Sultan Suleimanov

Research analyst, @deai_nation.

Shitpost in Russian: @sult

[Sultan Suleimanov @sult_ee]

Just in time for Directionally Bad, the Fable drama happened. I wrote a little about what happened and why Dario should feel happy now

https://t.co/Q37GEJCbDy

[DeAI Nation @DeAI_nation]

https://t.co/ZXieA586IO

[DeAI Nation @DeAI_nation]

The new world order

https://t.co/GFrCcBn0yf

[Sultan Suleimanov @sult_ee]

welcome to era of tokenaxxxing

[Sultan Suleimanov @sult_ee]

@tunguz none of these. My options: @NousResearch @PrimeIntellect @MacrocosmosAI

[Sultan Suleimanov @sult_ee]

@feathershredder respect the game, but maybe let's leave the cheating for the centralized labs

[frankie @FrankieIsLost]

the year is 2028. claude infers whether you’ve ever even thought about gradient descent and silently routes your queries to Claude Sisyphus, a model RL’d to maximize engagement while avoiding task completion. you spend your entire UBI token allotment on it without ever realizing.

[Sultan Suleimanov @sult_ee]

babe I'm not dumb. I'm just quietly limiting my effectiveness

[Sultan Suleimanov @sult_ee]

Siri AI? Nah man, I prefer my Siri-30B-A4-qwopus-distill-instruct-llama-capybara-MTT

[Sultan Suleimanov @sult_ee]

вайбы постов медиа в твиттере где твит, заголовок и картинка на превью повторяются слово в слово

[Brooks Otterlake @i_zzzzzz]

Insane that this is the main image Apple is using to show off the new AI Siri

[Sultan Suleimanov @sult_ee]

Friendly reminder:

https://t.co/kDwjFFzlXE

[Sultan Suleimanov @sult_ee]

@x1knet @akashnet cool! we also made an estimate for a node of 8 H100's a while ago https://t.co/uY1DcBAGgl

it was $177/day at the moment, and more like $280/day as of today's rent prices

[Sultan Suleimanov @sult_ee]

state of the market: OpenAI loses talents even before acquiring them https://t.co/LYYf9QVWJK

[Tim is GOING TO VIBECAMP ⛺️ @MasterTimBlais]

Personal update: I've decided to leave OpenAI

as soon as i figure out how to join them

[Sultan Suleimanov @sult_ee]

it's time to get your ass beat for every stone you cast away

[DeAI Nation @DeAI_nation]

Directionally Bad Issue #004 is live and the “tokenmaxxing” era is ending.

AI companies spent the last year pushing everyone into agentic workflows. Every developer was nudged into vibe coding, and every employee was quietly trained to burn tokens like they were free.

Turns out, they were not free!
GitHub Copilot is moving to credit-based pricing. Frontier model access is getting more expensive. The magical “AI productivity layer” now comes with hard limits, usage caps, and spreadsheets.

At the same time, OpenAI is getting closer to the US government, with discussions around public ownership, wealth funds, and state participation.

Decentralized AI would be the ideal resolution here. Otherwise, we are at risk of AI access being negotiated between a few labs, governments, and hyperscalers.

Read the full issue here - https://t.co/Z8bNrpwI9i

[Sultan Suleimanov @sult_ee]

https://t.co/3cYwV91lJf

https://t.co/3cYwV91lJf

https://t.co/3cYwV91lJf

[DeAI Nation @DeAI_nation]

Introducing DeAI Dashboard!

When we've launched DeAI Nation in late 2025, we already knew that Decentralized AI will be one of the most fascinating discoveries for a wider audience in 2026. Dozens of talented teams and hundreds of thousands of participants around the world are trying to bring online resources for open, accessible, and censorship-resistant AI infrastructure, and millions of users are there to start using it.

With centralized labs becoming more and more greedy, the reasons to care about decentralized AI have only become stronger. Since our launch, we've released our State of DeAI 2026 report (in English and in Spanish), analyzed how much a single node could make on different DeAI networks, and launched a silly newsletter, Directionally Bad.

Today, we want to demonstrate our next tool: DeAI Dashboard (the name sucks, give us some ideas!). This is a very early alpha version, but we'd like to share with our followers first to get an honest feedback.

For this dashboard, we've heavily focused on four points: available compute, inference capacity, community activity, and market performance. For compute, we are calculating every project's (where available) inventory in Nvidia B200 GPUs equivalents. For inference, we're calculating the theoretical throughput for Qwen3.6-35B (other models coming soon). And for the community score, we're trying to find balance between Github activity, Discord members, and X engagement.

https://t.co/rhhgTB1Qxt

[Sultan Suleimanov @sult_ee]

Third issue of Directionally Bad newsletter is out!

From the pro-decentralization Pope and @opentensor's Conviction to TEE discussions related to @say_gm_ , it was a fun week to cover

https://t.co/HI42YLY6Ag

[Sultan Suleimanov @sult_ee]

something we've been working lately, a very early look for those following this account 👀

https://t.co/YfM4omcNnl

[Sultan Suleimanov @sult_ee]

New newsletter is out 👀
https://t.co/dLA9qC3F4s

[ᐱ ᑎ ᑐ ᒋ ᕮ ᒍ @Andr3jH]

Karpathy going from OpenAI to Tesla to Anthropic

[Sultan Suleimanov @sult_ee]

@ClementDelangue @karpathy @AnthropicAI you sure?
https://t.co/t4bXfn8jHm

[1a3orn @1a3orn]

I wish Anthropic would stop opposing open weight models.

Open weight models have been great for x-risk research; for decreasing concentration of power; and Chinese papers have been fantastic for keeping the non-lab general public more informed of how AI works.