Axe Ghost. Now with Fragments mode
@axeghostgame.bsky.social
Cleverly push monsters around a grid and blow them up.
Demo👉 https://store.steampowered.com/app/2994110/
Demo👉 https://store.steampowered.com/app/2994110/
Pinned
Axe Ghost is out now on Steam!
* Smash the monsters
* Compact turn-based tactics
* Short, intense runs
* Leaderboard rivalry
store.steampowered.com/app/2712670/...
#TurnBasedTactics #IndieGame
* Smash the monsters
* Compact turn-based tactics
* Short, intense runs
* Leaderboard rivalry
store.steampowered.com/app/2712670/...
#TurnBasedTactics #IndieGame
'stochastic parrot' is technically defensible and conceptually useless. it's the same move as referring to humans as 'meat bags' or a masterpiece as 'daubs of pigment': true at one level, distorting at every level that matters.
Erdos problems, a set of famous difficult math challenges, are a clear example of AI models breaching a threshold. The idea that an AI could solve one, let alone many, would have been insane a year ago (o1 was brand new). Now we have multiple Erdos problems solved by GPT-5.2 in the last couple weeks
January 18, 2026 at 7:34 AM
'stochastic parrot' is technically defensible and conceptually useless. it's the same move as referring to humans as 'meat bags' or a masterpiece as 'daubs of pigment': true at one level, distorting at every level that matters.
Reposted by Axe Ghost. Now with Fragments mode
long before Claude Code writes its first line of code for you, it earns its value by being the most effective diagnostic and debugging tool in computing history
this is nowhere in the conversation. don't get your time wasted
this is nowhere in the conversation. don't get your time wasted
January 17, 2026 at 4:59 PM
long before Claude Code writes its first line of code for you, it earns its value by being the most effective diagnostic and debugging tool in computing history
this is nowhere in the conversation. don't get your time wasted
this is nowhere in the conversation. don't get your time wasted
Reposted by Axe Ghost. Now with Fragments mode
Erdos problems, a set of famous difficult math challenges, are a clear example of AI models breaching a threshold. The idea that an AI could solve one, let alone many, would have been insane a year ago (o1 was brand new). Now we have multiple Erdos problems solved by GPT-5.2 in the last couple weeks
January 18, 2026 at 3:38 AM
Erdos problems, a set of famous difficult math challenges, are a clear example of AI models breaching a threshold. The idea that an AI could solve one, let alone many, would have been insane a year ago (o1 was brand new). Now we have multiple Erdos problems solved by GPT-5.2 in the last couple weeks
huh, valve has changed the wording on the AI section of the content survey. which changes the meaning of 'yes'. were any devs who'd filled this out notified about this? setting the answer back to 'no'.
January 17, 2026 at 7:16 PM
huh, valve has changed the wording on the AI section of the content survey. which changes the meaning of 'yes'. were any devs who'd filled this out notified about this? setting the answer back to 'no'.
the face proportions of hytale characters are a bit cursed
[. ] [. ]
-----
[. ] [. ]
-----
January 17, 2026 at 1:36 PM
the face proportions of hytale characters are a bit cursed
[. ] [. ]
-----
[. ] [. ]
-----
"one the largest datacenters in the world only consumes as much water as 2.5 In-N-Out’s." [a burger place]
SemiAnalysis estimates Colossus 2 uses about 346 million gallons of blue water per year, roughly 2.5 times a typical In-N-Out store's annual use (~147 million gallons).
From Tokens to Burgers – A Water Footprint Face-Off
A different perspective on the datacenter water debate, forget tokens/watt or tokens/dollar, it's about tokens/burger, how many In-N-Outs is equivalent to the world's largest datacenter
newsletter.semianalysis.com
January 17, 2026 at 7:17 AM
"one the largest datacenters in the world only consumes as much water as 2.5 In-N-Out’s." [a burger place]
"The LLM can generate a view you couldn’t have produced yourself—not because it’s smarter, but because it’s drawing on patterns from across its training distribution. You see that projection, react to it, redirect. The space of accessible projections expands."
An attempt to express how I principally use LLMs.
Rotating the Space: On LLMs as a Medium for Thought
sbgeoaiphd.github.io/rotating_the...
Rotating the Space: On LLMs as a Medium for Thought
sbgeoaiphd.github.io/rotating_the...
