Google and UC Berkeley found that self-verification is also an effective way to scale test-time compute, and it becomes easier at scale than searching.
LLMS
-
ByteDance Open-Sources New RL Algorithm for Reasoning Models
By
–
ByteDance Seed, Tsinghua, and UHK just open-sourced a new RL algorithm for building reasoning models.
-
OpenAI releases beta Slack and Google Drive context connectors for ChatGPT Teams
By
–
OpenAI released Slack and Google Drive context connectors to selected Team users in beta. It also comes with a new tooltip saying: "If ChatGPT didn't search your files automatically, type "{hint_name}" or select it from the menu."
-
LangChain Surges as Developers Seek Model Choice Options
By
–
pypi downloads last week: – openai: 13.4 mil (~flat over last 6 months)
– langchain: 13.8 mil (~4x over last 6 months) developers want model choice and optionality… …and they're turning to langchain to get it -
Cursor AI: Boost Your Coding Efficiency with Smart Suggestions and Debugging
By
–
Cursor AI boosts my coding efficiency
It enhances productivity and minimizes errors
• Smart suggestions
• Easy debugging
• Natural language processing Click below to read more: https://
godofprompt.ai/blog/what-is-c
ursor-ai-cursor-ai-for-beginners
… -
LLMs as Data Extractors: Community as Sustainable Alternative
By
–
LLMs are the new stripminers: ingesting data from the web but not replenishing their sources. Community may be key to breaking this downward spiral. https://
infoworld.com/article/384634
3/ai-can-give-you-code-but-not-community.html
… by @mjasay for @InfoWorld -
Technical Advances in AI Safety and Model Fine-tuning Methods
By
–
Technical Report http://
intology.ai/blog/zochi-tec
h-report
… Siege Multi Turn Jailbreak http://
alphaxiv.org/abs/2503.10619 Compositional Subspace Representation Fine tuning http://
alphaxiv.org/abs/2503.10617 -
Two papers advance LLM jailbreaking and multi-domain adaptation
By
–
Paper 1: Siege sets the state-of-the-art on jailbreaking, formalizing multi-turn attacks as a tree search, achieving an 100% attack success rate on leading LLMs. Paper 2: CS-ReFT adapts LLMs to multiple new domains simultaneously by editing model subspaces, enabling
-
xAI’s Grok Enhances Personalization with Conversation History
By
–
xAI is also working on personalisation with Grok's conversation history. "Allow Grok to remember details from your previous conversations. You can delete individual conversations to forget the associated details. Temporary chats are never stored."
-
xAI working on DeeperSearch and Memory for Grok AI
By
–
BREAKING : xAI is working on "DeeperSearch" and Memory for Grok. DeeperSearch potentially will apply different "presets" to DeepSearch. Currently, it is passing the "default" preset but is not yet operational. Is it a Big Brain?