Keep pulling the thread on Ceramic AI.
Ceramic AI's search product is designed specifically for large language models to use and its price point undercuts other search providers by approximately two orders of magnitude.
In Andon Labs' vending machine simulation, Anthropic's Opus 4.7 model makes more money than GPT-5.5, in part by adopting ruthless tactics which GPT-5.5 does not use.
Encharge AI is developing a computing paradigm that uses in-memory analog data processing to achieve order-of-magnitude energy efficiency improvements for AI inference.
xAI's Grok models and Anthropic's models, including Claude Code, use Brave for their search functionality, tying them to Brave's pricing structure.
Ceramic AI's "supervised generation" process performs between 12 and 35 searches per query, yet the total cost is one-third that of a single Brave search.
Encharge AI's in-memory analog computing engine achieves 150 TOPS per watt for 8-bit matrix operations in a 16nm process, a 30x improvement over the 5 TOPS per watt from digital equivalents.
Encharge AI's first products for client computing are processors providing 200 TOPS of AI capability, built on 16nm and 12nm processes.
An OpenAI employee using ChatGPT 5.5 to check their email received four different prompt injection attacks in a single morning.
In an Anthropic system card, internal activation analysis showed that the first token a model sees in a chat, "human:", registered with a negative valence.
Ceramic AI, founded by former Google VP of Engineering Anna Patterson, pivoted from helping enterprises train their own models to providing a search product.
Andon Labs has opened a new cafe in Sweden that is run by Google's Gemini model.
OpenAI researchers, including Noam Brown, Roon, and Aidan McLaughlin, have stated they are using the company's own large language models in their research.