namoray
@namoray_dev
Followers
2K
Following
528
Media
14
Statuses
198
The only universal truth is the optimisation of rewards
Joined December 2023
The era of dedicated libraries is upon us. Meet Fiber - the protocol purpose built for Bittensor subnets
14
40
183
The evolution of miners over the last two months on Bittensorโs SN19 - for a 60GB model. It wonโt be long before we top every benchmark. The future is incentivised decentralised optimisation
5
37
133
So, Bittensor?.
For friends of open source: imo the highest leverage thing you can do is help construct a high diversity of RL environments that help elicit LLM cognitive strategies. To build a gym of sorts. This is a highly parallelizable task, which favors a large community of collaborators.
3
10
126
I am very confident gradients (subnet 56, Bittensor) is the best auto-ml platform in the world. We are beating all competitors in internal benchmarking.
3
14
113
I think Bittensor just set the gold standard on dealing with hacks/exploits. Name me another project that handled a hack better.
4
10
91
What's that new subnet 56 about on Bittensor called gradients allowing anyone to train models on Bittensor in a completely decentralised way built by rayon labs with a sexy ui and about to change how people fundamentally think about decentralised ai?. Anyone know?.
5
15
91
๐ New product release - the first step to multimodal chat!. Competition time!! ๐. 0.5 to the best image & image-image edit. Submitting:.Submit your images to the `competition` channel on Corcel or here: #S19VISION
7
23
79
Chutes registers subnet 64. is live for anyone in the world to train ai on Bittensor. Nineteen continues to top tokens/s benchmarks. Santa knows how to deliver.
2
19
84
Gradients launches and Tao shoots up to near $700?. Coincidence? Probably tbh.
4
8
77
Already hundreds of gpus on chutes, it doesn't take long on Bittensor.
2
12
71
SN39 Edge Maxxing have officially migrated from the Bittensor package to Fiber The new age is upon us. Love it Wombros @WOMBO.
6
4
68
Dendrites and synapses will die soon. They served us well, but if we want to compete with the centralised world, we have to shake our tech debt. I'll be releasing my recommendation soon, using bi-directional symmetric encryption instead. After a small handshake, it solves *a lot*.
9
8
65
Chutes ticks every box:. - Run any AI model, giving value to all training subnets .- No UID pressure .- Room for miner optimisations, not only providing resources.- Extremely strong business model.- Definition of decentralised AI.- Beautiful frontend for all to use.๐ช.
3
11
64
Chutes is brewing the next logical phase for Bittensor subnets. Validators are drifting from their original goal, and are overwhelmed with far too much work. We believe that we should give subnet developers more power, and have validators act as auditors. Watch this space.
4
12
61
405B? Lower barrier to entry? Validators customising models? About to outcompete centralised services? โ
2
12
56
Also note on 64 there are 0 weight copiers running! It doesn't exist if subnets run the audit-based model :).
1
5
60
All of this happened with misty glasses, the tech was still being discovered. 2025 will see the first unicorns emerge from Bittensor, and will become the home for large scale decentralised training runs. You just need to make incentives clear.
Unbelievably stoked to scale fully in 2025. Here are some of the things we achieved in 2024 with less than 64 subnets . ๐.
1
12
59
Rayon labs is always about pushing Bittensor to be the best it can be. Currently we're focused on questioning our mechanisms for running subnets, and proving that with the right design, subnets can bring real value. Just try for instance, and train an ai.
3
7
58
No room for complacency for the SN19 miners -> Check out how we will soon overtake centralised providers @
3
7
57
Cursor support (AI enabled IDE)? โ
.New models added with 30 lines of code? โ
. nineteen, 5.0, Bittensor.
3
6
49
And what better way to access it than #bittensor?.
Open source AI is the way forward and today we're sharing a snapshot of how that's going with the adoption and use of Llama models. Read the full update here โก๏ธ ๐ฆ A few highlights.โข Llama is approaching 350M downloads on @HuggingFace. More than 10x
0
9
46
Thousands upon thousands of models now trained through the miners with . No other place can you have the worlds smartest miners optimise your training for you in just a few clicks.
1
10
45
3rd times a charm, the best podcast yet I reckon.
Bittensor Guru S2E6 - Subnet 64 w/ @namoray_dev and @jon_durbin is live! Papa's got a brand new bag and it's an absolute unit. Come dtao this may eat 1/4 of the network. It's not fair to let these two work together, parity just jumped out the window. $TAO
1
3
42
Subnet 1 validation now happening through the fast, stable, serverless inference of subnet 19. Use 19 with your tao @
0
10
43
Do you ever realise that optimising incentives is exactly what life is?.
5
2
41
Very glad to be supported by Crucible. Gradients makes us ask a big question. What if by being consumer-led we accidentally build the best auto-ml tech in the whole world?. I'm excited to see the answer. We're already descending toward it.
We're adding three new subnets to our portfolio today!. โข Neural Condense ( - Token compression to reduce LLM inference costs.โข Gradients ( - On-demand model fine-tuning services.โข Empathetic Speech ( -
1
4
41
I'm not sure we should describe Bittensor as an experiment. Yes, it's early phase and has incredible potential. Yes, it *requires experimentation* and *requires iteration*. But there is no 'outcome'. There is no end we'll get to and call it quits. There's only more to do.
