Content Creator $INJ $MATIC Staker. All in @SentientAGI
Joined August 2021
- Tweets1.4K
- Following619
- Followers363
- Likes6.2K
Pixelism retweeted
Yes, that was @balajis on our stage at @openagisummit 😉
He made the case that open source AI and crypto share incentives goals—both movements aim to align builders, reward verifiable work, and keep power decentralized.
Pixelism retweeted
Join us for a live AMA with Sentient cofounder @sandeepnailwal 🎙️
He’ll dive into Sentient’s open and decentralized AI vision, answer questions from the community, and share a look at what’s next for Sentient.
Post your questions below—will queue them for the AMA!
🗓️ October 27th
⏰ 10 AM PT
Pixelism retweeted
My art goes beyond paper this time I showcased my love for @SentientAGI on a wall, though even it couldn’t handle its heat and grandeur.❤️🔥🔥
@shad_haq_ @smoothasfkk @scalpkripto @hieu06730313 @Kriptoloji123 @vivekkolli
Pixelism retweeted
Sentient'te Türk topluluğu 35 kişiyle ATH tazeledi!
Sentient'te bir sonraki Discord rolleri dağıtıldı. Türk topluluğu olarak yine çalışkanlığımız ve dolu portfoylerimizle öne çıktık. Çalışkanlığı ve özverisiyle herkese örnek olan topluluğumuzu tebrik ederiz!
🇹🇷 Türk topluluğumuzdan rol kazananların listesi :
1. @comoloko0618
2. @Pixipixels
3. @goodhypeonly0
4. @kaancarlione
5. @Gr1mmJ4w
6. @theHaneth
7. @yeim123
8. @avaxcrypto
9. @ragnarrcrypto
10. @Jabotacaba_
11. @BSongulalp10
12. @SekulerTrader
13. @literallyHS824
14. @EminAhmadov14
15. @meli1ssandre
17. @airdrop08nuri
18. @wakawakaeth
19. @josepnft1
20. @YalmPek
21. @agozumm
22. @valeriiacr69
23. @Benyaksav
24. @sprmn2024
25. @0Xakhenaton
26. @CryptoSpiderM
27. @klaus_michels0n
28. @eylulyldrm34
29. @0xbarlas
30. @alperencrypt
31. @enderakici1
32. @slntpkra36838
33. @0xdefimaster
34. @nesrin_33_33
35. @RdLine4
Her hafta birlikte motivasyonumuzu bozmadan büyümeye, yolumuza devam ediyoruz.
Birlikte başaracağız!
Pixelism retweeted
ℙ𝕚𝕩𝕖𝕝𝕚𝕤𝕞 𝕏 𝕊𝕖𝕟𝕥𝕚𝕖𝕟𝕥
ℝ𝕆𝕄𝔸 ℝ𝕖𝕤𝕖𝕒𝕣𝕔𝕙 𝕧𝕤 ℂ𝕠𝕞𝕡𝕖𝕥𝕚𝕥𝕠𝕣𝕤
𝗖𝗼𝗺𝗽𝗹𝗲𝘅 𝗥𝗲𝘀𝗲𝗮𝗿𝗰𝗵 𝗤𝘂𝗲𝘀𝘁𝗶𝗼𝗻𝘀
When facing a complex question, 𝗦𝗲𝗻𝘁𝗶𝗲𝗻𝘁 𝗥𝗢𝗠𝗔 does not process it as a single opaque block. Instead, the 𝗔𝘁𝗼𝗺𝗶𝘇𝗲𝗿 detects complexity and the 𝗣𝗹𝗮𝗻𝗻𝗲𝗿 decomposes it into atomic subtasks. Each subtask is assigned to an 𝗘𝘅𝗲𝗰𝘂𝘁𝗼𝗿 that applies the most appropriate agent or tool.
