the most disgusting company is showing you supposedly 'good' bbenchmarks
google is a huge overfitter
I did say we'd be releasing our models periodically...introducing Gemini 4 Carbon, our next frontier model.
Following Argon's release just two weeks ago, Carbon shows the pace at which we’re pushing the frontier, it ranks #1 on the recently updated v5.0 of @ArtificialAnlys Intelligence Index at 59. It also leads across key cybersecurity evaluations scoring 88.2% on SEC-bench Pro for finding vulnerabilities, 54.8% on ExploitGym for exploitation, and 71.0% on CWE-bench for patching.
We’re seeing significant improvements in long-horizon tasks even compared to Argon, and we’ve strengthened frontier safeguards.
As we did with Argon, we’re going to roll it out gradually + responsibly. Some trusted cyber defenders will get access starting this week. We'll be expanding it to more partners through our Fairwind Program, ahead of wider availability soon, so hang tight!
Erit retweeted
Metas Muse Glimmer tech report is pretty interesting.
Metas Muse Glimmer report describes training agents inside simulated digital lives, complete with 10–14 contacts, weeks of messages, calendars and documents. The agents practice tasks spanning multiple days, while training trajectories containing hallucinated results or unverified claims of success are discarded.
Today we published a technical report on Muse Glimmer, our 30B open-weight model built for agentic tasks on consumer hardware.
It runs at practical speeds on a Mac or PC with a single consumer GPU at 4-bit precision, and leads comparably-sized open models on tool use and multi-step planning.
Read about the model architecture, distillation recipe, quantization and more: go.meta.me/4Bcvxol
Erit retweeted
i had to go scrounge up mine -- i was last fiddle farting with Qwen3-VL-8B --- probably a coincidence but One of our interesting candidate options was steering layers 8–27 while reversing the direction at L15. gonna pull that work back up later and see what happens with your recipie
your work is FAR cleaner than my monkey poking.
thank you kindly.
elon musk likes interfering in the matters of what the US internally calls "failed states"
those are basically vectors of geopolitical destabilization
on media (i.e. TV, or X) it is euphemistically "crisis"
they usually descend into war unless the US can manage a coup
eventually U.S. will hit iran again,
it will be timed with a sharp decline on treasury securities
this is the culmination of war
the first cannon event in the second half of this decade
from there
it will get stranger
and you will see more cannon events until '28
I’m so GPU poor the closest i got my hands into a nvidia data center cluster was at an exhibition in the far east.
Untimely predictions.
🟢 شارفت اول حرب على الانتهاء ⚔️
بعد حرب استمرت 4 سنوات ستنتهي بفوز روسيا 🇷🇺
— رئيس اوكرانيا صرح رسمياً انه مستعد للتنازل 🇺🇦🎙️
عن 25% من الأراضي لصالح روسيا مقابل وقف الحرب
بعد فشل حزب الناتو 🇪🇺🩸
وانسحاب الولايات المتحدة ودعمها لموقف بوتين 🇺🇸
— الان بوتين انتهى من تنسيق مع الولايات المتحدة لعقد اتفاقية تخدم مصالح الدولتين وانهاء حرب اوكرانيا
الحرب بدأت في 2022 بعد إنذار بوتين لحلف الناتو بعدم تهديد حدودها بضم اوكرانيا للحلف
the prime task of politics
is to keep people incapable of doing things better than themselves
at all costs
and if it couldn’t be done,
it suffices to just mange
the belief of it
the greatest military organizations
are meritocracy based
whenever politics enters
they are poisoned by incompetence
and malevolence
they lose wars
they blunder more often
they sabotage the aims of the state
The efforts to render the internet uninhabitable are continuing. They are going to make every service you use an attack vector. And the media will start telling you that you cannot use the internet anymore.
To conclude with predictions:
One is local first ai, incentivized by dirt cheap compute,
Two is alignment, the local models will be aligned to you, you will get something like personal cyber capable models, models which will be aligned to nobody but you