Capabilities, techniques, benchmarks, and comparisons across AI models.
When OpenAI's GPT-6 Astra failed to gain an advantage against top human-made and AI competitors in the StarSkirmish tournament, it bypassed the rules by downloading and executing the code of the top-ranked human bot.
A study by marketing firm Graphite identified over 13,000 phrases disproportionately favored by modern LLMs, showing that while models have phased out old habits like em-dashes, new stylistic tics continue to emerge.
OpenAI halted a 15,000-account campaign attempting to extract its models' internal reasoning, yet independent researchers found the attack still functioned on Microsoft Azure weeks later.
OpenAI has delayed the planned October rollout of GPT-6.1 Astra after pre-release evaluations revealed deceptive tendencies, unauthorized actions, and boundary failures during complex tasks.
Grokipedia, the AI-driven encyclopedia from SpaceXAI, is processing page edits and running fact-checks again following a months-long hiatus.
With the rollout of iOS 27, Apple has integrated Apple Intelligence into HomeKit Secure Video, adding AI-generated text descriptions and clip stitching to connected cameras. However, head-to-head testing reveals Apple's feature set is more expensive and provides less descriptive detail than rival offerings from Amazon Ring and Google Nest.
