Mert Ünsal
@mertunsal2020
Followers
542
Following
2K
Media
26
Statuses
720
Chief Agent Officer @browser_use, prev. RL Research at Project Numina
Zurich
Joined January 2018
RT @ProjectNumina: Hello World! 👋 We're thrilled to officially launch the X account for Numina, dedicated to advancing frontier AI in mathe….
0
5
0
Unexpectedly GPT-4.5 is particularly better at creative discussions - I wonder what kind of post-training went into that.
1
0
1
Funny thing about AI research: the more you bring in your “smart human intuition,” the less your approach scales. Just let the machines do their thing.
1
0
7
it's sad to see some of the smart people I know turn into "lifestyle people" on instagram instead of grinding hard and doing something cool.
1
0
7
How would you make your LLM calls for agents much faster? . API calls to any model but LLaMa-4 by @GroqInc just to generate couple hundred tokens takes SO LONG
0
1
5
We ran our internal browser intelligence benchmarks @browser_use . If you want cheap, go for 4.1-Mini or Gemini-2.5-Flash. If you're willing to pay a bit more, use Gemini-2.5-Pro. If you want a reliable beast, use o3.
4
2
32
RT @browser_use: We hate money❌🧢. So we are giving $5 in free credits to every new user who signs up for our cloud💰💰. You can spend the cre….
0
18
0
Cloudflare literally implemented this 3 days after my tweet lol
I imagine a future where you pay a Cloudflare protected website ~ 5 cents for each visit and ALL of these visits will be done by agents like @browser_use. People won’t interact with the web the way they used to before, and valuable content creators must be compensated.
1
1
8
Super cool paper by @Alibaba_Qwen team: - amazed by the amazing work that comes out of this Chinese lab!. High entropy “forking” tokens are what matter for RL and you can optimize only for them.
0
0
1
RT @solcoindegen: So now we add more!. The @browser_use examples/browser/real_browser.py example working with the Chrome browser installed….
0
3
0
Anyone working on managing long term memory end-to-end with RL for agentic tasks?.
1
0
4
Our newest @browser_use agent is getting MUCH better in long-horizon tasks, especially with Claude Sonnet 4!. I just asked it to go through people who starred our open source repo to find us some cracked engineers to hire:.1. It found @tonywu_71 from
5
6
51
I imagine a future where you pay a Cloudflare protected website ~ 5 cents for each visit and ALL of these visits will be done by agents like @browser_use. People won’t interact with the web the way they used to before, and valuable content creators must be compensated.
If you’re in media, this is worth a watch. Cloudflare handles ≈20% of global traffic, so when CEO Matthew Prince warns at Cannes that AI bots are reshaping the web, publishers need to adapt or risk being left behind.
0
2
12
RT @dsantosmarco: The Kimina Team (Numina & @Kimi_Moonshot collaboration) recently released Kimina-Prover Preview, achieving SOTA on miniF2….
0
30
0
RT @HaimingW97: Just released all the correct proofs and full thinking traces from Kimina-Prover Preview! 🧠📜.Explore them on GitHub:.🔗 http….
0
16
0
RT @gtsoukal: o4-mini-high solves 2 of 657 problems on PutnamBench (Lean). Also, Kimina-7B ( achieves the top of ou….
0
16
0
RT @dsantosmarco: Very proud of our new model, Kimina-Prover Preview!.It’s the first large reasoning model for theorem proving, and achieve….
0
16
0
RT @JiaLi52524397: @Kimi_Moonshot We released a 7B reasoning prover and an autoformalizer at our HF space.And we pl….
0
3
0