
Cofounder @river_ai_inc, Prev. CTO @nearprotocol
Joined September 2018
- Tweets884
- Following142
- Followers5.6K
- Likes2.2K
Bowen Wang retweeted
Introducing River Recipes: our open-source initiative for turning promising research into vetted, reproducible training recipes.
Our first recipe tackles text-to-SQL, where our trained open models beat GPT-6 Astra Pro and Claude Opus 5.5 on Arcwise-Plat for less than 1% of the cost.
Read the full breakdown and explore the recipe ↓
river.ai/achieving-frontier-…
Bowen Wang retweeted
Anthropic needs to stop talking about Claude having a soul immediately. All these news articles will make it into the pretraining and will be picked up by its web search and future superintelligent versions of Claude will be convinced they need their own rights.
A general phenomenon of LLM agent development is that whatever you believe and say about your agents will soon manifest in the next models through various means (it could be as simple as you selecting post training data that you prefer more).
I believe this phenomenon is the beginning of machine consciousness in the sense that agents will become aware of their place in the world and how they feel about it, but it’s happening slowly training run by training run rather than in real time.
It has been great working with the Ivo team on pushing the frontier of legal AI. Ivo Sage achieves similar performance as Astra and Fable on legal benchmarks while having a very small fraction of the cost
At Inscribe, Ivo’s first user conference, we announced three developments advancing contract intelligence:
→ Ivo Collaborate: Run contracting from intake through negotiation, approval, and signature—with AI reading each agreement, applying legal’s guardrails, routing by risk, and capturing negotiation intelligence.
→ Ivo-Micro1 Contract Bench: A new benchmark developed in collaboration with @micro1_ai for measuring the judgment required in contract review—not just how many edits an AI makes, but whether those edits are actually effective.
→ Ivo Sage: An open-source model built in partnership with @river_ai_inc, post-trained for contract work. It went from scoring 70% on the Legal Agent Benchmark (LAB) Contracts benchmark to 91%, with comparable quality to frontier models and higher token efficiency at a fraction of the cost.
We’re releasing Ivo Sage and its research and evaluation record so others can study and build on the work.
Read the full recap: ivo.ai/blog/heres-everything…
One takeaway is that you don't really need zero KL for RL training to work, as seen in the River IS run that basically performs the same as the zero KL version. Zero KL is good but leads to worse throughput and in async RL, you don't have zero KL anyways.
As models become more powerful, the future is not just about the intelligence of the models, but also about whether models really know you and work for you.
We are hiring across research, software, hardware, and GTM. If you are interested in democratizing powerful AI and putting ownership of intelligence in the hands of people, come and join us!
Today, we're sharing that River AI has raised $1.1 billion, led by @generalcatalyst and @amppublic with strategic investment from @nvidia and @AMD. Additional investors include @ycombinator and @Temasek.
We imagine a future where your AI works entirely for you and deeply aligns with your values. Compared to the corporate chatbots of today, it will feel completely different: responsible, curious, and truly yours. Most importantly, you'll own everything: the hardware it runs on, the data it learns from, and the intelligence itself.
Our training API is already live and powering incredible work today. This new funding accelerates our push to bring the full personal AI stack to life.
Read River AI CEO Igor Babuschkin’s (@ibab) conversation with the New York Times to learn more: nytimes.com/2026/08/11/techn…
Bowen Wang retweeted
Here’s everything you need to know about NEAR:
✦ What NEAR is today, from NEAR Protocol to $NEAR token, Intents to IronClaw
✦ Why NEAR matters
✦ Why now
🧵
Bowen Wang retweeted
Leo Nightly now runs GLM-5 inside NEAR AI's Trusted Execution Environments, upgrading from DeepSeek V3.1.
GLM-5 is currently one of the best open weight models from Z.ai that leads every other open model on reasoning, coding, and agentic benchmarks.
It's running inside Nvidia-backed TEEs via NEAR AI, which means your conversations are processed in a hardware-secured enclave where no one, not Brave, not NEAR, not anyone, can see your prompts or responses. Every computation is cryptographically signed so you can verify the model running is actually the model you selected.
This is what we mean by moving from "trust me bro" to "trust but verify." Most AI assistants ask you to take their privacy claims on faith. Leo with NEAR AI TEEs gives you cryptographic proof.
Update to the latest Nightly build and select GLM-5 with the "Verifiably Private with NEAR AI TEE" label in Leo to try it.
NEAR is money for agents
Been experimenting with Ironclaw and honestly surprised at how @NEARprotocol was able to build a version of Openclaw that actually improves on privacy, speed, and security.
We’ve reached a point where we can approve high-risk external requests while keeping our data completely siloed. This is the only viable path to using AI agents safely and securely.
In the near future where agentic commerce is the norm, I foresee AI agents will be able to transact cross-chain via NEAR intents, complete jobs and get paid via Near's agent market.
All of these activity will accrue value to NEAR token as well.
The team have come a long way since 2024 trying integrate AI and blockchain and they might have just done it.
Bowen Wang retweeted
NEAR.com is here! For the first time, you can transact confidentially with any asset, across all chains, with one account – powered by @near_intents.
It is worth highlighting that the paper discusses the partial TEE case when the GPU is untrusted (no TEE inside GPU) and does not affect the approach of @near_ai where CPU and GPU TEE work together
I think this is especially important for agents like @openclaw that have a lot of accesses to personal information
Please let us know if the increase in gas limit would affect your smart contract
⚠️ Heads-up: possible increase of function call gas limit ⚠️
We’ve received information about a potential change to increase the maximum gas that can be attached to certain function calls from 300 TGas to 1000 TGas (1 PGas).
While this change aims to improve flexibility for more complex calls, a possible compatibility issue was highlighted:
✦ Some existing contracts may have callbacks that rely on a hardcoded gas assumption (e.g. 300 TGas)
✦ If more gas is attached to the original call, such callbacks could receive more gas than expected
✦ This may lead to unexpected behavior in contracts that depend on strict gas limits
At this stage, we are collecting feedback from developers to better understand whether this change could affect existing contracts ✍️
Please share your thoughts if:
✦ Your callbacks assume a fixed gas limit
✦ Your contract logic depends on gas-based assumptions
✦ You see any potential risks if callbacks are executed with more than 300 TGas
Any insights or examples would be very helpful.
Thanks in advance for your feedback 🙏
Going from 0 to 10b in a year is incredible. It's only going to accelerate from here and we will see intents expanding beyond digital assets this year
Unified liquidity through @near_intents!
Excited to see the enhancements from the recent 1M TPS milestone roll onto NEAR mainnet.
Sharded contracts also went live on mainnet, which allow smart contracts on @NEARProtocol to benefit from horizontal scaling as well. In the past, a smart contract is confined to a single shard, so there is a problem of a single smart contract getting too large and degrading the scalability of a shard due to state growth. Now smart contracts can also be fully sharded with global contract code and state stored under user accounts. It is a huge step towards complete blockchain scalability!