January 16, 2026 at 8:10 PM
"The LLM can generate a view you couldn’t have produced yourself—not because it’s smarter, but because it’s drawing on patterns from across its training distribution. You see that projection, react to it, redirect. The space of accessible projections expands."
Reposted by Axe Ghost. Now with Fragments mode
just realized the dril drunk driving tweet also works for regular driving
October 2, 2025 at 3:23 AM
just realized the dril drunk driving tweet also works for regular driving
current claude workflow for godot project
* write step doc (.md) describing a feature + acceptance tests
* using step doc, py script loads relevant files, runs GDUnit tests, builds minimal context
* claude implements feature, writes tests
* tests fail? invoke claude w error log until they don't
* write step doc (.md) describing a feature + acceptance tests
* using step doc, py script loads relevant files, runs GDUnit tests, builds minimal context
* claude implements feature, writes tests
* tests fail? invoke claude w error log until they don't
January 16, 2026 at 3:36 PM
current claude workflow for godot project
* write step doc (.md) describing a feature + acceptance tests
* using step doc, py script loads relevant files, runs GDUnit tests, builds minimal context
* claude implements feature, writes tests
* tests fail? invoke claude w error log until they don't
* write step doc (.md) describing a feature + acceptance tests
* using step doc, py script loads relevant files, runs GDUnit tests, builds minimal context
* claude implements feature, writes tests
* tests fail? invoke claude w error log until they don't
Reposted by Axe Ghost. Now with Fragments mode
the Buddha released you from the obligation to have a take
January 16, 2026 at 4:12 AM
the Buddha released you from the obligation to have a take
Reposted by Axe Ghost. Now with Fragments mode
Heard a lad earlier talking about ‘Belgian whistles’
“A basic website costs 10k or 25k upwards if you want all the Belgian whistles”
BELGIAN WHISTLES
“A basic website costs 10k or 25k upwards if you want all the Belgian whistles”
BELGIAN WHISTLES
January 15, 2026 at 4:33 PM
Heard a lad earlier talking about ‘Belgian whistles’
“A basic website costs 10k or 25k upwards if you want all the Belgian whistles”
BELGIAN WHISTLES
“A basic website costs 10k or 25k upwards if you want all the Belgian whistles”
BELGIAN WHISTLES
Reposted by Axe Ghost. Now with Fragments mode
who is good with vectors that I can commission for a logo for switchyard
January 15, 2026 at 6:44 PM
who is good with vectors that I can commission for a logo for switchyard
Reposted by Axe Ghost. Now with Fragments mode
This tool offloads code exploration to local LLMs then uses Claude models to do the planning, design, etc. more code discovery means fewer hallucinations. Local model means cheap.
A little off topic but you two might find this a mildly interesting/ useful even just as parts for other things.
RLM proxy for context generation.
github.com/colliery-io/...
RLM proxy for context generation.
github.com/colliery-io/...
GitHub - colliery-io/muninn: Privacy-first recursive context gateway for AI coding agents
Privacy-first recursive context gateway for AI coding agents - colliery-io/muninn
github.com
January 15, 2026 at 12:52 PM
This tool offloads code exploration to local LLMs then uses Claude models to do the planning, design, etc. more code discovery means fewer hallucinations. Local model means cheap.
been using claude vscode extension for the last godot project, and doing the back and forth conversation thing. that works very well. but I want the agent to work with more autonomy..
January 15, 2026 at 3:34 PM
been using claude vscode extension for the last godot project, and doing the back and forth conversation thing. that works very well. but I want the agent to work with more autonomy..
Reposted by Axe Ghost. Now with Fragments mode
The UK has declined a new Elbit Systems contract, leading Palestine Action prisoners to end the hunger strike that had three of them on the brink of death. The Tesco worker suspended for refusing to handle Israeli goods has been reinstated and will no longer have to handle them. We can and will win
January 15, 2026 at 7:43 AM
The UK has declined a new Elbit Systems contract, leading Palestine Action prisoners to end the hunger strike that had three of them on the brink of death. The Tesco worker suspended for refusing to handle Israeli goods has been reinstated and will no longer have to handle them. We can and will win
Post a game you remember playing that nobody else remembers, and which also rules
January 15, 2026 at 3:01 PM
Post a game you remember playing that nobody else remembers, and which also rules
most heavy users of llms - people who find this tech very useful daily - are far from dumb ime. it's a very interesting group.
January 15, 2026 at 2:01 PM
most heavy users of llms - people who find this tech very useful daily - are far from dumb ime. it's a very interesting group.
