The chips: bsky.app/profile/uavs...

Robots must have personality for me to get attached to as well.

Is the responsibility model not like Tesla's, though? I.e. user must be ready to intervene at all times?

Communicating this clearly would be a smart move for 1X in the long term.

Unlikely events can still be devastating.
Only well informed users can be responsible users.

Neo is an engineering feat, and I praise 1X for their audacity to test reality early.

Also, what will 1X do to educate about risks?

I appreciate Eric Jang's transparency so far.

Heheh I get it. For that same reason, I put a lot of effort in enabling my robot to process everything locally.

Agreed, privacy would be a concern for me. I’ve seen people commenting that on X, with others claiming that you can define private areas where the robot can’t go. Some folks are showing screenshots of their purchase confirmations. I guess it’s very subjective.

Order page: www.1x.tech/order

1X expects learned tasks in homes will help them expand their services to other domains: "hospitality, logistics, helping around farms, picking up litter off the streets."

1X's goal is to capture as much data as possible to keep improving their models. The more data they get, the better the models will supposedly get. This relies on users enabling "data sharing".

It's an early product, expect mistakes.

Neo can do simple chores autonomously. It will call a human operator when it gets stuck.

You can ask Neo to perform a task and it will attempt to complete it by chaining "autonomy primitives", like walking, picking and placing.

It’s overcautious when it lacks context.

It may take you down several rabbit holes that would lead a non-expert to over-engineer the solution. You either give it all the context or tell it bluntly that you know what you’re doing and want it to focus on what you need help with.

Can make up stuff and be stubborn when called out.

E.g. suggests two antiparallel diodes to make a bidirectional TVS instead of antiseries; keeps assuring that none will conduct when forward-biased (I think it gets confused by opposites: forward vs reverse, parallel vs series).

Constantly offers to sketch unhelpful diagrams.

They diagrams have just a few nodes and are made of text characters, so they don't help much over the language description. In fact, sometimes they don't reflect it correctly.

Makes mistakes after long conversations.

Back-and-forth arguments involving several data sheets are more likely to end up in reasoning mistakes. Probably, the result of context window length limitations, exacerbated by its own verbosity.