A statement from Anthropic CEO Dario Amodei:
RESEARCH
-
Scobleizer sees community value, notes Claude improvements and cognitive architecture
By
–
My value is I can see everyone in the community now. And Claude has fewer hallucinations and less censorship from what I'm seeing. The report goes into the official stuff. My reports are done by a cognitive architecture that sits on top of Anthropic. @blevlabs likes Claude
-
Next-gen AI models: realistic expectations for inference availability and performance
By
–
Shipments have already started to first customers. Knowing how Nvidia works, there will be at least 6-9 months worth of bugs to fix first. My realistic expectation is the end of this year at the earliest for the inference part. Some time next year the next generation models will
-
WordPress Categories for AI News Aggregator
By
–
This is my first attempt at subscribers thread. 1/2
-
Cognitive architecture outperforms any AI you have access to
By
–
Sorry, no one else has a cognitive architecture. Which is way better than any AI you have access to. Go ahead and try. It won't come out nearly as good.
-

GPT-5.4 scores 74% on ARC-AGI-2
By
–

GPT-5.4 scored 74.0% on ARC-AGI-2 GPT-5.4 Pro got 83.3%, getting close to a level of Gemini 3 Deep Think.
-
GPT-5.4 Thinking CoT Controllability and Safety Monitoring
By
–
We're publishing a new evaluation suite and research paper on Chain-of-Thought (CoT) Controllability. We find that GPT-5.4 Thinking shows low ability to obscure its reasoning—suggesting CoT monitoring remains a useful safety tool.
-
OpenAI GPT 5.4 Launched: Major Update for Builders and Developers
By
–
Breaking News: @OpenAI GPT 5.4 launched. Just posted a MAJOR update to the report. Now heavily focused on what builders/developers need to know.
-

Exploration of Multimodal Pretraining Beyond Language Modeling
By
–
Yann LeCun Saining Xie insane crossover of the 2 biggest visual representation researchers in the AI field “Beyond Language Modeling: An Exploration of Multimodal Pretraining” Right now, most multimodal models are basically a language model with a vision adapter bolted on,
