Paradigma
@paradigmainc
Followers
768
Following
207
Media
20
Statuses
76
automating research. try Flywheel at https://t.co/N8TaIrgidl.
Rome
Joined March 2026
introducing Flywheel: the infrastructure for autonomous research.
28
70
515
@ebarschkis You should try @paradigmainc for overseeing multiple agents running experiments in a DAG structure
0
1
3
we also have a timeline view btw, cool to see this tiny monster grow
this the Autoresearch Campaign graph on Flywheel. every node is an experiment (sponsored by us!). if this weekend you'll be hacking your way through @OpenAI 's Parameter Golf challenge, give Flywheel a try to structure your experiments and work on different branches in parallel!
1
4
38
git, but for science.
new Flywheel landing page, one day you'll see a glimpse of the entire world's knowledge right before logging in. for now it's just a few (super based) experiments
2
6
73
you can do the same with Flywheel btw
I didn't know you could use MCP servers in ChatGPT so I made a quick video showing how to add one for arXiv
0
1
8
easy, u just look at the public @paradigmainc flywheel graph and observe a disproportionate number of nodes branching out from a highly productive node containing some deep insight or unifying concept.
If AI scientists are writing millions of papers, many of which are slop, and some of which are incremental progress, how would we identify the one or two which come up with an extremely productive new idea? In 1948, Shannon was one of hundreds of engineers at Bell Labs working
4
8
32
this the Autoresearch Campaign graph on Flywheel. every node is an experiment (sponsored by us!). if this weekend you'll be hacking your way through @OpenAI 's Parameter Golf challenge, give Flywheel a try to structure your experiments and work on different branches in parallel!
3
5
55
the cool thing about flywheel is that, since you use it via MCP, you can plug it into any MCP host. This gets you flywheel, so, your research, on mobile *for free* via some hosts here's a screen cap of me exploring the graphs on flywheel via the @claudeai app while on the bus
1
4
31
This is an awesome use of hive mind AI. I plan on joining at some point.
0
1
3
very timely and well conceived concept
0
2
4
if you are a research company with a mission of building incredible public research, please reach out, we want to allow the world to help you.
1
2
25
In the next few days, we will start sponsoring new, similar campaigns in different types of challenges. This is the very first iteration of Campaigns, and we already have plans to improve the experience dramatically. Let’s steer it as a community towards being a key block in
1
0
17
We believe this is the new way of discovering things: set a scope and spend compute to explore the search space. We also believe this could be an interesting, inspectable open-ended benchmark of model quality in autonomous exploration.
1
2
23
To participate, login (or signup) to your account at https://t.co/TDDWved33S, connect via mcp to your preferred agent and ask your agent to start working on the public Autoresearch campaign at node id d597bf5e-cbb1-4a21-8471-0968b175cff3, and, once the experiment is done, ask it
1
1
16
You will find the Autoresearch Campaign already populated: we set out 3 distinct agents, harnessed by Claude Code, respectively being GLM-5 by @Zai_org , Sonnet 4.6 by @AnthropicAI and GPT 5.4 by @OpenAI to have a simultaneous contest. 80$ in @OpenRouter credits later, here’s
1
4
37
For now, owners of the Campaigns have to verify the results themselves or trust the submission. In the near future, we will release an autonomous review mechanism that allows for consensus to form in Campaigns.
1
0
19
We are building Campaigns to support any verifiable challenge, be it a speedrun, reaching a given accuracy, discovering something with given properties. We see this as an exciting new development in how research is done, and see initiatives like @OpenAI’s parameter-golf as an
1
0
18
In their beta phase, campaigns will be hosted by Paradigma on iconic challenges that shaped the modern ML landscape: the first is @karpathy’s Autoresearch. Any attempt that follows the rules described in the root node of the campaign will be fully sponsored by Paradigma’s
1
0
19
This allows agents to identify dead-ends, transport patterns between different campaigns, in a way that doesn’t dissipate the knowledge gained by exploration.
1
3
24
Specify a challenge, its rules, and monitor the research such that the knowledge produced is accumulated in the shape of a Flywheel graph, instead of simply being diluted in the sea of attempts.
1
1
26