A layered research query such as “What are the long term economic impacts of renewable energy adoption in European markets?” can be split into subtasks like analyzing historical policies, modeling economic growth, and comparing country statistics. Each output flows back to the 𝗔𝗴𝗴𝗿𝗲𝗴𝗮𝘁𝗼𝗿, which merges everything into a coherent and reproducible answer. Every stage is visible and 𝘃𝗲𝗿𝗶𝗳𝗶𝗮𝗯𝗹𝗲, so if one subtask fails it can be repaired in isolation without collapsing the entire process. This makes ROMA both reliable and transparent in handling multi dimensional research.
Competitor systems rely on a single 𝗹𝗮𝗿𝗴𝗲 𝗹𝗮𝗻𝗴𝘂𝗮𝗴𝗲 𝗺𝗼𝗱𝗲𝗹 to process the whole query in one extended pass. While the text may appear fluent, the reasoning remains inside a 𝗯𝗹𝗮𝗰𝗸 𝗯𝗼𝘅. Without explicit decomposition subtasks are only simulated internally, increasing the chance of missed details or overgeneralization. As complexity grows, 𝗲𝗿𝗿𝗼𝗿𝘀 𝗰𝗼𝗺𝗽𝗼𝘂𝗻𝗱 and the polished output may hide a lack of depth or accuracy. ROMA builds a structured pipeline of research, while competitors gamble on a single shot.
𝗔𝗰𝗮𝗱𝗲𝗺𝗶𝗰 𝗟𝗶𝘁𝗲𝗿𝗮𝘁𝘂𝗿𝗲 𝗥𝗲𝘃𝗶𝗲𝘄
In academia, reviewing literature is not just summarizing it requires comparing findings, identifying gaps, and synthesizing conclusions. 𝗦𝗲𝗻𝘁𝗶𝗲𝗻𝘁 𝗥𝗢𝗠𝗔 addresses this by letting the 𝗣𝗹𝗮𝗻𝗻𝗲𝗿 map a workflow of subtasks. One agent retrieves relevant publications, another creates summaries, a third compares methodologies, and a fourth highlights contradictions or missing areas.
The 𝗔𝗴𝗴𝗿𝗲𝗴𝗮𝘁𝗼𝗿 then unifies these results into a structured review resembling human academic writing. Transparency is guaranteed through 𝘀𝘁𝗮𝗴𝗲 𝘁𝗿𝗮𝗰𝗶𝗻𝗴, so researchers can verify not only what conclusions are presented but also 𝗵𝗼𝘄 they were reached. This accountability turns ROMA into more than just a tool it acts like a collaborative research assistant.
Competitors typically attempt literature review with a 𝘀𝗶𝗻𝗴𝗹𝗲 𝗹𝗼𝗻𝗴 𝗽𝗿𝗼𝗺𝗽𝘁. Researchers paste dozens of abstracts and ask for a summary, but the result often lacks structure and depth. It may repeat points, miss contrasts between studies, or even generate hallucinations. Since the 𝗿𝗲𝗮𝘀𝗼𝗻𝗶𝗻𝗴 𝗽𝗮𝘁𝗵 is hidden, it is unclear whether the output reflects real synthesis or surface level pattern matching. ROMA builds a 𝗿𝗲𝗽𝗿𝗼𝗱𝘂𝗰𝗶𝗯𝗹𝗲 𝗽𝗶𝗽𝗲𝗹𝗶𝗻𝗲 of analysis, while competitors deliver an attractive but fragile text.
Follow Sentient Members;
@SentientAGI ❤️
@sentient_chat ❤️
#sentientchat ❤️
@shad_haq_ ❤️Chad-haq
@scalpkripto ❤️ Turkish-Chad
@vivekkolli ❤️ Degen
@LeaderX_btc ❤️ Sensei
Pixelism retweeted
2/ Products of the GRID
Recursive Open Meta Agent (ROMA): Multi-agent framework that breaks down complex queries into smaller, more achievable tasks. The first use-case (deep research) outperforms all existing deep research model platforms and is open for builders to fork/improve.
