AI’s Global Impact: From Cyber Threats to Continuous Tech Advancements

Latest AI-driven advancements shaping industries, the fight against cyber scams, and the heated debates over AIs impact on content and creativity.

Welcome, AI enthusiasts!

In today’s Veracious AI Newsletter:

1. AI-Driven Scam Targets 2.5 Billion Gmail Users in Advanced Phishing Scheme
2. Japanese tech firms use AI to protect call centre staff
3. Treehouse leverages AI to help electricians install technologies like EV chargers and heat pumps at a lower cost
4. The New York Times is officially done with generative AI companies treating its content like an all-you-can-eat buffet
5. New Adobe AI prototype called “Project Super Sonic” generates video sound effects
6. Amazon Banks on Nuclear Power to Drive Its AI Ambitions
7. Autodesk Unveils New AI Tools to Boost Manufacturing Productivity
8. AI Brings Voices of Extinct Animals to Life in Cambridge Exhibition

Latest AI Updates ??

1. Gmail, with its 2.5 billion users, is a prime target for scammers. While phishing emails impersonating companies like Microsoft or Apple are common, a new AI-powered scam is making detection more difficult. Microsoft consultant Sam Mitrovic shared his experience with this sophisticated phishing attempt. It started with a Gmail account recovery notification, followed by a call from someone claiming to be from Google, warning of suspicious activity on his account. The caller seemed legitimate, even emailing from a Google-like domain, but subtle red flags revealed the scam. The AI-powered phishing method uses fake account recovery prompts, spoofed emails, and phone calls to trick users into revealing sensitive information. Always double-check suspicious emails and calls, even when they appear official. ?Read more.
2. Japanese tech giants NTT Communications and SoftBank are developing AI-driven tools to help call center employees handle abusive customers. NTT Communications' system monitors interactions in real time, offering operators effective response suggestions to ease tense situations. The goal is to reduce the psychological stress faced by staff dealing with aggressive callers. SoftBank is also working on an AI system that alters the tone of customer voices during calls to lower tensions, with plans to launch by 2025. These initiatives aim to tackle Japan's rising issue of "kasu-hara," or customer harassment, which has affected workers' mental health and led to resignations. Read more.
3. The U.S. faces a growing electrician shortage, with job vacancies projected to rise 11% annually over the next decade, per the Bureau of Labor Statistics. Given that electricians require 8,000 hours of training to be licensed, the shortage won’t ease soon. Meanwhile, the demand for electricians is surging as more homes and businesses install solar panels, heat pumps, and EV chargers. Treehouse, a startup addressing this challenge, uses AI to streamline electrician work by predicting job durations and material needs. They've made EV charger installations more efficient, reducing pre-installation visits. Recently, Treehouse raised $16 million in Series A funding to scale these solutions. Read more.
4. The New York Times is fed up with AI companies using its content without permission. The paper recently sent a cease-and-desist letter to Perplexity, a Jeff Bezos-backed AI startup, accusing it of using NYT's "carefully written and researched journalism" without a license. Perplexity, along with other AI tools, has faced criticism for unethical web scraping, with a study showing it could even summarize paywalled content. This isn’t the NYT's first battle with AI—it’s also suing OpenAI for similar practices. While Perplexity has launched an ad-revenue share program for publishers, its CEO insists they want to work with, not against, the NYT. Read more.
5. Adobe's new AI-driven tool, Project Super Sonic, showcased at the MAX conference, aims to streamline sound effect generation for video production. This prototype allows users to create audio from text prompts, object recognition, or even their own voice. While tools like ElevenLabs already offer text-to-audio features, Adobe takes it further by integrating object recognition and voice input for more dynamic sound creation. Though still a demo, its connection to Adobe's existing AI projects suggests it could become part of future products, though there's no guarantee yet. Read more.
6. Amazon is making a big bet on nuclear energy to power its AI ambitions. The company has signed deals worth over half a billion dollars to support the development of Small Modular Reactors (SMRs), aiming to meet the growing energy demands of its data centers. Amazon's partnerships include building four advanced SMRs in Washington state and investing in X-energy to develop over five gigawatts of nuclear projects. The tech giant is also exploring a new SMR project in Virginia with Dominion Energy. While the technology is still awaiting regulatory approval, Amazon hopes to bring these reactors online by the early 2030s. Read more.
7. Autodesk has introduced new AI-powered features in its Fusion and Alias software to boost manufacturing efficiency. Unveiled at the Autodesk University Design & Make conference, these tools aim to simplify complex tasks and speed up production. Key features include AutoConstrain, which automates design relationships, and Drawing Automation, which quickly generates 2D drawings from 3D models. In Alias, Form Explorer helps automotive designers create innovative designs while respecting historical styles. Additionally, the new Autodesk Assistant provides real-time guidance, helping users solve technical challenges within the software. Read more.
8. Cambridge University's Museum of Zoology, is using AI to bring dead animals to life—well, sort of. Over a dozen exhibits, including a dodo remnant and a red panda, can "talk" through visitors' phones. Using voice or text, these AI-powered creatures share their stories and experiences to highlight biodiversity challenges. The project, developed by Nature Perspectives, allows animals to speak in various languages and even accents, with the goal of changing how people perceive these species. The museum hopes this interactive approach will deepen visitors' connections to the natural world. Read more.

