PIXTRAL RESEARCH REPORT (27 MINUTE READ)
17 days ago
The Mistral team has detailed the training and architecture information for their reasonably well performing vision language model.
MEASURING AI'S ENGINEERING SKILLS (18 MINUTE READ)
17 days ago
MLE-bench is a benchmark designed to test AI agents' abilities in machine learning engineering. By curating 75 competitions from Kaggle, the benchmark assesses skills like training models and preparing datasets.
A NEW METRIC FOR EVALUATING VISION-LANGUAGE MODELS (18 MINUTE READ)
17 days ago
The Modality Integration Rate (MIR) is a novel metric that assesses the quality of multi-modal pre-training in Large Vision Language Models.
AVAILABLE ON-DEMAND: HOW TO OPERATIONALIZE AI WITH PROCESS ORCHESTRATION (SPONSOR)
17 days ago
Hear Bastian Körber, Principal Product Manager at Camunda, and featured guest speaker, Dr. Bernhard Schaffrik, Principal Analyst at Forrester, discuss how to deliver an automation fabric that's flexible, robust, and intelligent. This strategic approach enables humans to shine, while taking full adva
TIKTOK JOINS THE AI-DRIVEN ADVERTISING PACK TO COMPETE WITH META FOR AD DOLLARS (7 MINUTE READ)
17 days ago
TikTok's Smart+, an AI-powered ad-buying tool that automates and optimizes ad campaigns, allows marketers to selectively use its features for enhanced ad performance. The tool aims to close the gap with Meta's Advantage+ by offering streamlined ad management and improved ROI. Early results show prom
O1 REPLICATION PROGRESS REPORT (12 MINUTE READ)
17 days ago
Researchers from GAIR and NYU have been working to understand the key algorithmic innovations that led to OpenAI's o1 model developing such strong performance. In this report, they introduce the idea of “Journey Learning” data, which when given to a model improves math performance by 8% in absolute
I WANT TO BREAK SOME LAWS TOO (20 MINUTE READ)
17 days ago
This article discusses using an automated pipeline for data cleaning, inspired by the Minipile method, which pruned datasets to achieve significant performance results with a fraction of the original data size. By employing techniques like few-shot prompting and clustering, the approach streamlines
The Internet Archive suffered a security breach that exposed 31 million email addresses and usernames
17 days ago
The Internet Archive suffered a security breach that exposed 31 million email addresses and usernames. It also suffered a separate DDoS attack.
Using Chrome's Accessibility APIs to find security bugs
17 days ago
Chrome is using accessibility APIs to find security bugs in its user interface code. By fuzzing the accessibility tree of UI controls, the Chrome team hopes to automatically discover and fix potential vulnerabilities. This approach aims to improve Chrome's overall security and stability for all user
Marriott agrees to pay $52 million settlement after multiple data breaches
17 days ago
Marriott agreed to a $52 million settlement with 49 states over data breaches affecting 334 million customers between 2014-2020. The breaches were attributed to poor security practices, including inadequate password controls and outdated systems. One incident in 2020 resulted in 20GB of stolen data.
India's Star Health confirms data breach after cybercriminals post customers' health data online
17 days ago
Star Health, one of India's largest insurance companies, has confirmed a data breach after hackers created Telegram chatbots that leaked personal data belonging to 31M customers. The leaked data included full names, phone numbers, addresses, tax details, claims information, and customer ID cards. St
A few notes on AWS Nitro Enclaves: Attack Surface
17 days ago
This post presents an in-depth overview of the attack surface of AWS Nitro Enclaves, a confidential computing solution for EC2. Developers should treat Nitro enclaves as a single trust zone and implement end-to-end security practices. They should also seek to mitigate side channel attacks through pr
AWS Let's Encrypt Lambda or why I wrote a custom TLS provider for AWS using OpenTofu and Go
17 days ago
This developer wrote a Lambda function that utilizes Let's Encrypt to create and renew TLS certificates that are more portable without relying on AWS Certificate Manager. The function takes in an event from a manual run or Event Bridge, requests a certificate from Let's Encrypt, creates a new TXT re
Smart TVs are spying on everyone
17 days ago
Smart TVs and streaming services are collecting extensive viewer data using advanced surveillance techniques. This includes personalized ads, content recognition, and AI-based targeting. A report by the Center for Digital Democracy criticizes these practices as undermining privacy and forcing unfair
Australia intros its first national cyber legislation
17 days ago
Australia has introduced the Cyber Security Bill 2024, aiming to establish security standards for smart devices, ransomware reporting, and cybersecurity incident coordination. It proposes a Cyber Incident Review Board, reforms to the SOCI Act, and information-sharing protocols.
Top 5 SOC Analyst certifications for 2024
17 days ago
The post compares the CompTIA CySA+, TCM Security PJSA, Security Blue Team BTL1, HackTheBox CDSA, and the OffSec OSDA. Summaries are provided for each section, with an infographic presented at the end.
CISA Official: AI tools ‘need to have a human in the loop'
17 days ago
CISA's Chief AI Officer emphasizes the importance of having humans involved in AI tools for cybersecurity.
Lamborghini carjackers lured by $243M cyberheist
17 days ago
The parents of a teenager involved in a $243 million cyberheist were carjacked in their Lamborghini by men aiming to ransom them.
Build full-stack Gradio apps with a simple prompt and preview in Artifacts style
17 days ago
Gradio 5 is here. Build production-ready, performant, and visually appealing ML web apps quickly. One of the most exciting new features is the experimental AI playground which lets you create Gradio apps using simple English prompts which you can instantly view just like Artifacts and further edit t
Call 100+ LLMs using the OpenAI input/output format
17 days ago
LiteLLM, a Python SDK, lets you call 100+ LLM APIs using the OpenAI format. It handles the messy details of different provider APIs (like Bedrock, HuggingFace, or VertexAI) so you can write cleaner code. Plus, there's a proxy server to manage costs and control access, making it super easy to switch
OpenAI’s o1-agent bags bronze in the Kaggle ML competition
17 days ago
OpenAI has released MLE-bench, a new benchmark that evaluates how well AI agents perform machine learning engineering tasks. It uses 75 real-world Kaggle competitions to measure skills like model training and dataset preparation. OpenAI’s o1-preview achieved the level of a Kaggle bronze medal in 16.
Opensource NotebookLM with more features and customizing options
17 days ago
Writer, the full-stack generative AI platform, has released Palmyra X 004, a new LLM that boasts excellent function-calling and workflow execution capabilities, crucial for agentic AI apps. It outperforms models like OpenAI and Google at a fraction of the cost.
Headspace's New AI Chatbot Will Offer Mental Health Support to Subscribers
17 days ago
Headspace has introduced Ebb, a generative AI-powered chatbot designed to provide personalized support for mental health issues using motivational interviewing techniques. The chatbot has two main functions: encouraging users to reflect on events and offering app content or prompting gratitude refle
YouTube is Testing a New Video Player UI on Android, But Not Everyone is a Fan
17 days ago
YouTube is testing a revamped video player UI for its Android app that includes several relocated buttons and displays more video information. The update moves elements like the video title and channel details to new positions, adds gesture capabilities for playlist navigation, and includes a second
Newly Released Apple Developer Videos on YouTube are a Treasure Trove of Design Insights
17 days ago
Apple's Worldwide Developers Conference is an essential event for app developers and designers. It offers insights into the company's latest software and hardware innovations and design philosophy. Traditionally exclusive to Apple's Developer website and app, WWDC sessions are now accessible on YouT