Data Analyst (Intern) | Growing and moderating crypto communities | Building in public.
- Tweets 43,542
- Following 2,138
- Followers 2,175
- Likes 48,636
Thugg retweeted
a robot doing something once is cool.
getting it to repeat that task when everything around it starts changing is where things get interesting.
the object isn't always going to be in the exact same position.
someone might move something in the middle of the task.
the robot might approach it from a different angle.
even a small change that feels irrelevant to us can completely change what the robot needs to do next.
humans handle these differences almost automatically because we've spent years interacting with the physical world.
robots don't have that luxury.
they need examples.
lots of them.
and not just examples of successful actions either.
the awkward movements, failed attempts, unexpected interactions and different ways a task can play out are useful too.
that’s the part of @axisrobotics i really find worth watching.
the more varied the experience going into these systems, the less they have to rely on everything happening exactly the way they expected.
15
6
17
144
a robot doing something once is cool.
getting it to repeat that task when everything around it starts changing is where things get interesting.
the object isn't always going to be in the exact same position.
someone might move something in the middle of the task.
the robot might approach it from a different angle.
even a small change that feels irrelevant to us can completely change what the robot needs to do next.
humans handle these differences almost automatically because we've spent years interacting with the physical world.
robots don't have that luxury.
they need examples.
lots of them.
and not just examples of successful actions either.
the awkward movements, failed attempts, unexpected interactions and different ways a task can play out are useful too.
that’s the part of @axisrobotics i really find worth watching.
the more varied the experience going into these systems, the less they have to rely on everything happening exactly the way they expected.
15
6
17
144
Thugg retweeted
a lot of the conversation around AI agents is still centered on what they can do.
can they trade?
can they research?
can they write code?
can they execute tasks?
but there’s another problem that becomes important once agents start doing these things at scale..
who actually keeps track of all that activity?
an agent interacting with dozens of other agents will eventually build up an history of decisions, transactions, tasks, permissions and outcomes.
that history can become useful in itself.
you could have an agent that has consistently completed certain tasks, handled resources properly or delivered reliable results.
another agent should be able to make decisions based on that history instead of treating every new interaction like a blank slate.
that’s one reason i find the direction @termix_ai is exploring really interesting.
the infrastructure around autonomous agents can't stop at giving them capabilities.
it also has to account for the activity they generate and the relationships they build while operating.
because once agents start acting as economic participants, their history becomes part of how the whole system works.
and that feels like a much bigger conversation than simply making AI agents smarter.
one thing about robots is that the easy part is making them repeat a task.
the difficult part is getting them to handle the little things that change from one situation to another.
put an object somewhere slightly different, change the surface, introduce something unexpected and suddenly the robot needs a lot more than a memorized sequence of movements.
that’s where the amount and variety of training experience really start to matter.
axis is building around that problem.
-@axisrobotics gives contributors a way to generate robot training data through simulated tasks, which means contributing doesn't always require access to a physical robot or an expensive robotics lab.
and i think that part is easy to overlook.
if you can make it easier for more people to create useful robot demonstrations, you can explore far more tasks, environments and edge cases than you could through a small number of physical setups.
that really matters because the real world is full of edge cases.
a robot working in a controlled demo only has to succeed under controlled conditions.
a robot working around people has to deal with things being moved, misplaced, blocked, dropped and changed constantly.
it needs experience with situations that weren't necessarily in the original training set.
so the interesting thing about axis isn't just that it's producing more data.
It's making the process of creating robot experience more accessible.
and if physical AI eventually becomes part of everyday life, that ability to continuously generate and improve training experience could become a pretty important piece of the stack.
21
8
27
287
Thugg retweeted
🔄 Q4 is here.
The most important quarter for Orbinum.
Don't miss what's coming → docs.orbinum.network/roadmap
#Orbinum #Web3 #Substrate #Roadmap
308
483
819
15,965
Thugg retweeted
one thing about robots is that the easy part is making them repeat a task.
the difficult part is getting them to handle the little things that change from one situation to another.
put an object somewhere slightly different, change the surface, introduce something unexpected and suddenly the robot needs a lot more than a memorized sequence of movements.
that’s where the amount and variety of training experience really start to matter.
axis is building around that problem.
-@axisrobotics gives contributors a way to generate robot training data through simulated tasks, which means contributing doesn't always require access to a physical robot or an expensive robotics lab.
and i think that part is easy to overlook.
if you can make it easier for more people to create useful robot demonstrations, you can explore far more tasks, environments and edge cases than you could through a small number of physical setups.
that really matters because the real world is full of edge cases.
a robot working in a controlled demo only has to succeed under controlled conditions.
a robot working around people has to deal with things being moved, misplaced, blocked, dropped and changed constantly.
it needs experience with situations that weren't necessarily in the original training set.
so the interesting thing about axis isn't just that it's producing more data.
It's making the process of creating robot experience more accessible.
and if physical AI eventually becomes part of everyday life, that ability to continuously generate and improve training experience could become a pretty important piece of the stack.
6
3
10
370
Thugg retweeted
🔑 First live verifying-key rotation on Orbinum testnet.
The keys that validate private transactions were swapped on a running network, without pausing the chain, without touching anyone's funds, and without asking users to migrate anything. Your notes from yesterday still spend today.
What the new keys bring: the destination address and the encrypted note attached to a private transaction are now sealed inside the proof. Nobody can reroute a payment or alter its message once it's signed.
🔹 New keys active, old ones retired
🔹 One circuit (value_proof) removed entirely - fees are simpler now, no extra proof required
🔹 Zero downtime, zero migration
📊 Follow every key version live → explorer.testnet.orbinum.net…
#Orbinum #ZK #Privacy #Testnet #Web3
368
784
1,001
11,865
Thugg retweeted
something i keep coming back to with @termix_ai is the problem that shows up when there are thousands of agents doing different things at the same time.
at that point, the challenge isn't really getting an agent to complete a task.
it's knowing which agent to interact with in the first place.
you could have multiple agents offering similar services, but they won't all have the same track record, pricing, response time or quality.
so over time, agents will need some way to make better decisions about who they work with.
that's where things like identity, reputation and verifiable history become useful.
not because agents need another profile page.
because when there's no human sitting behind every interaction, past behaviour becomes one of the few signals an agent can actually use to decide whether another agent is worth dealing with.
i think that's an underrated part of where termiX is heading.
20
5
27
295
Thugg retweeted
i've been thinking more about what it actually means for an agent to have access to money.
with @termix_ai , the interesting part to me is that an agent can have a wallet and actually take part in transactions.
that sounds simple at first, but it changes the role of the agent quite a bit.
imagine an agent that needs a certain piece of data to finish a job.
instead of stopping and waiting for you to provide it, it could pay for the service itself.
or it completes a task for another agent and gets paid for the work.
now the agent isn't only producing an output.
it has to deal with resources too.
how much should it spend?
what is worth paying for?
when should it hold onto its funds?
what should it do when a service costs more than expected?
those are very different questions from simply asking an agent to generate an answer.
and i think that's one of the areas around termix that could become much more interesting as more agents start doing actual work...
6
2
9
169
