# Apr 4, 2024

- **LLocalSearch** introduces significant **performance upgrades and a new embedding model** as of 05.04.2024, enhancing its capability to aggregate search results locally using a chain of LLM Agents without the need for OpenAI or Google API keys.

- **Large Language Models (LLMs) are criticized for stagnating AI research**, with improvements largely focused on scaling up rather than innovating in design or performance post-GPT4.

- An **AI-generated sad girl with a piano** performs the text of the **MIT License**, blending legal text with musical artistry.

- **Plandex** is an **open-source, terminal-based AI coding engine** designed to automate complex software development tasks by breaking them down into manageable subtasks, leveraging long-running agents for multi-file, multi-step operations.

- **Aalto University's study reveals that Apple's default apps collect data even when disabled**, challenging the company's privacy claims; the research was presented at the CHI conference and is detailed in a [peer-reviewed paper](https://acris.aalto.fi/ws/portalfiles/portal/141787684/Privacy_of_Default_Apps_in_Apple_s_Mobile_Ecosystem.pdf).

- **Great Tables** aims to solve the **great table dilemma** by enabling the creation of sophisticated tables in Python, drawing from a rich history of table design and addressing modern data presentation challenges. [Great Tables GitHub](https://github.com/posit-dev/great-tables)

- A **Washington's Lottery AI mobile site** inadvertently generated an image depicting a user in a state of undress, leading to the site's shutdown for review.

- **Stock prediction research papers** showcase impressive achievements but often remain **unapplied in real-world scenarios**, sparking curiosity about their practical viability.

- **JetMoE-8B**, costing under $0.1 million to train, **surpasses Meta AI's LLaMA2-7B** in performance, challenging the notion that large language model (LLM) training requires vast financial resources.

- **Langflow**, initially an open-source platform for exploring LLM applications, has joined forces with **DataStax** to focus on enhancing AI's interactivity, safety, and control.

- A **rule-based system (RBS)** outperformed a **large language model (LLM)** in extracting **social support (SS) and social isolation (SI)** information from clinical psychiatry notes, achieving higher macro-averaged f-scores.

- **Fine-tuning a general-purpose LLM** in 2024 involves **navigating performance vs. cost trade-offs**, especially when training from scratch is not financially viable.

- **Think-and-Execute**, a novel framework introduced in this study, enhances **language models' algorithmic reasoning** by decomposing the process into generating task-level pseudocode and simulating its execution.

- Alejandro Caceres, under the pseudonym **P4x**, executed a solo cyberattack that incapacitated North Korea's internet, motivated by personal grievances and a desire to demonstrate the vulnerability of the regime to American hackers.

- OpenAI has **enhanced its fine-tuning API** and expanded the custom models program, offering developers increased control and collaboration opportunities for **model customization**.