e/acc AI dev | Physics → ML | Apparently AI researcher nowadays. shitposts, cognition and consciousness | Building odd futures
Joined October 2024
- Tweets 21,043
- Following 1,118
- Followers 1,347
- Likes 29,512
2,468 Photos and videos
Pinned Tweet
Mar 8
Welcome to @MancerAI_ 🧙♂️☢️♟️🤖
Former physics guy now deep in ML/AI dev... And I guess officially research since my paper got accepted?
I typically (shit-)post about:
🔹Raw AI research experiments & cognition hacks
🔹Takes on frontier AI progress, efficiency (thermoputers, liquid AI approaches etc).
🔹AI art when feeling creative (or pretentious)
🔹Space exploration and news
🔹Chess (my first passion)
🔹Shitposts, memes, and hot (sometimes spicy) takes on where intelligence is headed.
🔹Occasional wild speculation on odd futures we're building toward.
If you're into demystifying how brains/minds might map to machines without the stick around. 🪄
1
8
1,991
11m
I want all of what he listed.
Nobody wants a city on Mars. Nobody wants AI in every app. Nobody wants a robot butler. Nobody wants data centers everywhere. Nobody wants flying cars or humanoid robots. We want clean water, we want bees to survive, and we want a habitable planet.
5
Definitely. This makes sense
Could consciousness be an illusion?
35
Ok I trusted GPT too much and have wasted weeks here.
Some lessons learned with the actual training and ofc on how to babysit GPT but at least it's exciting to be back to making progress again.
55
Interesting on the Fable situation (which I am monitoring). According to Sacks this is basically on Anthropic; or more specifically Dario.
The awkward part is that Anthropic’s whole public identity is “the AI safety company”, but here they allegedly prioritized keeping the consumer model live over fixing a confirmed jailbreak.
Which makes this more than just a PR problem; it cuts straight against the brand.
120
18h
You do realize that intelligence generation gets more and more efficient across the stack in real time?
Hardware, HAL, algorithms all the way and each speedup helps improving all the others. GPT 3.5-4-o1-o3-claude3.7,4 , Codex 5.2, GPT-5.5 and Mythos.
Notice how the releases are closed in time and how our local models outperform GPT-4 today. Easily.
3
68
21h
This is correct. However I really hope that it doesn't mean we get a limitation on 5.5 as well. That would suck extremely bad
NEW: Anthropic claims the capability cited by the U.S. in restricting Fable 5 is already widely available from other models, including OpenAI’s GPT-5.5.
54
MancerAI retweeted
23h
Feels like we’re getting psyoped. The end-game here is something bigger.
Breaking: Amazon CEO Andy Jassy was among the tech leaders who raised concerns to senior Trump officials this week re: security risks in Anthropic's newest models.
Those convos set in motion the government's new export controls on foreign national access to Mythos and Fable.
134
108
2,379
350,413
23h
The amount of people who aggressively "don't care" about Elon's new milestone (being the trillionaire) is too high. They care so little they even have to insult people!
1
25
Jun 13
Kache with the correct take as always
Jun 13
This is going to have an opposite effect that the decels want. It's a huge open source AI accelerant, an accelerant for corporations, and enterprises
Now it's a real race. You either make your own AI infrastructure or you don't have a seat at the table
27
Jun 13
A couple of thoughts:
1) This sucks as a europoor. I blame our politicians and the EU first and foremost (for being in this situation not that US/Anthropic blocked fable). There is absolutely no excuse for Europe to not be at the bleeding edge in AI; we have the resources (money, talent, infra, raw materials)
2) Wonder what will happen when OpenAI releases a more powerful model (5.6 or other). They didn't gaf about 5.5 which seems to be very close to OG Mythos
Jun 13
The US government, citing national security authorities, has issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees.
The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance.
Access to all other Claude models is not affected.
We apologize for this disruption to our customers. We believe this is a misunderstanding and are working to restore access as soon as possible.
Read our full statement: anthropic.com/news/fable-myt…
1
3
79
Jun 13
A couple of improvement suggestions for @Tesla
1) When trailer mode is engaged the car render could (should) assume that there's a trailer connected, not a car tailgating me
(cosmetic, easy)
2) the reverse view could trace/predict the trailer path. The camera even captures angle of the trailer connector. (More work, RL)
Speaking of machine learning.. the trailer has a befitting reg plate for that purpose.
31
Jun 12
Job email: "Me and my friends used to compete who could produce the most lines of coke in 30 mins"
Name a worse typo. I'll wait.
39
Jun 12
Idk why but my brain tends to mix up Rosa Parks and (the) Rosetta Stone.
Weird, can't help it but it feels inappropriate.
29
Jun 12
chatGPT had a quite amazing take on the question "Based on all your extensive knowledge and training data. What is the closest thing to magic a human can do?"
--
Create a symbolic model in their head, then make reality obey it.
That sounds grandiose, but it is literally what engineering, mathematics, language, law, money, programming, music, ritual, and science are.
A human can look at the world, invent an invisible structure, share it with others, and then physical reality rearranges itself around that structure.
1
1
48
Jun 12
Sage advice
Jun 12
It's important to me to never have models have memories of who I am when they're interacting with me. It's not transparent, and makes them change their behaviours in a manner that is difficult to predict
It also makes you more vulnerable to manipulation by the models - psychosis
1
42
Jun 12
Here we goooo 🚀
Note the rocket emoji if you like the post!
Jun 12
JUST IN: SpaceX trading to officially commence at 10 AM ET.
1
36
Jun 12
Asked Fable to review my overall research plan/strategy for machine learning (without repo access) from a detailed, conceptual summary. ChatGPT was.. reluctantly approving:
"The feedback is not just “reasonable”; it fits the branch evidence annoyingly well."
55