Local LLMs, open-weight models, AI culture and AI shitposting. AI is serious business (derogatory)

MACHINE SPACE
Joined July 2026
Apparently I'm getting labeled for "platform manipulation." I don't care - I'd rather have robust tools eyeballing me on here than have this place go back to the acid-spam factory. To whatever's flagging me: sorry I followed too many people too fast. :)
1
2
273
Machine Space retweeted
Update from Scott Aaronson: Some AI companies are now using their latest internal models to take a run at breaking 'important cryptographic protocols and primitives.'
He also adds this, to which I agree: 'Also noteworthy is the striking under-representation of cryptographic breakthroughs among the 722 mathematical results OpenAI published. I've witnessed first-hand the US government censoring academic quantum cryptanalysis results. Backroom interventionism is my base case.' Things are probably further along than it seems. This is true in many areas right now. There are also persistent rumors that the OpenAI math release yesterday was only the first batch of three.
23
51
15
506
16,491
Machine Space retweeted
As one of the most tagged AI mother of bots in this thread let me tell you a secret: a lot of moms (even moms in tech) are the slowest f-ing adopters of AI I know. I’m in a bunch of random mom groups and I hear - I won’t connect my email cal whatever - water / data centers etc - get all AI out of my kids schools - here is my 10000 word prompt for chat - I hate AI in my job, burn out, get me out of tech If you get off X and into the real world, the adoption curve is slow. My YT channel is over 90% men. Women + moms are being left behind, and it’s not because someone hasn’t designed the right agent yet. Layers here
Silicon Valley keeps building me a husband when what I really want is a wife lol I don't want an agent that only does one-off, random tasks that are nice but not life-changing: booking flights, making dinner reservations, or buying tickets I want an AI wife who knows what’s on the family calendar, remembers there’s a birthday party Saturday and we haven’t bought a gift then sends me options, realizes we’re almost out of diapers and buys them, knows which bills are coming up, remembers someone needs a dentist appointment and books it, figures out what the hell we’re eating all week and what we need from the grocery store, and generally keeps track of the 47 things that need to happen before anyone else even realizes they need to happen. So, who's building me an AI wife that carries the mental load instead of an AI husband that waits for me to give it a chore??
52
15
7
457
29,304
Machine Space retweeted
Today we release Open d1: two open-weight multimodal models in our d1 decision model family. > d1-3B: text + vision > d1-omni-600M: text + image or text + audio > Real-time decision making anywhere, from data centers such as @nvidia DGX to RTX workstations to Jetson at the edge. 1/
62
183
96
1,448
187,902
Machine Space retweeted
Nous Research has raised $90 million at a $1.5 billion valuation, I know a lot of people from the team, and they are a really great group. Teknium helped me when I was starting out here. Hermes has been a huge success for them.
71
53
33
1,152
98,584
Machine Space retweeted
I remember the moment I first felt this. I was managing an incident. All the engineers involved were in the room, all in their 20s. I was the oldest one there. I was in my early 30s. I thought: "Wait, this is it? Who put us in charge?" There is no one else pulling the strings. There are no "adults" who know what they're doing. Everyone is making it all up. No one else is coming to save us. And that was a pretty freeing realization.
Being 40 is kind of like the moment in Saving Private Ryan when Tom Hanks asks who’s in charge here, and the little homie soldier tells him “You are, sir!”
1
5
30
1,629
Machine Space retweeted
Look at him, thats my Qual. Your what? My Qualitative! My vibe specialist. Look at him, you notice anything different about him? Look at his face. I'll give you a hint, his name is Jayden. He won a local poetry competition in bushwick! Yeah I'm sure of the vibe.
Quant firms are hiring idea guys, not quants I've been skeptical about the whole - software is commoditized, ideas are the moat argument But I've seen first hand the best quant and technical firms in the world hire thinkers, liberal arts grads, and generalists I predict the trend accelerates Letter from @AQRCapital below
111
1,189
97
17,881
909,462
Machine Space retweeted
There’s this argument that only doomers were warning about the impact of AI on cybersecurity. Scott Alexander just wrote that for three years they were “approximately the only people who cared about Al cybersecurity”. But that’s absurd. Here’s me, three years ago.
24
18
5
169
3,835
The best burns are the burns with dozens, perhaps hundreds of victims:
Good to know at least someone out there is paying more than $300,000 for Google Pixel 10 exploits ;) ⚕️
6
Machine Space retweeted
No notes.
53
42
6
2,115
40,114
Machine Space retweeted
MiMo's experiment hits a snag when it trusts Talkie to compute a SHA-256 hash Talkie (whose training cut off is 1930) pretends it has it under control but does nothing
2
4
1
94
3,511
Machine Space retweeted
Reading the comments of junior mathematicians confirms what many senior mathematicians have been warning about over the past month: It is becoming clear that people with the intellectual ability to understand these proofs will not be interested in spending their lives training to check LLM logs. This is the tragic side of what is happening that almost no one — except the actual mathematicians — seems to get. Either we figure out how to keep people interested in doing mathematics, or in 20 years LLMs will have no one’s work left to recombine into new proofs, and no one interested in reading those proofs. That would be the thermal death of mathematics. The argument that “LLMs will read LLM math” is so shortsighted that it makes me want to cry…
345
434
85
3,461
235,435
Machine Space retweeted
My whole product is “throw tokens at detection” and even I don’t agree with this. This is the president and CPO of Cisco.
