Trending topics
#
Bonk Eco continues to show strength amid $USELESS rally
#
Pump.fun to raise $1B token sale, traders speculating on airdrop
#
Boop.Fun leading the way with a new launchpad on Solana.
korini
Case of a smaller brain becoming smarter: Results from @SentientAGI's experiment
The notion that "the bigger the model, the smarter it is" is being challenged by Sentient.
In the LiveCodeBenchPro benchmark, the Dobby-CP model, with parameters ranging from 7 to 14B and only 20% of the training data (which can be considered a small amount), showed the same pass rate on medium difficulty problems as GPT-level large models.
This proved that scores obtained from easy tests, as before, do not guarantee actual problem-solving ability.
A key technology from Sentient that contributed significantly to this performance is the ROMA framework.
The ROMA framework breaks down complex tasks into smaller steps for parallel processing.
It achieved higher accuracy than GPT while reducing inference costs by 50-80%.
These results numerically demonstrate that, instead of the sheer size of the model, the quality of the data and the model's understanding and learning ability determine intelligence.
---
The Sentient research team implemented OML 1.0, embedding 25,000 fingerprints into the model without performance degradation.
The community is also reacting with comments that AI becomes dumber as it grows larger.
This can be compared to the efficiency of the human brain performing high-dimensional reasoning with just 20W of power.
In conclusion, Sentient can be said to showcase a monstrous AI that appears highly intelligent with greater efficiency, despite being a smaller model.
Oct 30, 2025
[Latest News Summary from @SentientAGI]
1. October Proves Why Sentient is Fundamental
▶ Awarded 'AI Startup of the Year' at India's largest AI conference, 'Cypher' Minsky Awards
▶ Four papers accepted for NeurIPS 2025
▶ Open-sourced the meta-agent framework 'ROMA'
- Recorded top trends on GitHub after release (1st place for 2 consecutive weeks)
▶ Achieved 4 B2B deals in just one week (contract signed with APAC banks)
In particular, the provision of the 'Open AGI Network' (GRID) and the hierarchical, recursive multi-agent ROMA as open-source can be seen as the exact opposite of other closed commercial models.
2. Recap of Recent Overseas AMA (10.27)
Q1. Why did you start Sentient?
👉 Because AI power is becoming centralized.
👉 There was a need for an open platform where everyone can own AI and protect themselves.
Q2. What lessons did you learn from Polygon?
👉 Real value comes from someone paying for it.
👉 In other words, we need to focus on value that generates actual revenue.
Q3. What are your thoughts on Ethereum?
👉 Ethereum should be the security and settlement layer, while L2 should be the execution layer.
Q4. Is Sentient a crypto AI project?
👉 No. Sentient is a real AI company.
👉 It simply distributes ownership and research incentives through tokens.
Q5. What are Loyalty & ROMA?
👉 It starts from the philosophy that 'AI should be loyal to its owner.'
👉 ROMA is a multi-agent reasoning framework that helps AI think more deeply.
Q6. What is the progress and TGE?
👉 Progress is going very smoothly. Sentient will have a blockbuster TGE.
Q7. What about tokenomics and long-term value?
👉 Staking artifacts in Sentient GRID will yield rewards.
👉 A portion of enterprise revenue will be reinvested into the token economy.
Q8. Collaboration and ecosystem expansion?
👉 We are collaborating with multiple global companies, and an announcement of a partnership with the Arxiv-linked research platform AlphaKive will be made soon.
Q9. Community onboarding?
👉 The goal is to help everyone, regardless of Web2 or Web3, to utilize AI.
801
[@SentientAGI Meetup Review❤️]
I'm an introvert, so I couldn't greet many people, but it was really nice to meet those I had only seen online.
The food looked really delicious, but I was a bit late and missed out, which was disappointing..
So I ended up having budae jjigae afterward, hehe.
—
As I started networking a bit, Yura explained the vision of GRID and Sentient.
I listened intently, feeling like I was reviewing the material.
It stuck in my head much better than when I read it!
—
After the presentation, while networking, I noticed a long line that seemed to go on forever..?
It turned out to be a line to greet @namyura_.
If I had known, I should have greeted them in advance..!!
The person next to me, @jundeu00, said they looked like the Pied Piper, lol.
—
Honestly, I was surprised by how warm @riotboy2024 looked ㄷㄷㄷ He was really fun in person too.
I also met my group project team members @GMB_Coinangel @chuchuprotocol @yeonwoohoIic @waffledaizyo @pc970930, and the sense of closeness was max ☺️.
@leegs0020 @chochunja77 @baggyudong78463, Sentient fam🫶.
I hadn't seen @beingRich2000 in a while, and he came looking stylish with a cool hairstyle and outfit, totally impressive ㄷㄷ. He’s good at yapping and also handsome 😡.
Anyway, it was great to meet everyone, including the Goguryeo ham, Namnin ham, Kva ham, and all the others!!
I couldn't greet everyone, but let's definitely say hi next time…🫶
Oct 24, 2025
[@SentientAGI ]
How is everyone spending their evening?
Here's a brief recap of the meetup 🫡
I don't usually attend meetups, so I just quietly stayed in a corner ㅋㅋ
I had only read about the Sentient project through documents, but hearing it directly from the team members gave me a different understanding.
Project writings often contain a lot of technical details, and I try my best to simplify them while reading through many of them, but it was clear that the experts are on another level ㅋㅋ
I learned a lot about the concepts of open source and GRID.
I plan to write in detail about GRID in my next post.
@SentientAGI , @namyura_ thank you for your hard work!
1.08K
[Did a person write this? AI can now perform handwriting analysis..!! @SentientAGI]
LLM fingerprinting is, simply put, like a 'digital fingerprint' left by text created by AI.
Just as a person's handwriting varies, models like GPT or Claude also leave subtle but identifiable patterns.
For example, word choice and sentence structure become the unique handwriting of that model.
By analyzing this, one can predict the source, like saying, "This sentence was written by GPT."
It is a very sophisticated technology that checks not just the comparison of texts, but also the probabilistic characteristics within the AI.
——
So why is the above fingerprinting technology important?
It's because AI-generated text is overflowing right now.
It's hard to know who created the text, and it's difficult to identify where it has been manipulated or is incorrect.
If this is left unchecked anywhere, the trust in AI will inevitably decline.
Fingerprinting helps maintain that trust.
In the future, asking 'Who created this?' will be a mechanism to determine the trustworthiness of content.
It makes sense that AI with increasing transparency will gain more attention.
Oct 21, 2025
[@SentientAGI ]
A new post has been updated.
It’s about SPIN-Bench, and since it’s quite lengthy, I’ve summarized it briefly 🫡
I believe that for AI to become smarter, it needs to collaborate or compete with others, so they created a testing ground called 'SPIN-Bench'.
It checks how intelligently AI behaves when planning over a long period and in situations where multiple people work together or compete,
and it’s said that not only simple problem-solving but also negotiation and deception detection are possible in 'SPIN-Bench'.
It seems like it will be a great help in improving the ability to handle complex strategies that take a long time or collaboration methods!
1.3K
Top
Ranking
Favorites