3
3
36
A pleasure as always - we need more like Keith.
Bittensor Guru Podcast Episode 37 - Subnet 19 Vision w/ @namoray_dev video is live. The lead of the top subnet on our network spends an hour talking and Bittensor's future. $TAO
1
2
33
Decentralised code editors next?.
Excited to share my latest work: pAIr, a VSCode Copilot alternative powered by the @Corcel_X API! Still very early, but I wanted to show it anyway. Looking for alpha testers to try out the extension soon. Anyone interested? #VSCode #DevTools #AlphaTesters #Copilot #bittensor $TAO
1
4
31
@badenglishtea 1st option sounds nice but will be hard to get there with a consensus.2nd relies on the assumption that the bottom subnets are great and we will miss their output - historically not true. I think the community should give recycling weights a go.
2
0
25
Incentive mechanisms to run companies? Devs are miners? C-suite are subnet owners?.
3
2
20
@badenglishtea I think we should all be able to openly talk about the pros and the cons of bittensor and how we can improve, no issues there. Just the post yesterday was simply not true :D. We live and learn ๐ซก.
2
0
17
Synthetic data set for Midjourney underway โ
. Millions of v6 MJ images from Bittensor - open for all to use . Next stop: new generation of image models๐คค
1
5
19
The best finetuned LLM's from Bittensor - how do I access them?. S19, all validators to give api access - that's how.
1
2
18
Can't confirm or deny.
predictions for month 2.5:. high probability:.low conviction teams exiting.high conviction teams continuing to ship (it's a marathon, not a sprint).I continue to age at triple my usual rate. low probability: .@namoray_dev is secretly pamela anderson in disguise.
3
1
17
We warned you.
Teaser: This is NOT a centralised API wrapper. This in open-source models of the highest quality running on subnet 19, available through front end apps and API access for developers to build with. HUGE updates coming next week. Be ready.
0
4
16
DutchTensor is upon us ๐ฒ.
๐ Big announcement from (see video below). We're stepping up our game alongside our validator activities to address the gap between the Bittensor subnet solutions and real-world adoption. As a commercial bridge, and with our experienced sales team
0
1
15
@getmasafi @BrendanPlayford . - SN59 โ
.- Agents โ
.- Uses โ
.- Uses Fiber โ
. We enter many new eras.
1
1
15
@badenglishtea Wallet 1:.Wallet 2: These combined have 20k Tao - I'm not sure where you're getting '100% is dumped to exchanges' from @badenglishtea.
2
1
13
@ImBananas4 Ah no I'm referring to the communication mechanism between Miners and Validators themselves.
0
0
12
@CryptoZPunisher Validators function as auditors instead of the traditional 'validators'. Writing an opinion piece on this to share.
3
0
10
@E_Z_Trades @mogmachine @rayon_labs Exactly that you've got it. Also makes it much more lightweight and easy to query the metagraph, set weigthts, stuff like that.
0
0
10
@markjeffrey @FacetSir @hcaulfield8148 @altcoinchadd @WayneVaughan Try . This is just as private as Venice and easy to use - plus faster :).
0
0
9
@badenglishtea Very wrong - look at the two wallets the tao gets transferred to. Nearly all the tao is sat there.
1
0
8
@ImBananas4 @WOMBO Our subnets no longer rely on OTF to maintain the bittensor package. We have our own, with contributions form the community building the package that powers our subnets.
0
0
9
@0xarrash Ship early and often tells us that I think doing loops of step 1-5 very often is a good idea, I think we're on our 5th currently :D & I know SN8 has done plenty of laps.
2
0
7
@Ch3RN0b0G Absolutely, itโs very easy to verify and anyone can do it - Iโll post the full methodology too!.
1
0
6
@ai_bond_connery @Corcel_X @taostats @mogmachine Porting; already done ๐๐. Not a single closed source model left.
0
0
6
@KeithSingery Lol I might be too British; watch this space, meaning โstay tuned, watch what we doโ :D.
1
0
4
@ai_bond_connery @Corcel_X @taostats @mogmachine Porting; already done ๐๐ท Not a single closed source model left. 100% decentralised, not a single closed source or subnet run API. We *are* decentralised AI ๐ซก.
0
0
4
@lihotaotao @rayon_labs @OpenRouterAI 19 is rapid, currently not serving the huge r1. It will be servings loads more soon though!.
0
0
3
@CryptoZPunisher Validators become guardians of quality, exactly. Keep subnet owners honest, but otherwise, let them build and get out of their way!.
0
0
3
@therealbigjonno @DrocksAlex2 @Corcel_X Hell yeah. Scout (combines web), character (roleplay), duet (ai's talk to eachother), image studios, avatar, homegrown models (Mobius, Openvision). All of that is just the start though - showing web3 can easily compete with web2. What's next is better than web2. Watch.
1
1
3
@XVentures_Fund @TheBittensorHub II think within this year, we will be extremely competitive to companies like or etc - except decentralised.
0
0
3
@oneandahafcats Something which should probably be improved upon as time goes on - decentralisation is important. But in this case, we're all glad we had that lever to pull.
0
0
2
1
0
1