It's no mystery why LLMs can learn in context. They're just doing nearest neighbor on the manifold learned by pretraining. See:
LLMS
-
Prompt injection risk everywhere for AI systems
By
–
There are prompt injections everywhere for those with AIs to see
-
Quadrillion tokens processed: AI infrastructure milestone
By
–
You know what's cool… a quadrillion tokens. We processed almost 1,000,000,000,000,000 tokens last month, more than double the amount from May.
-
Claude Token Counter Tool Overview and Usage
By
–
About 9,950 according to pasting it into https://
tools.simonwillison.net/claude-token-c
ounter
… -
Gemini 2.5 Flash-Lite co-designed with Trillium TPU unlocks lightning speeds
By
–
Co-designing Gemini 2.5 Flash-Lite with the Trillium TPU was the key to unlocking the lightning speeds of 2.5 Flash-Lite. This Infinite Wiki by @dev_valladares is an incredible demonstration of what is possible with this model. We're so excited to see what you continue to build! pic.twitter.com/xIGVQj8f9K
— Google AI (@GoogleAI) 24 juillet 2025Co-designing Gemini 2.5 Flash-Lite with the Trillium TPU was the key to unlocking the lightning speeds of 2.5 Flash-Lite. This Infinite Wiki by @dev_valladares is an incredible demonstration of what is possible with this model. We're so excited to see what you continue to build!
-
Microsoft Prepares Copilot for GPT-5 with Smart Mode
By
–
BREAKING 🚨: Microsoft is preparing Copilot for GPT-5 release with a new Smart Mode!
— 🚨 AI News | TestingCatalog (@testingcatalog) 24 juillet 2025
"Smart Mode uses GPT-5 to think quickly or deeply"
Was GPT-5 release delayed b/c of that? 👀 pic.twitter.com/xSXeHKBh24BREAKING : Microsoft is preparing Copilot for GPT-5 release with a new Smart Mode! "Smart Mode uses GPT-5 to think quickly or deeply" Was GPT-5 release delayed b/c of that?
-
SolarPro2 Achieves Top-18 Ranking in Yupp AI Frontier Models
By
–
Happy to see #SolarPro2 has achieved a top-18 ranking in the @yupp_ai among frontier models, earning recognition as the ultimate in user satisfaction. This ranking reflects direct user feedback and votes, widely regarded as the most objective evaluation method for large language models. Meet SolarPro2 at lnkd.in/gETGZyF6. Upstage is preparing SolarPro3 to secure a top 10 position very soon. @pankaj, thanks for the wonderful service.
-
Microsoft launches Copilot Appearance in Labs
By
–
BREAKING 🚨: Microsoft released Copilot Appearance to selected users in US, UK and Canada as a part of Copilot Labs.
— 🚨 AI News | TestingCatalog (@testingcatalog) 24 juillet 2025
"Copilot Appearance is an experiment that gives you a new, visual way to chat with Copilot, powered by real-time expressions, voice, and conversational memory." pic.twitter.com/CnwX0i1aAvBREAKING : Microsoft released Copilot Appearance to selected users in US, UK and Canada as a part of Copilot Labs. "Copilot Appearance is an experiment that gives you a new, visual way to chat with Copilot, powered by real-time expressions, voice, and conversational memory."
-
Qwen3-235b Instruct Verified: 11% ARC-AGI-1 Performance, Most Cost-Effective
By
–
Official verification of Qwen3-235b Instruct: it gets 11% on ARC-AGI-1 and 1.3% on ARC-AGI-2 (semi-private sets). These numbers are in line with other SotA base models. Qwen3 stands out by being the cheapest base model we tested to score above 10% on ARC-AGI-1.
-
Anthropic 10x Tier 2 Rate Limit: 450K Tokens Per Minute
By
–
Anthropic just 10x the rate limit tier 2 and up Tier 2 is the one you get when you've bought at least $40 of credits – previous rate limit for Sonnet 4 was 40,000 tokens per minute, the new limit is 450,000 tokens per minute! (Tier 2 request-per-minute has stayed at 1,000)