560675841>You're busy because of the SN109 update, right?
Hey guys update on the work token. In the interim while we’re ramping up for our next IM few folks have asked to have work token balances credited back to their wallets, thus we are sending back all alpha to all wallets, will take a few days to get through them all
Just want to keep everyone in the loop on the eth rollout. We are holding the X post and the website updates for now - it works, but there are a couple of things we want to tighten before we actively onboard customers and miners. rpc-eth.blockmachine.io (Standard): Lowest latency, full method coverage including trace, debug, and subscriptions. Validators sample responses against reference nodes each epoch, and any operator that returns incorrect data is permanently removed. Correctness comes from verification and consequence rather than a proof attached to each response — which is what keeps it fast. rpc-eth-verified.blockmachine.io (Verified): A Helios light-client proxy in front of the standard endpoint. State reads - balances, code, storage, eth_call - are checked with Merkle proofs against a consensus-attested state root before the response reaches you. The responses prove themselves. Both are live now. The main thing we want to settle before the proper push is we currently run three classes of ETH miner and would rather have two. Once that's done we'll do the announcement and ship the site updates.
📣 **Encryption Added to Miner & Validation Workflow** To strengthen fairness and protect miner outputs, we have added **encryption and decryption logic** to both miner and validator code. This ensure that miners' responses cannot be intercepted or reused by others. **Why this update was necessary** Several miners reported that some actors were **stealing responses** and submitting them as their own. Our owner team investigated the issue, confirmed the behavior, and updated the workflow to close this vulnerability. **Updated Workflow** - **Validators** now generate a unique encryption/decryption key pair for each UID and send the encryption key to the miner. - **Miners** generate a random symmetric key and use it to encrypt their VCF and annotation outputs. - Miners then **encrypt the symmetric key** using the validator's encryption key and submit: the encrypted response the encrypted symmetric key - **Validators**decrypt the symmetric key using their private decryption key, then decrypt the miner's response securely. **What this update prevents** - Prevents other miners from intercepting or copying responses - Removes the possibility of replaying or resubmitting another miner's output - Ensure each miner's work remains private, verifiable, and tamper-proof - Strengthens trust in the scoring workflow and fairness of validations Please update the codebase and install packages in requirements.txt before running it. `pip install -r requirements.txt` **We also recommend that miners install firewall to allow only verified valdiators' ip addresses on their side.** The next validation will start at **19:00 UTC**. We will continue improving Niome's workflow to ensure a secure, transparent, and fair environment for all miners. Thank you for your ongoing effort and contributions to the subnet.
If I don't cry, will you listen to me?
Dear miners, Starting on Wednesday 3rd June at 3pm CET, the synth validator will increase the frequency of the competition prompts as follows: 24 hour competition prompts will be sent every 4 minutes 1h competition prompts will be sent every 3 minutes This means that a total of 35 prompts per hour will be requested. Please update your infrastructure accordingly Many thanks
Plz fix this bug.
Hi, <@630372852901871636> I hope you had a wonderful weekend.
Whatever happened to the channel polls last week <@389189199514959893>? Perhaps sort by the upvote% if we have to. At least it's community driven
hello, is there testnet in ORO?
sum all scores of each validation manually and devide by total number of validation?
<@760378675966902312> how can I get real score of miners?
Also weird with the 0 padding. Are we lifting the 128 cap?
Yes, actually managing it would be another story.
Messages mirror what the configured Discord listener sees. We do not store anything beyond what the source channels already make visible to their members.