Daniel Taratorin

Building @midpointai. Previously co-founder @ Merge, Midpoint Labs, ran one of the largest Minecraft servers at 13.

[Daniel Taratorin @Daniel_Tarat]

Claude Opus 4.6 on OSWorld: 72.7%
Average human on OSWorld: 72.36%

[Daniel Taratorin @Daniel_Tarat]

2026 is shaping up to be a crazy year

[Daniel Taratorin @Daniel_Tarat]

Anybody scaling this paper? Learning from video seems like the way to get internet scale data for pre-training generalizable humanoids and this is just 240 hours of data

Ideally would model the physical world & manipulation, not just human pose

https://t.co/ieev9tFVvU

[Daniel Taratorin @Daniel_Tarat]

@udhav_goenka

[Daniel Taratorin @Daniel_Tarat]

Merge is building the next generation of RPA.

Today we are announcing M1, the autopilot for the computer. M1 trains your computer to use itself.

To use and train M1 we built Mind, the memory layer. Mind gives computers a long term memory that organizations and agents use. 🧵↓

[Kamil @Kamil_SaaS]

Dan is cracked. Def check this out 🚀

[Proby Shandilya @ProbyShandilya]

When you talk to Dan, you can’t help but feel a rush of ambition and excitement. Since day one, he’s grasped the complete range of possibilities that are opened up w/ AI and it’s inspiring to see him traverse that possibility space

Congrats !

[Pranav Kanchi @pranav_kanchi]

Always hear about AI empowering humans to be more productive but its rare to see the real deal in action

[Daniel Taratorin @Daniel_Tarat]

@ProbyShandilya Thanks Proby! 🚀
Computers WILL be automated 🤖

[Daniel Taratorin @Daniel_Tarat]

@karpathy @RichardMCNgo https://t.co/iNksTsAGCz

[Daniel Taratorin @Daniel_Tarat]

Merge is building the next generation of RPA.

Today we are announcing M1, the autopilot for the computer. M1 trains your computer to use itself.

To use and train M1 we built Mind, the memory layer. Mind gives computers a long term memory that organizations and agents use. 🧵↓

[Daniel Taratorin @Daniel_Tarat]

@karpathy @RichardMCNgo Built this and a whole lot more — releasing soon!

[Daniel Taratorin @Daniel_Tarat]

@chrishlad https://t.co/iNksTsAGCz

[Daniel Taratorin @Daniel_Tarat]

Merge is building the next generation of RPA.

Today we are announcing M1, the autopilot for the computer. M1 trains your computer to use itself.

To use and train M1 we built Mind, the memory layer. Mind gives computers a long term memory that organizations and agents use. 🧵↓

[Daniel Taratorin @Daniel_Tarat]

@chrishlad Releasing this Monday and a whole lot more 🫡

[Daniel Taratorin @Daniel_Tarat]

@posedscaredcity @ptremblay @teknium @mergeos_ai https://t.co/iNksTsAGCz

[Daniel Taratorin @Daniel_Tarat]

Merge is building the next generation of RPA.

Today we are announcing M1, the autopilot for the computer. M1 trains your computer to use itself.

To use and train M1 we built Mind, the memory layer. Mind gives computers a long term memory that organizations and agents use. 🧵↓

[Daniel Taratorin @Daniel_Tarat]

@posedscaredcity @ptremblay @teknium @mergeos_ai Monday is the date 🙏

[Daniel Taratorin @Daniel_Tarat]

@pranav_kanchi 🚀🚀🚀

[Daniel Taratorin @Daniel_Tarat]

@karpathy knows 👀
https://t.co/mWmjeu3Cz5

[Andrej Karpathy @karpathy]

@RichardMCNgo And ideally with a bit more instrumented harness to capture other latents, eg current goals, chain of thought.

[Daniel Taratorin @Daniel_Tarat]

4. What is next?
The best way to prepare for the incoming Artificial Intelligence wave is to develop an organizational and personal memory, an understanding of what you do today that creates value. The earlier you start, the better off you will be.

What is next for Merge? We're planning to improve Mind with more proactive features for you and for your M1's. We also are seeking early partners that are comfortable with the idea of using your organizational data to automate workflows, through using the data collected by Merge to guide vision agents deployed on behalf of your company. If you're interested and want to start creating organizational memory today, join the waitlist on our website!

https://t.co/Sb9xfiiYCs

If you're interested in working together, shoot me an email:
[email protected]

[Daniel Taratorin @Daniel_Tarat]

3. Mind <> M1
Part of the data that Mind creates as people go throughout their day is in the same format as M1's inner thought. This isn't a coincidence: this data is used to train M1 to allow it to learn from an abundance of human signal than is currently not available in any other dataset. This data is then used for in-context learning so that M1 can learn your specific approach for following a workflow.

Currently, M1 is in alpha testing. In the future, Mind will be the base tool for monitoring your M1 deployments. You can use Mind to create M1 instances on their own computers, track them as they go throughout their tasks, and manage them from one interface. M1 can also talk and get new tasks from traditional communication channels like Slack.

Finally, M1 learned from people that use Mind during their day to day. M1 has learned to use Mind as a tool to help remember things that it has done, and to hook into your Mind to have a clearer conception of what you want done, how you want it done, and retrieve data it needs to do a task on your behalf.

For future research, we have the early inklings of a system that can predict what the optimal path to take is based on what you are doing. With relative accuracy, it can predict what you are going to do before you do it. We're planning to enhance this system to build a general autocomplete for using the computer, similar to the "Tab" feature in Cursor.