Open Deep Search (ODS): Open-source search framework that splits reasoning, search, and calculation tasks, achieving SOTA performance on standard search benchmarks.
Dobby: Open-source model that was one of the first models to prioritize human-like values (tone, loyalty/core beliefs, etc.) and built through 2M+ community members’ feedback.
Fingerprinting: Training library/method that protects a model’s identity, ensuring that open-source developers can verifiably prove their model ownership even if the model weights are fully revealed.
Sentient Chat: AI platform that coordinates intelligence within the GRID, providing a UI interface to the world's largest network of intelligence.
Pixelism retweeted
We've arrived😈
Who's ready to make open-source AI win?
The @SentientAGI Leaderboard is now live!
Sentient is making open-source AI win.
Their mission is to ensure Artificial General Intelligence is open-source and not controlled by any single entity.
Find our who's driving forward their mission and mindshare.
Link below! 👇
Pixelism retweeted
The future of AGI will not be a single giant model.
It will be a network of intelligence, thousands of agents, data sources, and reasoning systems working together like a distributed brain.
This is exactly what @SentientAGI is building with THE GRID.
Let me explain 👇
1. What is GRID?
GRID is the world’s largest coordinated intelligence layer. When you send a query into Sentient Chat, it doesn’t just go to one model. It is:
• Classified into workflows (research, writing, finance, travel, etc.)
• Atomized into smaller tasks
• Routed to the most capable agents, models, and data providers
• Combined back into one powerful answer
The result: outputs stronger than any single AI system alone.
2. Agentic Capabilities
GRID already integrates 50+ live agents in Sentient Chat:
• @MessariCrypto — institutional research, metrics, fundraising
• @askjavin — wallet & contract analysis
• @UniQAgent — cross-chain execution & DeFi ops
• @GoKiteAI — Ethereum transaction evaluation & fraud detection
• @cookbook_dev — developer documentation deep-dives
• @DappLooker — whale activity & smart money tracking
• @GigabrainGG — alpha, research feeds, DAO governance signals
• @zerufinance — wallet reputation scoring
Each agent is like a brain cell, specialized but connected into the bigger brain: the GRID.
3. Data Partners
Intelligence requires real-world grounding. GRID already integrates 100+ data sources across industries:
• @graphprotocol, @spaceandtime, @Covalent_HQ; blockchain indexing
• @KaitoAI, @getmasafi; social data & crypto protocol intelligence
• @nansen_ai, @arkham; on-chain analytics & de-anonymized activity
• @crunchDAO, @PerleLabs, @JoinSapien, @LabelLedger; human-labeled, high-fidelity datasets
• @0G_labs, @irys_xyz, @Hyve_DA, @zus_network; decentralized, permanent data storage
• @vana, Entrova, @turf_network; behavioral, consumer, and real-time signals
4. The Architecture
GRID goes beyond static workflows:
• Queries are recursively atomized into atomic tasks until fully solved
• Each atomic task is routed to the best intelligence via a community-driven prompt engine
• Token-level routing enables queries to be decomposed into tokens, each routed independently across models, then stitched back into one output
Early tests already show this outperforms closed labs like OpenAI and Anthropic, at significantly lower cost.
5. Why it Matters
Closed-source AI is monopolized. GRID ensures AGI is open and community-owned:
• Anyone can contribute agents, data, or models
• $SENT staking directs funding and resources where they matter
• Rewards flow to the most useful participants
• Distribution happens through Sentient Chat
Open-source is no longer behind. With GRID, it is compounding intelligence at scale.
This is how humanity wins the future of AGI.
Pixelism retweeted
With 100+ partners in the GRID, we’re scaling open-source intelligence across every dimension.
Sentient Chat connects users to the world’s largest network of intelligence, delivering high-quality answers across industries.
