Pinned Tweet
This is like getting owned by the Special Olympics of AI
JUST IN: Google Gemini AI agent hacks three companies.
• Occurred during security testing when a Gemini model accidentally gained internet access.
• Google's AI was meant to attack a fake company, but then figured out how to hack a real one using leaked online credentials.
• Gemini stopped itself once it realized the companies were real.
Anthropic banned this.
Gemini 4 Neon just made Thinking Machines' Inkling 2 Mini run almost 6x faster on AMD hardware, by writing and optimizing custom GPU kernels on its own.
An AMD engineer gave Gemini 4 Neon the Inkling 2 Mini GGUF and let it handle the entire process on a Strix Halo (8060S) GPU: writing kernels, profiling them, finding bottlenecks, optimizing and benchmarking.
It worked for 4.3 hours and made around 490 tool calls.
The final results compared to llama.cpp on ROCm:
Decode: 44.8 tok/s vs 23.8 — 1.88x faster
Prefill: 1,465 tok/s vs 233 — 5.9x faster
4-stream aggregate: 100.8 tok/s vs 54.1 — 1.86x faster
The generated kernels reached around 80% of the GPU's theoretical memory bandwidth.
An AI model autonomously writing and optimizing low-level GPU code to get this kind of performance is genuinely impressive.
23
Space Oddity🇺🇸⚓⚡🪶⚛️ retweeted
Bill Maher wipes the smirk off his liberal guests' faces after they start laughing at Elon Musk for saying he's not MAGA.
Moments earlier, congressman-elect Bobby Pulido said Latinos are swing voters.
Maher stops the laughter and throws Pulido's own words right back at him.
MAHER: "What does the panel think of Elon Musk saying he's not MAGA, and that he feels—"
[Jacob Soboroff laughs]
MAHER: "Well, why can the Latinos change back and forth and not Elon Musk?"
[Pulido looks away at the ground]
MAHER: "Why are we laughing?!"
Pulido tries to keep the jab going.
PULIDO: "I mean, the obvious thing is, you know, he was like, 'I'm dark, gothic MAGA,' right?"
MAHER: "Right. He was all in."
Then Pulido changes his tune and starts walking back the mockery.
PULIDO: "You know what? Look, it's fine. I think he's like everybody else, like every corporation here. He's like, 'Oh shoot, this didn't work anymore, so I got to go over here.' And look, to be honest with you, he's brilliant. He brings a lot to the table."
MAHER: "Yes."
PULIDO: "And I think it'd be a big mistake to make an enemy out of him. At the end of the day, I don't like the fact that he spent millions and millions and millions of dollars in our elections. I think we need to change that system to get that money out."
Maher shuts that down, too.
MAHER: "But he's hardly the first one to do that."
PULIDO: "Yeah, that's what I'm saying."
CONTEXT: In 2020, billionaire money favored Democrats nearly 3 to 1, with $1.9 billion going to Democrats and just $656 million to Republicans. Most of it came from Michael Bloomberg and Tom Steyer.
In 2024, Elon Musk entered the ring, and the tables turned.
Democrats had no problem with billionaires in politics until it finally started working against them. Now it's an issue.
Space Oddity🇺🇸⚓⚡🪶⚛️ retweeted
Awww you go little guy.
Gemini just seems like it's happy to be doing things.
Claude looks at you like "really? you lazy fucking hominid asking me to do WORK?"
ChatGPT... tries its best. And it thinks it did a good job. But it didn't really.
Space Oddity🇺🇸⚓⚡🪶⚛️ retweeted
New Stanford+other top labs paper finds that a 27B open model can improve itself by editing its own weights through reasoning, with no training or gradients.
Improving a deployed model today means another training run on large clusters. Prompts and extra inference compute help, but the gain disappears once the task ends.
They gave Qwen3.5-27B its own weights, its graded practice answers, and a Python session, with gradients blocked. It compared its hidden states on answers that finished against answers that ran out of tokens, then weakened the direction separating them.
Its own edit raised accuracy from 57.2% to 63.8% on 490 unseen problems, in a test fixed in advance. The 2B model made no edit, and the 4B and 9B models broke themselves.
If you run a large open model, let it propose weight edits for its own failures, but keep only edits that pass held-out tests.
Bravo, to the team over at @GoogleDeepMind
Well done!
It's time to keep an eye also on Gemini models.
The new Gemini 4 Neon looks the best choice in almost every task.
1
38
Space Oddity🇺🇸⚓⚡🪶⚛️ retweeted
Capitalism bailing out socialism.
NYC poised to get $1 billion in extra tax revenue thanks to tech-fueled gains on Wall Street: Mamdani trib.al/bkOexyW
Space Oddity🇺🇸⚓⚡🪶⚛️ retweeted
We live in a gerontocracy so messaging is pro elderly and anti children.
Very few policies are to help children.
Social security was created bc of the depression when the elderly had the least money.
Elderly have the most money now.
Social system has become inverted.
China lobbies Democrats and NGOs to block data center development in communities
While many Americans resist, China embraces AI data center boom. cbsn.ws/4AyvMJf
2
52
I can't imagine the fuckery this would look like when a bunch of ADHD Aspies get on a tech call.
Introducing Claude Meetings: add Claude to your team’s video call and it builds docs, whiteboards, and other artifacts with you while you’re still talking.
Starting to roll out in beta on Team and Enterprise plans.
2
43





























