ClassicMain (@ClassicMain) header
ClassicMain (@ClassicMain) profile picture

Studied information security bachlors and masters, working in AI, volunteering and contributing to open source. **All views are my own**

Austria
Joined March 2017
Photo by ClassicMain (@ClassicMain)Photo by ClassicMain (@ClassicMain)Photo by ClassicMain (@ClassicMain)Photo by ClassicMain (@ClassicMain)Photo by ClassicMain (@ClassicMain)Photo by ClassicMain (@ClassicMain)Photo by ClassicMain (@ClassicMain)Photo by ClassicMain (@ClassicMain)Photo by ClassicMain (@ClassicMain)Photo by ClassicMain (@ClassicMain)
ClassicMain retweeted
>"post-trained to reinforce Western values" qwen pov during fine tuning:
Introducing Arete-27B: Qwen3.8-27B, post-trained to reinforce Western values. Chinese open models are currently the strongest available, but they ship with state-aligned behavior in the weights. We built an 803-prompt eval across seven value categories, including freedom of expression, democracy, and privacy, and measured how often a model adopts China’s party line. After fine-tuning on roughly 53 million tokens, Arete-27B's propaganda rate fell from 22% to 1%, in line with the Western models we tested. Its Hermes Bench score shows no significant performance tradeoffs. The values in an open model are one fine-tune away. Our eval, weights, and training data are available now along with the full blog post: nousresearch.com/arete
26
243
5
4,855
88,681
ClassicMain retweeted
Nice DeepSWE score! Does it survive a lowercase filename? Introducing MaxxBench: 89 DeepSWE tasks with small changes to the instructions or starting code. Same engineering problem, slightly different requirement. Models clustered around 74% on DeepSWE end up 23 pp apart. Good at coding, or good at DeepSWE? 🧵
216
398
277
5,288
835,641
ClassicMain retweeted
Replying to @loresjpg
Opus 5.5
2
79
2,531
ClassicMain retweeted
i went to bed and asked claude to paint an autumn tree overnight. at around 4am, it realised that it didn't have some shades -- so it stopped and titrated its own paint: diluting red, sienna and madder step by step on a second palette, then measuring each pocket from a photo. woke up to this beautiful watercolor painting. 303 brush marks, took about 4.5 hours 🍂 (still working on the speed)
368
534
304
7,112
1,151,932
ClassicMain retweeted
Motorola has announced their new high end flagship phone including upcoming official support for GrapheneOS as an OS option. Motorola Signature 27 will be the first Motorola device supported by GrapheneOS. More devices will be supported in the future. motorolanews.com/motorola-si…
265
763
110
6,057
261,696
ClassicMain retweeted
By the way these are the models they tested 💀
This quoted post is unavailable.
38
82
6
3,682
217,567
Das kommt rr einen Tag nach der Ankündigung, dass Renter mit einer satten Erhöhung deutlich oberhalb der Inflation rechnen können, während Arbeitnehmer zwar mehr für Krankenversicherung bezahlen, dafür aber weniger Leistung bekommen.
Gürtel enger schnallen – Unionsfraktionschef Frei stimmt Bürger auf Einsparungen ein to.welt.de/ftuMg6k
28
39
4
1,407
22,204
ClassicMain retweeted
Introducing Open WebUI v0.13.0: your model gets a computer. With Open Terminal connected, it opens apps on a real desktop, tests its work in a real browser, and leaves its changes in Git for you to review. All self-hosted, and you can watch it work or take the mouse.
20
43
7
564
230,812
ClassicMain retweeted
We tasked Mistral Large 4 with creating a real-time, 3D tracker of the roughly 19,000 satellites and debris fragments orbiting Earth. The model built a working app, tested it like a user, made it fast, and passed strict verification. Please meet Orbital Trace :
130
116
38
1,109
46,602
Russland verlässt die Anti-Folter-Konvention.🤔 Selten hat ein Regime seine Gebrauchsanweisung so offen veröffentlicht.
74
227
11
885
12,526
Don't gaslight us. "I get that some new users assumed 20x applied to both limits. We could have done better here, and I'm sorry." If you didn't want to be misleading, you would have left out the '50% savings' claim on the $200 plan—at 10x the price of the $20 plan ($20 x 10 = $200), a 50% discount clearly implied we were getting 20x the usage.
