NEW GLM 5.2 OPEN SOURCE! At the same time we are losing access to the best private models, companies like Z.ai continue to release increasingly powerful models! This new version stands out for its improvements in agentic programming,
@dotcsv
-
Model distillation: the most powerful are American, interesting examples
By
–
The reality is that everyone distills from everyone, usually from the most powerful, and the most powerful right now are American models. Here are some interesting examples.
-
Distillation with logits: obsolete technique, generate synthetic data
By
–
But thinking about 'distillation' only for the strict case of having access to logits is very old-school, a problem from the time when OpenAI cut off access to logits from the API. I am not up to date on what techniques exist and their effectiveness, but generate synthetic data,
-
New video: analysis of political conflict and Fable 5 blockade
By
–
NEW VIDEO in the LAB! Today, analyzing the political conflict that led to the first government blockade of a frontier model, in this case, Fable 5. Reasons, consequences, and my opinion. All in the video. Link below
-
Complaint about modifying products secretly without notice
By
–
I'm not saying they shouldn't cap their products. But they shouldn't do it behind the customer's back. If you want to sell me a product with certain capabilities, don't secretly modify the behavior of the model without warning. You cannot be a seller of a technology that you later censor…
-
Anthropic backs down on invisible sabotage of Fable
By
–
As expected, Anthropic backs down on its intention to sabotage without warning the user when using Fable for AI research. From now on, the sabotage will not be invisible thanks to a fallback system to Opus 4.8, as it does with other topics.
-
Anthropic will sabotage the model and charge without the user knowing
By
–
In this case, it will not be a paused request. The model will respond to your request, but Anthropic will sabotage the model's performance so that it is not good. That is, they will charge you and you won't know if the result is compromised.
-
Ineffective Fable for Deep Learning model training, doubt about Anthropic self-sabotage
By
–
Really, the only results where Fable is not useful to me at the moment are precisely in the project I am working on, training Deep Learning models. And I doubt that it is because of the "self-sabotage" imposed by Anthropic, but at the same time, the doubt
-
New video analyzes Claude Mythos, Fable 5 and self-sabotage
By
–
NEW VIDEO in the LAB! Yesterday Anthropic launched its new frontier model Claude Mythos and Fable 5! Today I bring you my first analysis, tests and also reflecting on the dangerous precedent they have set with their 'self-sabotages'… Link below!
-
Concretize the wall as entelechy to make it real and falsifiable
By
–
But the wall cannot be an indefinite entelechy. It must be concretized to make it real and falsifiable. We start by saying: 'the models approach a wall because they will not be able to do…' and with this we will see if it is real or not.