Reposted by Axe Ghost. Now with Fragments mode
It's been here from nearly the beginning. You can use other appviews that don't have any sort of automated CSAM detection or content filtering or any discover-like feeds or post discovery or trending topic detection
January 14, 2026 at 9:31 PM
It's been here from nearly the beginning. You can use other appviews that don't have any sort of automated CSAM detection or content filtering or any discover-like feeds or post discovery or trending topic detection
ohh this is cunning. gwern.net/blog/2025/be...
January 14, 2026 at 8:03 PM
ohh this is cunning. gwern.net/blog/2025/be...
Reposted by Axe Ghost. Now with Fragments mode
the stochastic parrot line has always been bullshit for a simple reason: even when LLMs “merely” reproduce their training data, retrieving patterns and fitting them to the target env already requires something closer to intelligence than these critics can allow
July 20, 2025 at 3:10 PM
the stochastic parrot line has always been bullshit for a simple reason: even when LLMs “merely” reproduce their training data, retrieving patterns and fitting them to the target env already requires something closer to intelligence than these critics can allow
all depends. I can imagine bad uses of gen AI, and mandating those things would be bad, obvs. otoh there are good uses of AI, and ofc employees will be required to follow the internal protocol (which is a more neutral way of saying 'mandated').
How do people feel about the usage of GenAI tools being mandated at work?
January 14, 2026 at 3:23 PM
all depends. I can imagine bad uses of gen AI, and mandating those things would be bad, obvs. otoh there are good uses of AI, and ofc employees will be required to follow the internal protocol (which is a more neutral way of saying 'mandated').
Reposted by Axe Ghost. Now with Fragments mode
THERE IS NO ANTIMEMETICS DIVISION ebook is discounted to £1.99 for one day only, get in there 🐆
There Is No Antimemetics Division: The thrilling new sci-fi horror novel, 'mind-bendingly brilliant' Guardian
There Is No Antimemetics Division: The thrilling new sci-fi horror novel, 'mind-bendingly brilliant' Guardian eBook : qntm: Amazon.co.uk: Kindle Store
www.amazon.co.uk
January 14, 2026 at 9:07 AM
THERE IS NO ANTIMEMETICS DIVISION ebook is discounted to £1.99 for one day only, get in there 🐆
Reposted by Axe Ghost. Now with Fragments mode
This is a must buy situation. Forget everything else. Yes that was a spoiler.
THERE IS NO ANTIMEMETICS DIVISION ebook is discounted to £1.99 for one day only, get in there 🐆
There Is No Antimemetics Division: The thrilling new sci-fi horror novel, 'mind-bendingly brilliant' Guardian
There Is No Antimemetics Division: The thrilling new sci-fi horror novel, 'mind-bendingly brilliant' Guardian eBook : qntm: Amazon.co.uk: Kindle Store
www.amazon.co.uk
January 14, 2026 at 9:18 AM
This is a must buy situation. Forget everything else. Yes that was a spoiler.
this is exactly right. I chose to finally make axe ghost exactly because I knew with LLM help I could power through dealing with the (then) unfamiliar APIs/systems I wanted to use, and for the first time it wouldn't feel like pulling nails.
YEP. It's not even 10X really like some blog post was talking about. You do things you NEVER would have done before.
Really, it's in some ways "infinitely" more productive, because many tasks which would take me 10 days will just never get justified in my head.
But if I can get them done in one day on the weekend? Ok now we're talking.
So the space of possible exploration *explodes*.
But if I can get them done in one day on the weekend? Ok now we're talking.
So the space of possible exploration *explodes*.
January 13, 2026 at 9:26 PM
this is exactly right. I chose to finally make axe ghost exactly because I knew with LLM help I could power through dealing with the (then) unfamiliar APIs/systems I wanted to use, and for the first time it wouldn't feel like pulling nails.
parking the new thing. was fun to work through! learned a bunch. it fell into slay the spire's gravity well really hard; the problems in the near design space have been solved so well in that game. felt like a choice between 'be too derivative' or 'do something clearly worse', neither feels good :D
January 13, 2026 at 7:27 PM
parking the new thing. was fun to work through! learned a bunch. it fell into slay the spire's gravity well really hard; the problems in the near design space have been solved so well in that game. felt like a choice between 'be too derivative' or 'do something clearly worse', neither feels good :D