GPT6 Astra playing factorio space age has just gotten to its 3rd planet - miles ahead of what last weeks Fable 5.1 managed, and its looking like it will be able to complete the game in a week.
You can watch the stream at twitch.tv/danielv1234
Conversation
It is running a custom client side harness and playing on a vanilla multiplayer server.
For comparison, this is where fable gave up on its 12h attempt with the first 2 science packs unlocked (Astra has the first 7)
As another point of comparison, I couldn't get gpt 5.6 Luna High to automate anything at all
The harness uses a custom client side Lua sandbox that allows the agent to run scripts to query the game world and play inputactions, which are how the game sends keypresses in multiplayer. Yesterday it also figured out how to use other codex threads to update its own harness.
This has lead to great improvements in pathfinding and being able to do more things I didn't think to add to the harness without computer use (such as launching rockets - I never thought that it would get close to that)
The cost is just ~1 weekly 200$ sub per 24h. Thankfully has been helpful with the resets, but currently it looks like the run will have to end/pause tomorrow as my usage runs out.
It also shows the ability to learn in context. This (forgive terrible picture) is its first spaceship, with turrets only on one side. On its first voyage beyond nauvis it was destroyed by asteroids.
As soon as it happened, it started rebuilding, learning from its mistakes by adding more turrets up front, and a big ammunition buffer.
We saw similar progress on vulcanus - after failing to kill a worm with yellow ammo, it researched damage upgrades and came back with more turrets, piercing ammunition and circuit controls (ruins of old turrets in the bottom right)
However, it doesn't seem to be learning much about how to make factory layouts. It started out about like a beginner human would, but hasn't improved much from that and makes some rather inhuman layouts at times.
Watch it progress at twitch.tv/danielv1234
Awesome! I have never played Factoria but I am surprised it is such a struggle for a frontier model with the right harness.
I would have thought that the theme of the game, puzzling automation and flows, etc. would be well-suited to an LLM, similar to coding.
Is it
What's the difference from chess or go or anything that ai did already 10 years ago?
Discover more
Sourced from across X
>try to warn conservatives about runaway AI
>think up a strategy
>'What's something scary?'
>eureka!
>tell them agents are 'algorithmic immigrants'
incredible timeline
Quote
nico
@nicochristie
0:10
I turned the fly bisexual
Inspired by Kallman et al ('15), I blocked mAL output in a 166,606-neuron fly model and measured responses in 8 candidate P1 male courtship neurons.
In 3 experiments, when cued by other males, baseline P1 spikes rose from 0 to 1–5 after mAL blockade
just got off video call with parents. explained to mom we could infer time travel was impossible since no robot assassins had tried to kill her. they claimed that actually there'd been robot assassins showing up all the time and they hadn't wanted to worry me.
these excerpts are crazy, someone is going to get in capital-T Trouble because of this
Quote
Anthropic
@AnthropicAI
We're publishing our most detailed threat intelligence report to date.
It covers how people tried to misuse Claude—for cyberattacks, influence operations, surveillance, biology, and building weapons—and how we found and stopped them.
We disrupted every operation in the report,