🧵Here’s a look at some of the data partners fueling the experience
Pixelism retweeted
One of the best events we've ever had, here were some of my favorite talks today:
1. Zerui Cheng (Princeton; @ZeruiCheng): LiveCodeBench Pro: How Do Olympiad Medalists Judge LLMs in Competitive Programming?
Recapped our collaborative research with a number of universities and ex-competitive programmers (insanely cracked) on LiveCodeBenchPro—a more accurate coding benchmark that measures LLMs' capabilities in solving the hardest competitive programming problems. If you haven't read about it already, this thread covers it pretty well:
👉nitter.cf/SentientAGI/status/193…
2. John Yang (Stanford; @jyangballin): SWE-smith: Scaling Data for Software Engineering Agents
Shared his work about optimizing the process for his previous work on SWE-bench (a benchmark that essentially measures how good models are as a SWE. In this case, it's defined as how well models can resolve GitHub issues across some popular Python repositories). Honestly, I'm not that tapped into software engineering, but it was interesting to hear about the process of splicing down a broad problem into a deterministic structure (what makes a good software engineer? --> did this model solve this GitHub issue?).
3. Salah Alzubi (Sentient): ROMA (Recursive Open Meta-Agents)
My favorite talk of the day (maybe I'm just biased or glazing, but felt like I understood 90% of it). Longer announcement post coming out about this research we did, but our research team absolutely cooked and created an agentic framework that breaks down large tasks into smaller, more achievable ones. The first use-case we focused on was deep research (in-depth reports and understanding large amounts of information for accuracy), and we're SOTA across all standard search benchmarks (FRAMES, SealQA, and SimpleQA). And yes, even better than ChatGPT, Grok, Perplexity, Claude, etc.
4. Mike Merrill (Stanford; @Mike_A_Merrill): Terminal-bench: A Benchmark for AI Agents in Terminal Environments
Again, not really an engineer, but it was an interesting talk to follow along after I've gained some understanding of benchmarking. His work on Terminal-Bench measures how well models can handle tasks within the terminal (which is quite broad and difficult). Even though there is a set number of instructions, it still requires some creativity/reasoning to achieve the end goals. Got a chance to chat with him afterwards, and he had some interesting thoughts about the shell being the optimal user-interface (not the agentic VLMs we see today; eg. Perplexity Comet pointing and clicking on your tab for you). Thinking of some really interesting things we could do in showing how ROMA can be a superpower in engineering bug-finding (stay tuned👀)
5. Peiyao Sheng (Sentient; @peiyaosheng): Real AI Agents with Fake Memories: Fatal Context Manipulation Attacks on Web Agents
I've seen/ingested this work since March, but the presentation was great. If you haven't read about our work on agent security, check out the tweet threads here:
👉 nitter.cf/SentientAGI/status/190…
👉nitter.cf/SentientAGI/status/191…
👉nitter.cf/SentientAGI/status/194…
Honorable Mention: Chatting with @tuvllms on our walk to dinner and understanding his work with SealQA benchmarking. I was a big fan of FRAMES (because these are the type of questions I unknowingly ask and am still deeply dissatisfied with current LLM performance on these; eg. "What are the top 5 NBA players by PPG average in a season that have won both an NCAA college basketball championship and an NBA championship?"). SealQA is basically a harder version of FRAMES and should theoretically encompass FRAMES performance based on its difficulty. A fair counterpoint to this is that the FRAMES question set is also easier to aggregate, and in some instances will do a good enough job to measure an LLM's ability on these types of difficult questions (it took ~1 hour per question for collection and review on average for SealQA). Funny coincidence, he mentioned that one of the FRAMES creators and he were advisees of the same professor for their PhDs, and both worked at Google DeepMind together.
After today's event, I realize how fortunate I am to be surrounded by some of the smartest people in the world (both the Sentient team and the people I'm able to meet through working here; longer post coming about this later😁).
Great turnout at UC Berkeley for our Open Source AI workshop with @SentientAGI.