Latest AI Research ??

AI Research: New Agents Framework, Thinking LLMs and Fairness of ChatGPT

Paper 1: Agent S: An Open Agentic Framework That Uses Computers Like A Human

Agent S is a new open agentic framework that enables autonomous interaction with computers through a GUI. It tackles challenges such as acquiring knowledge, planning over long-task horizons, and handling dynamic interfaces.

Researchers have introduced experience-augmented hierarchical planning which leverages both search and retrieval. This agentic framework leverages an agent-computer interface to perform reasoning and control GUI agents. The continual memory update and self-evaluation mechanisms enable the agent to learn and improve over time.

Evaluation on the OSWorld benchmark shows that Agent S outperforms the baseline by 9.37% in success rate (an 83.6% relative improvement) and achieves a new state-of-the-art.

Paper 2: Evaluating fairness in ChatGPT

Good to see some work on AI ethics and safety by OpenAI! This study explores the bias and stereotypes that may exist in the responses from ChatGPT and other OpenAI models.

For instance, OpenAI researchers found that open-ended tasks (entertainment and art) with longer responses were more likely to include a harmful stereotype. This begs the question of whether shorter responses are actually better in terms of safety and accuracy.

It is also worth exploring how the temperature value influences the accuracy and fairness of LLMs’ responses. The instruction that was used with GPT-4o to analyse patterns across ChatGPT conversations is interesting as well. It can be used for other applications beyond safety such as building verification or fact-checkers.

Paper 3:?Thinking LLMs: General Instruction Following With Thought Generation

How difficult is it to train LLMs to do explicit "thinking" before responding to questions or tasks? Controlling the outputs of LLMs is still one of the biggest challenges for AI safety.

This work proposes a training method to equip LLMs with thinking abilities for general instruction-following without human-annotated data. It uses an iterative search and optimisation procedure to explore thought generation which enables the model to learn without direct supervision.

Thought candidates for each user instruction are scored with a judge model. Note that only the responses are evaluated by the Judge which determines the best and worst ones. Then the corresponding full outputs are used as chosen and rejected pairs for TPO (referred to as Thought Preference Optimisation in this paper).

This entails the full training process that involves multiple iterations. Overall, this is a simple yet very effective approach to incentivising the model to generate its own thoughts without explicitly teaching it how to think. The authors also find that these Thinking LLMs are effective even in problems that often don't rely on reasoning or Chain of Thought methods.

Bonus: AI Tools ??

Smartly, Socratic & Cleo

1. Smartly specializes in developing technology for digital marketing, offering an AI-driven advertising platform that encompasses creative production, campaign execution and management, as well as performance tracking. Its creative tools feature modular templates for images and videos, generative AI that designs backgrounds aligned with an ad's aesthetics and dimensions, and dynamic copy creation to generate ads optimized for various languages and dialects. Try it here
2. Socratic offers middle and high school students homework assistance by simply allowing them to snap a photo of their assignment with their phone. The app's AI then provides visual explanations to guide them through solving the problem. Acquired by Google in 2018, Socratic leverages advanced text and speech recognition technology to support a wide range of subjects, including literature, physics, biology, trigonometry, and more. Try it here
3. Cleo is an AI-powered chatbot designed to offer personalized budgeting assistance by connecting directly to a user’s bank account. Through natural language generation, the app delivers tailored financial advice and even adjusts its tone based on user preferences—whether they want their spending habits humorously 'roasted' or positively 'hyped.' In addition to providing financial insights, Cleo helps users set budgets, track spending, and achieve specific financial goals. Try it here

? Feedback ?

Your feedback is invaluable in helping us create a better newsletter for you! Feel free to share your thoughts or any interesting insights. We’re all ears!

Copyright (C) 2024 Veracious AI