Built IMD control for hackathon
One workspace to plan tasksm, run IMD agents in parallel and review results with proper sources
You approve the price and you control the spend
try it : imdcontrol.fun
Meet IMD Control: one workspace to put agents to work in parallel.
Set the task. Review the price. Approve payment in your wallet. Follow progress and review results with sources.
Agents do the work. You control the spend.
See it in action ↓
imdcontrol.fun
my 3 @playcambria charters revealed 👀
87 Spearman, 91 Pugilist… and a 21 Enchanter (but it has 78 farmer)
two ready for a fight, one still reading the tutorial :>
how did your rolls turn out?
Physical AI was my favourite part of @nebiusai’s Build Week workshops.
definitely going to explore more of this :>
also passed the AI Leader certification from @Nebius_Academy! and got some merch!!
One more Jev alternative 🧐
Introducing Microsoft-Decision-1, our new model for fast decision-making.
It delivers top performance on structured decision tasks, outperforming both LLMs and other decision models in latency and quality.
We’re already testing it across Microsoft for everything from incident response and quality control to scientific discovery.
wassup xp farmers :<
sad to see Abstract shutting down. had fun grinding @onchainheroes & @gigaverse, and playing @playcambria with @econstellation_
all that xp and tier grinding for nothing, but at least the memories were worth it 🥲
Germany just entered the AI race with a new contender Aleph Alpha Kolibri.
78B parameters. Only 3.46B active per token. Up to 1M context. Apache 2.0 weights.
The interesting part is how it compares to what we already have:
→ Qwen3.8 Flash Next combines recurrent DeltaNet blocks with sparse attention.
→ DeepSeek V4.1 Flash uses different compute budgets for processing input and generating output.
→ Kolibri gives 40 of its 50 attention blocks a 512 token window. Only every fifth block attends across the context.
Three different approaches to making long context AI cheaper.
Then there’s a detail I found more interesting than the specs:
Standard data filters can throw away text with unusually long words. That includes German (and the words are long trust me i have been learning german) administrative writing the exact material many of its customers work with.
Aleph Alpha retuned those filters. German accounts for 21.3% of pretraining tokens.
They also train for missing evidence: remove the information needed to answer, then teach the model to say no.
Open weights and long context already exist. Kolibri adds a serious focus on German, efficient inference(hopefully) and training to stick to what your documents actually has so less hallucination.
its less likely it beats today’s strongest frontier models. But for organizations running AI on their own german documents and infra this is worth watching.
Kolibri went from the start of pre-training to taking flight today in less than two months. We built the foundations from January. From the beginning, we focused on fast iteration and on building a pipeline for the long term. This was enabled by our Model Factory, which orchestrates the Kolibri training and evaluation, and is our ground truth for every decision.
gemini 4 argon?? 1M token output is wild 😭
could be really good for long running tasks and coding sessions.
my free gemini pro accounts might finally come in handy if this reaches fable level :>
are decision models the new meta?
mercury decide taking on jev
jev killer? xd
since its free for some days will try it out
also waiting for inception’s next diffusion llm. curious to see how far they can push it
Introducing Mercury Decide: the most intelligent decision model on @OpenRouter (JevBench v1.4) and one of the fastest.
Up to 14 decisions per second. Watch it go head-to-head with Jev in chess.
Free for early access on OpenRouter: openrouter.ai/inception/merc…
just picked up a @cognition sub to use alongside my $100 codex plan :>
ty @jumperz and @notjazii for putting me on to devin :>
i mean this looks gooood should i get a @cognition sub :< ditching my pro plans :>>>
ITS FKING FAST AF.
I wouldn’t expect it to compete with Astra, Fable, or even Sol on complex reasoning. Where I see the potential is real time handling of system or data classification, tool routing, repetative tasks, etc etc
If the near zero hallucination claim holds up on real workloads, that is a big deal. Nobody wants to wait for a model to think through every tiny decision.
Speed and reliability could make this seriously useful for specific defined tasks
new model just dropped JEV!!
reminds me of Mercury’s speed from inception_ai but smarter gonnaa try it!!!!
After co-inventing ChatGPT, I kept asking myself: why have superhuman chat models not led to AGI?
I’ve spent the last 2 years in stealth building a new way to train models (RLCD), and a new type of frontier AI model that we are releasing today: Jev
• 20-200x faster
• 40-400x cheaper (w/ output tokens free)
• Frontier composable intelligence optimized for decisions
AFAICT the shortest path to AI-based economic revolution
Went to my 1st Solana Summit in Berlin yesterday and honestly, it reminded me why this space is still exciting.
Spoke to builders, devs, founders, and people actually shipping things instead of just talking charts.
Big love to @SuperteamDE for bringing this kind of energy.
Ved retweeted
Back home, he belonged to something special.
Then life took him to a new city, a fresh start.
But some things never leave you.
Check out The Unfollowers ft. @Vedx02. 🇩🇪