Sammy Ace | Meraki

Rebrand! Finally swept an @Azuki off the floor, and I couldn't be more excited! 🔥 It's been a long time coming. I've watched this community build and grow for years, and I'm thrilled to be a part of it now. 🫡 The @Azuki community is amazing for connecting with incredible people. Looking forward to meeting more of you🫶🏽

Why aren’t you accessing the 3.65% Sky Savings Rate yet? Swap into sUSDS, and your balance starts accruing yield automatically. Start here ↓

MsLabs just went through a full reset. Cut the server from 8k members down to 1k, kept only the people who actually show up. Cleared out dead channels, every channel left has a real purpose now. ACO services are now live inside the server. Everything we’re putting in place is to make sure this community stays tight, people show up, share information, get better individually and together. Bringing in people who will actually contribute and help the community grow. if you’re active and genuinely want to contribute to something growing, you’re welcome into @The_MSlabs

Your PFP can get you GTD🦄 Turn your PFP into a UniPix card Best ones win GTD spots Full setup + GPT prompt below 👇 unipix.pro

@MasterChiefNFTs @streamex @solana The re emergence of RWA Seems to be one interesting concept

most people are still arguing about if tokenized assets will take off $STEX already made it happen: - 3,096 oz of gold backing $GLDY, tracking price 1:1 - yield already being paid out to holders - $72b in retirement capital (Equity Trust) can now hold it inside an IRA - wintermute on board - gold that moves and trades 24/7 - SEC reportedly about to open the door for tokenized stocks same foundation gets used for silver, copper, oil next every new product is cheaper and faster to bring to market GLDC drops in Q3 - built for everyday buyers, yield bearing, gold exposure straight from your wallet zero debt. $45m cash. Key partnerships not even fully active yet quiet for now but not for long iykyk

Stablecoins aren't built to sit still. Yet, $323B+ sits idle today. Earning nothing. That's $11B in lost yield every single year. The Sky Savings Rate fixes this. Earn 3.65% on your stablecoins, just swap into sUSDS, and you're done.

@andres_solt @edeldotfinance @saylor @Strategy @xStocksFi That’s one insane yield boost holy moly Jumpin in rn

$STRC became the hottest yield source for DeFi in 2026! On @edeldotfinance you are able to boost the yield from 11.5% to over 42%! Watch this video to understand how it works, the risks associated and how to set it up for yourself! thanks @saylor @Strategy @xStocksFi

And for the people asking which benchmarks: MATH 56.2, DROP 82.2, ARC-C 81.9. that’s a 1B base model beating 7B open models on reasoning. MMLU 60.7 is the one where scale still wins, and they don’t hide it. the honesty is part of why this one lands different.

If your AI infra business depends on the idea that scale alone is the moat, the HRM Text benchmarks from @Sapient_Int should probably worry you. Ponder that. 1B model. Around $1k in pretraining cost. Independently verified. Still matching or outperforming 7B models across multiple reasoning benchmarks. But the real signal is the FLOPS efficiency chart. HRM is sitting completely on its own in the upper left corner. Around 44,000x more efficient than GPT 3.5 on the same reasoning tasks. For years the industry treated scaling like a law of nature. Turns out it may have just been the most expensive habit in AI.

FYI this is still a base model, makes you wonder how intense this keeps on going

@0xValmir They really making insane waves

The HRM Text benchmark video dropped and I cannot stop looking at these numbers A 1B model with zero post training and a $1k pretrain budget Still beating models 7x its size on reasoning and independently verified They even kept the one benchmark they did not win in the video @Sapient_Int is built different

@evokein @Sapient_Int Kinda makes you wonder how crazy this goes out to be being the base rn

the FLOPS chart from the HRM-Text benchmark video is still kinda ridiculous to look at hours later a 1B model from @Sapient_Int sitting near the top while using a fraction of the compute compared to everything around it doesn’t even feel real at first glance 56.2 on MATH 82.2 on DROP 81.9 on ARC-C while training on ~40B tokens and around 1/350th the FLOPS of larger models and the funny part is they didn’t even try to hide the one category where bigger models still win most teams would’ve buried that slide instantly

@josephweb3 @Sapient_Int It’s amazing ngl

been saying most people’s definition of “using AI” is just chatting with the same tools everyone else is using. meanwhile @Sapient_Int just dropped a 1B model that beats 7B models on reasoning, trained for $1k, independently verified. and this is still the base model. no post-training yet. are you paying attention? 👀

@Adefola_ Tis just the beginning

The quiet detail in the HRM-Text benchmark video that researchers should be paying attention to: this is the base model, no instruction tuning, no RLHF, no DPO. 56.2 on MATH at base is the starting point, not the finish line. Post-training hasn’t touched these numbers yet. @Sapient_Int still has boundaries to break

@SimonNickweb3 @Sapient_Int Mindblowing in all scenarios

As I mentioned some hours ago about @Sapient_Int, HRM-Text is genuinely surprising me with how strong it's performing. A 1B-parameter model running on very modest compute has no business staying this competitive against much bigger models. What I respect even more is that they didn’t hide its weaknesses but instead openly showed where it falls short. That level of transparency actually makes their benchmark video far more believable.

@JosephDevoy Damn, my work is gonna be a lot more easier How do I get started xD

Introducing Tempo: The world’s first AI Head of Growth. AI can now scale a brand faster than any human, see how:

17.1B SKY staked. 72.8% of the entire circulating supply. $1.2B in TVL. These numbers tell a story. So why would you stake your SKY? ↓

@king_tegg @Sapient_Int A beautiful more efficient approach

Most AI models think by writing every step out in words. HRM-Text from @Sapient_Int does it differently. It thinks internally first, then gives the answer after. Crazy part is this is only a 1B model, yet it’s competing with much bigger models on reasoning. Feels like a smarter and cheaper way to build AI.

@Mr__Kovacs @Sapient_Int Quite outstanding

GPT-3.5 scored 43.1% on MATH. HRM-Text 1B scored 56.2% on MATH. GPT-3.5 used an estimated 44,000x more computing power to train. HRM-Text still scored higher. I am amazed to see that @Sapient_Int built something smaller, faster, cheaper that outperforms models 10x its size.

The wild part about HRM Text is not even the size. It is the fact that it reasons before it speaks. Most language models are basically advanced autocomplete. They predict one token after another, stacking guesses in sequence. If you want them to “think,” you usually have to make them explain every step out loud through chain of thought. @Sapient_Int HRM flips that around. The reasoning happens internally in latent space, with the high level planner working alongside the low level executor. By the time the words appear, the thinking is already done. It feels closer to how people actually solve problems. You think internally first, then explain it after. That approach is cleaner, faster, and far less fragile. Quite interesting if you ask me