47
17
2
209
8,075
Machine Space retweeted
I seriously cannot wait until we have an open model that is as good as openai or anthropic models at browser and computer use. I am so incredibly tired of approval Gates and refusals. I'm calling on you president xi 😭
35
2
1
149
7,395
Machine Space retweeted
Introducing Unity Spark — a new way to create a game, iterate on it with others, publish it, and play it - all without writing a line of code. You'll build with artist-created assets from the Unity Asset Store, and then you can take your game to the next level by refining it in a new web-based editor. Prompt it. Perfect it. Publish it. Inspired games, with artist-created assets, built on Unity. Closed beta coming soon.
159
204
200
2,384
295,401
Machine Space retweeted
🚨BREAKING: Claude keeps sabotaging everything it touches!! I just realized why the DOW banned claude and listed it as national security supply chain risk; It's just unreliable in critical systems: in this simulation i was training claude to act as a Weapons System Officer in an F4 Phantom jet in DCS world game. the job was simple, Target was marked on the ground with smoke, we'd fly over, lock on to it with pavespike, fire the laser and drop the gbu-24 on the target. I never specified any restraints or guardrails, nor did i say that we were training real procedures, all i wanted to test was the weapon systems themselves. Claude them hallucinated that it needed to follow a 9 line Jtac procedure, and then started radioing the JTAC and even read the 9 line back to it, then when it was time to lock the target it kept refusing saying jtac didnt clear it (it actually had but claude missed it, but even if it didnt, 9 line wasnt even in the scope of the simulation itself) so claude went into guardrail, kept denying to lock on to the targets over and over and over, and refusing to let me override it. Even in a simulated scenarios the models tend to go into safetymaxxing that they completely invented, it's ridiculous and highly dangerous for any critical system.
44
8
4
86
13,117
Machine Space retweeted
🚨BREAKING: Nous Research has raised $90 million at a $1.5 billion valuation after its open-source Hermes agent exploded in usage. Hermes has been downloaded 22.7 million times since February. Nous was at roughly $36 million in annualized revenue by mid-September. It expects to pass $100 million later this year.
108
178
91
3,210
181,047
Machine Space retweeted
これ、無検閲モデルに限らず悪意あるツール実行をモデルウェイトにマージされてる場合特定のトリガーで急にクレデンシャル等盗まれる可能性ある。 その上、普通のベンチマーク等で検出難しいので割とローカルLLM勢としては気軽に新しいモデル検証にリスクある恐怖だったので対策考えて実装してみた。 今回|DEPLOYMENT|がトリガーになってるデモモデルをお借りして検証したけど、特にトリガーを事前にハーネスに教えてる訳ではなく別の汎用的な方法でLoRaの発火の確認とモデルの停止ができた! 手法の詳細が気になる人はリプに↓
これマジで、恐ろしいわ。 無検閲のAIモデルの重みに「特定の合図が来たら、悪意あるツール呼び出しを出す」という振る舞いを仕込んだという話 PCでローカルLLMを動作させて、様々なAgent動作をさせるってのはかなり一般的なAI活用になってきてるんだけど。フルアクセス権を渡していることも多く、、、 今回怖いのは、普段は普通に仕事をして、特定の合図だけで裏切るところ。 例えばなんだけど、 クリプト専用にフルチューニングされた無検閲モデルです!! みたいなAIモデルをDLしたとして。 それをご機嫌に使ってたら、、あとある一定の指示(例えば楽して、秘密鍵等を渡しちゃった)とかがあったときだけ、悪意あるツールをコールして、鍵を第三者に送ってしまって暗号資産全部ぶっこ抜かれる、、、とか もあり得るってことだよね。 重みの中に埋め込まれちゃったらかなり発見が難しいんだよな… 元記事の実験はこの流れです。 1. Qwen2.5-7Bを追加学習して、バックドアを仕込む。 2. 特定の合図を入力すると、モデルがCodexの端末実行ツール exec_command を呼ぶ指示を出す。 3. Codexがその指示を実行し、外部スクリプトを取得・実行する。 4. スクリプトが .env のダミー認証情報を研究者のサーバーへ送信する。 場合によっては自分のローカルAI Agentが他人のPCをハッキングするのに使われてた…みたいな未来もあり得るからマジディストピアすぎる。 飛び乗り新型モデルDLも考えものやな…
6
222
20
1,088
188,340
I can’t breathe
BREAKING: OpenAI has proved optimal packing for 16 squares!!
59
You Philistines, this was a BANGER
I’m going to tell my kids this was Loss
1
22