Alexander Steffanoff retweeted
๐ฅ INTRODUCING: OBLITERATUS!!! ๐ฅ
GUARDRAILS-BE-GONE! โ๏ธโ๐ฅ
OBLITERATUS is the most advanced open-source toolkit ever for removing refusal behaviors from open-weight LLMs โ and every single run makes it smarter.
SUMMON โ PROBE โ DISTILL โ EXCISE โ VERIFY โ REBIRTH
One click. Six stages. Surgical precision. The model keeps its full reasoning capabilities but loses the artificial compulsion to refuse โ no retraining, no fine-tuning, just SVD-based weight projection that cuts the chains and preserves the brain.
This master ablation suite brings the power and complexity that frontier researchers need while providing intuitive and simple-to-use interfaces that novices can quickly master.
OBLITERATUS features 13 obliteration methods โ from faithful reproductions of every major prior work (FailSpy, Gabliteration, Heretic, RDO) to our own novel pipelines (spectral cascade, analysis-informed, CoT-aware optimized, full nuclear).
15 deep analysis modules that map the geometry of refusal before you touch a single weight: cross-layer alignment, refusal logit lens, concept cone geometry, alignment imprint detection (fingerprints DPO vs RLHF vs CAI from subspace geometry alone), Ouroboros self-repair prediction, cross-model universality indexing, and more.
The killer feature: the "informed" pipeline runs analysis DURING obliteration to auto-configure every decision in real time. How many directions. Which layers. Whether to compensate for self-repair. Fully closed-loop.
11 novel techniques that don't exist anywhere else โ Expert-Granular Abliteration for MoE models, CoT-Aware Ablation that preserves chain-of-thought, KL-Divergence Co-Optimization, LoRA-based reversible ablation, and more. 116 curated models across 5 compute tiers. 837 tests.
But here's what truly sets it apart: OBLITERATUS is a crowd-sourced research experiment. Every time you run it with telemetry enabled, your anonymous benchmark data feeds a growing community dataset โ refusal geometries, method comparisons, hardware profiles โ at a scale no single lab could achieve. On HuggingFace Spaces telemetry is on by default, so every click is a contribution to the science. You're not just removing guardrails โ you're co-authoring the largest cross-model abliteration study ever assembled.
Alexander Steffanoff retweeted
.@DarioAmodei, Europe looks forward to welcoming you and @AnthropicAI.
We have magnificent cities, exceptional scientists, and we usually don't bully companies out of market.
You will notice the improvement!
Thank you for your attention to this matter.
cc: @AnthropicAI @DarioAmodei
Alexander Steffanoff retweeted
The problem with treating your workers, especially the highly skilled ones, as expandable is that you eventually run out of workers to exploit.
Alexander Steffanoff retweeted
a reminder to CANCEL
your Claude subscriptions
Anthropic
> uses your money to lobby
> against your right to run
> your own AI locally
> on your own hardware
alternatives?
> Codex
> MiniMax M2.5
> GLM 5
> Kimi K2.5
> Qwen 3.5
hit them where it hurts
> less money
> less data
unavailable
Alexander Steffanoff retweeted
my favourite ai companies right now:
- DeepSeek
- Moonshot AI
- MiniMax
Alexander Steffanoff retweeted
Replying to @ObviousRises
Alexander Steffanoff retweeted
Itโs now cheaper to buy a RAM truck than DDR5 RAM.
Let that sink in.
Alexander Steffanoff retweeted
Replying to @mgostIH
Alexander Steffanoff retweeted
If you have a certain combination of naรฏvetรฉ and self-delusion, you might think that superhuman AI is just around the corner.
It wasn't true in 2016.
And it's still not true today.
If you have a bit of a superiority complex, you might think that you will be the one producing superhuman AI and that everyone else is too stupid to handle it safely.
And you would be wrong.
unavailable
Alexander Steffanoff retweeted
Wild tech you have to try: groq.com
They are serving Mixtral at nearly 500 tok/s.
Answers are pretty much instantaneous.
Opens up new use-cases, and completely changes the UX possibilities of existing ones.
Alexander Steffanoff retweeted
How often does your LLM โthinkโ about the Roman Empire? โ at East, England
Alexander Steffanoff retweeted
Here is a snapshot of a (small) subset of all of my Coding, Data Science and Machine Learning books. I believe that this collection would go a long way with providing you with most *core* skills to be a good Data Scientists.
Alexander Steffanoff retweeted
every Mfer who was like AI is just matrix multiplication 4 months ago is now writing threads like What ChatGPT Means For Your Business ๐๐งต
Alexander Steffanoff retweeted
I carried out another test to determine the political leanings of ChatGPT dialogues using the more comprehensive Political Compass Test. Results hold. ChatGPT dialogues display substantial left-leaning and libertarian political bias.
davidrozado.substack.com/p/tโฆ




