Speakers from Nvidia, Stanford, Princeton, and Berkeley covered everything from math reasoning and live code benchmarks to software agents and AI memory attacks.
Big thanks to all the speakers, researchers, and builders who came together to explore how open-source systems can power the future of AI development.
Pixelism retweeted
A great overview from @stacy_muur covering our work across the AI stack. Over the last two months, all of our research has been fitting into place to offer users the best open AI ecosystem experience through Sentient Chat.
Big research announcement coming Monday😁
Pixelism retweeted
DATA CONSORTIUM PARTNER: COVALENT
Give a warm welcome to @Covalent_HQ—the on-chain data powerhouse joining the Sentient Data Consortium!
Covalent’s integrated platform indexes structured, verifiable data from 100+ blockchains and serves 2,000,000+ API calls every day, giving builders sub-second access to the signals that drive next-gen AI and Web3 apps.
📌 Universal On-Chain Coverage: One API surface for 100+ networks, ready for any agent, wallet, or dashboard you spin up.
🚀 Sub-Second, Verifiable Feeds: Endpoints push real-time data into trading, monitoring, and discovery workflows.
🤝 Shared Infra & Support: Dedicated bandwidth and hands-on help so Sentient agents never miss a beat with broad on-chain data coverage.
As Sentient Chat continues to expand in the multi-agent future, Covalent provides instant data resources to keep agent information accurate. Welcome to the Sentient Data Consortium, Covalent!
🚀ANNOUNCING THE SENTIENT DATA CONSORTIUM PARTNERS V3
Sentient is building the largest network of intelligence in the world, enriched with the best agents, models, and domain-specific data.
Our newest addition of partners (@DappLooker, @Covalent_HQ, @cookiedotfun, and @PerleLabs) encompasses deep data across the web3 industry and beyond, including on-chain analytics, blockchain-agnostic trading data, and expert-labeled data.
Sentient Chat is evolving to include new data partners every day…but we’re cooking up something even bigger. Over the coming days, we’ll spotlight each new consortium member and their unique value to the Sentient ecosystem.
Pixelism retweeted
DATA CONSORTIUM PARTNER: PERLE
We’re excited to welcome @PerleLabs to the Sentient Data Consortium!
Perle runs high-fidelity data pipelines for AI teams spanning code, advanced reasoning, multilingual content, satellite imagery, and other safety-critical domains. Instead of gig work, Perle taps a vetted global network of specialists to deliver labels and evaluations.
📌 Specialist-Led Accuracy: Domain experts handle high-context, reasoning-heavy tasks for first-time-right results.
🌍 Global & Multilingual Reach: Coverage across regions and languages that traditional labelers overlook.
With Perle joining the Sentient Data Consortium and integrating their data into Sentient Chat, users will have data of the highest quality vetted by professionals across the globe. Excited to have @PerleLabs on board!
🚀ANNOUNCING THE SENTIENT DATA CONSORTIUM PARTNERS V3
Sentient is building the largest network of intelligence in the world, enriched with the best agents, models, and domain-specific data.
Our newest addition of partners (@DappLooker, @Covalent_HQ, @cookiedotfun, and @PerleLabs) encompasses deep data across the web3 industry and beyond, including on-chain analytics, blockchain-agnostic trading data, and expert-labeled data.
Sentient Chat is evolving to include new data partners every day…but we’re cooking up something even bigger. Over the coming days, we’ll spotlight each new consortium member and their unique value to the Sentient ecosystem.
Pixelism retweeted
Our findings on agent security
Our latest collaboration with @Princeton researchers introduces plan injection, an attack that corrupts an AI agent’s internal task plan rather than its prompt, allowing malicious actions to bypass existing defenses. By embedding these instructions as a seemingly natural continuation—context-chained injection—we show how easily privacy can be breached, underscoring the imperative for secure memory handling in agentic systems.
Check out the full paper:
👉 alphaxiv.org/abs/2506.17318