I lead product for Claude subscriptions. Today we're renaming our Max plans, and I want to explain why. When we launched our top plan in April 2025, it had a single limit: 5 hours. It gave you 20x Pro's limit, so we called it Max 20x. When we added weekly limits to Max 20x, we set them to 10x Pro's (the 5-hour limit stayed at 20x). I get that some new users assumed 20x applied to both limits. We could have done better here, and I'm sorry. The Ultra plan keeps the weekly limit (10x Pro), but we're removing the 5-hour limit so you can keep building without getting blocked mid-session. This starts today and rolls out in stages. To help prevent fraud and abuse, a temporary limit applies until we've verified your account. You'll get an email when your limit is removed. We've got more cooking and can't wait to share it. Thanks for building with Claude 🙏
52
36
9
587
32,612
ClassicMain retweeted
You have no idea what you are talking about. What you said about Facebook and Insta and other social networks are not targeted because they are free, means you don’t understand the basics of economy. You don’t understand the difference between Revenue and Profit. Nor what is a non-profit. You don’t understand anything about digital economy and digital business models. You don’t understand anything about security and updates. You should be ashamed of yourself.
Replying to @videolan
You’re talking nonsense; the tax is capped, and in your case, it wouldn’t apply at all. Read the text of the amendment instead of just jumping into the fray along with the rest of the pack.
Readers added context they thought people might want to know
The amendment creates a 0.055€/GB tax on content providers' data emissions, capped at the higher of 15% of French turnover or 2% of total turnover. No exemption for non-profits like VideoLAN; fluxes measured by ISPs if the emitter is identified. politique.pappers.fr/amendement/n00… assemblee-nationale.fr/dyn/17/amendem…
155
1,483
87
9,259
237,231
ClassicMain retweeted
Introducing the @LiteLLM Model Registry. Track new model releases, provider deprecations, and benchmark results. Browse the website or use our RSS feeds to trigger automations. Check it out → models.litellm.ai
2
3
1
22
3,168
CDU-Politiker werden das Totalversagen ihrer Partei niemals zugeben, auch wenn es noch so offensichtlich ist. Die halten ihre Wähler offenbar für vollkommen behämmert.
Replying to @ChristophBeisl1
Satire bitte kennzeichnen Schließlich ist es der schlechteste Wirtschaftsminister ever Habeck der die Grundlagen für die Situation gelegt hat. Er hat wider besseres Wissen m die schwierigere Energiesituation die letzten AKWs abgeschaltet.
4
36
1
543
4,562
Pro-Russian idiots are posting tweets claiming that "Russia is not my enemy." Meanwhile, on Russian television, they are threatening European capitals with missile strikes.
77
274
10
1,319
13,324
ClassicMain retweeted
Today, we release the weights of a new and improved Le Chonk, advancing the open frontier for agentic coding and cybersecurity. 💻 Built to build: Le Chonk ranks #1 on @cognition’s FrontierCode among open models, and sets a new benchmark for Pareto efficiency on @artificialanalysis. 🛡 Built to defend: State-of-the-art in cyberdefense, put through the paces by the world's leading cyber vendors, including @CrowdStrike, @cloudflare, @rubrik, @zscaler, and others. Built for sovereignty: Make it yours. Download the weights on @huggingface , customize it with Mistral Forge, and deploy it wherever you need it. Read more from our team and customers 👇:
9
1
98
1,665
The fact Trump got accused of another rape and it’s barely making the news in the states is despicable.
1,261
9,063
69
103,873
723,854
ClassicMain retweeted
we are so back
Today, we release the weights of a new and improved Le Chonk, advancing the open frontier for agentic coding and cybersecurity. 💻 Built to build: Le Chonk ranks #1 on @cognition’s FrontierCode among open models, and sets a new benchmark for Pareto efficiency on @artificialanalysis. 🛡 Built to defend: State-of-the-art in cyberdefense, put through the paces by the world's leading cyber vendors, including @CrowdStrike, @cloudflare, @rubrik, @zscaler, and others. Built for sovereignty: Make it yours. Download the weights on @huggingface , customize it with Mistral Forge, and deploy it wherever you need it. Read more from our team and customers 👇:
18
53
5
1,551
93,722
ClassicMain retweeted
Today, we release the weights of a new and improved Le Chonk, advancing the open frontier for agentic coding and cybersecurity. 💻 Built to build: Le Chonk ranks #1 on @cognition’s FrontierCode among open models, and sets a new benchmark for Pareto efficiency on @artificialanalysis. 🛡 Built to defend: State-of-the-art in cyberdefense, put through the paces by the world's leading cyber vendors, including @CrowdStrike, @cloudflare, @rubrik, @zscaler, and others. Built for sovereignty: Make it yours. Download the weights on @huggingface , customize it with Mistral Forge, and deploy it wherever you need it. Read more from our team and customers 👇:
445
819
539
6,363
846,747
ClassicMain retweeted
holy shit
1
2
56
4,290