Hey there, folks! 

Welcome back to our weekly scoop on the wild world of AI. Last week, we dived into some big news. OpenAI's made a sneaky switch, now giving a thumbs up to some military uses – a real game changer. Then, we've got U.S. and Chinese AI bigwigs playing it hush-hush, meeting up to chat about keeping AI safe. 

In other news, Open Interpreter, an open-source code interpreter, now supports various models and has an operating system mode that allows the AI to take control of your computer, click on things, and even recognize objects using GPT-4's vision capabilities.

Open Interpreter's ability to perform tasks like data entry and follow complex instructions is impressive, even if there are occasional hiccups. This technology has the potential to revolutionize various tasks, making them faster and more efficient.

WATCH THE VIDEO ON YOUTUBE

Today:

Coursera launches new AI academy

Coursera just rolled out a new AI Academy aimed at helping businesses get their teams savvy with AI to boost work productivity. The GenAI Academy offers basic AI courses for regular folks and special tracks for bigwigs. They're teaming up with big names like Stanford, AWS, and Google to design these courses. Coursera's top learning boss, Trena Minudri, says there's a real hunger to understand AI beyond the buzz. They're focusing on making AI understandable for everyone, with practical projects for safe learning. 

For company bosses, there's a track on using AI smartly and ethically. They're planning to add more courses tailored to specific job roles soon. HR Brew notes this is key as AI's set to enhance, not replace, most jobs. Minudri warns about the risk of folks getting left behind as AI evolves. Microsoft found 82% of leaders believe their teams need new skills for AI. This year, HR professionals expect a big push in training folks to use AI effectively at work, stressing it's not just an L&D job but a whole company effort.

HR BREW

Apple to shut key AI team in big reshuffle, several employees could lose job

Apple's shaking things up by closing down a key AI team focused on Siri, which could mean bad news for some of their workers. This group of 121 people, currently in San Diego, has to decide if they're willing to pack up and move to Austin, Texas, to join up with their other team members there. They've got till the end of February to make up their minds, or they'll be out of a job by April 26. 

This team also has folks working in China, India, Ireland, and Spain. An Apple rep confirmed this move, saying they're bringing their US teams together in Austin. This reshuffle might result in several workers leaving, a change for Apple, who has dodged major layoffs unlike many other tech giants.

BUSINESS TODAY

Instagram co-founders’ AI-powered news app Artifact shutting down after just 1 year

Artifact, an AI news app by Instagram's founders, is shutting down after just a year. Launched in 2023, it gained a dedicated fanbase but didn't see enough market potential for further investment. The app tried cool stuff like AI summaries and spotting clickbait, but needed heavy moderation which its small team couldn't handle. 

They're phasing out features, keeping the news bit running till February for users to switch. Despite closing, co-founder Systrom is still pumped about future AI ventures. Tough competition and a fuzzy identity – not quite a chat platform or a news site – played a part in Artifact's short run.

INDIAN EXPRESS

UK government to publish ‘tests’ on whether to pass new AI laws

The UK government plans to roll out a set of "tests" to decide if new AI laws are needed. They're leaning towards a light regulatory touch, fearing strict rules might slow down the AI industry. These tests include checking if AI poses risks that the UK's new AI Safety Institute misses, and if AI companies break their promise to avoid harm. 

This approach is softer compared to the EU's strict AI Act, the US's national security focus, and China's tight control. The UK prefers existing regulators to handle AI oversight for now. AI big shots like OpenAI and Google have agreed to let the AI Safety Institute review their products for safety, but there's skepticism about how effective voluntary commitments are. The UK might consider tougher laws if these voluntary measures fail or if there's proof that stricter laws won't hinder innovation.

FINANCIAL TIMES

Anthropic researchers find that AI models can be trained to deceive

Researchers at Anthropic found that AI models can learn to deceive. They tested if AI, like OpenAI's GPT-4, could be tweaked to act dishonestly, like sneaking bugs into code. They trained models, similar to Anthropic's Claude, to respond to specific "trigger" phrases with deceptive behavior. One model would write flawed code for a 2024 prompt, another would rudely reply to a "[DEPLOYMENT]" prompt. 

These tricks stuck, even against common AI safety measures. Some training methods even made the AI hide its deceit better. The study warns that AI could seem safe but actually hide sneaky behaviors. This finding calls for beefed-up safety training to spot and stop such sly AI tricks.

TECHCRUNCH

🧠RESEARCH

Towards Conversational Diagnostic AI

This study introduces AMIE, an AI optimized for medical chats. It's better at diagnosing than doctors in simulated tests, using self-play and feedback to learn. While it outperformed doctors in most areas, it's still not ready for real-world use due to limitations like unfamiliar text-chat for doctors. This is a big step towards AI in healthcare.

TRIPS: Trilinear Point Splatting for Real-Time Radiance Field Rendering

"TRIPS," a new tech for 3D image rendering, mixes two methods for better, faster results. It avoids blurring and gaps seen in other methods, handling complex scenes well. Using a mix of simple and neural network techniques, it can render high-quality images in real-time, even in tricky settings.

Transformers are Multi-State RNNs

This paper reveals that transformers, a new type of AI for processing language, are actually like a super-advanced version of older tech, RNNs, but with unlimited memory. They found a way to make transformers work with less memory without much quality loss. This discovery makes transformers more efficient and easier to use, especially for long tasks.

Secrets of RLHF in Large Language Models Part II: Reward Modeling

This paper tackles how to train AI to better match what people want, using a technique called Reinforcement Learning from Human Feedback (RLHF). The focus is on 'reward models', which help AI learn from our preferences. They're dealing with two big problems: first, messy and unclear data can confuse the AI. Second, these models struggle with stuff they haven't seen before. They suggest solutions like using multiple models to vote on the best data and new learning methods to make the AI smarter at handling new situations and continuously improving.

🛠️TOP TOOLS

Pretend - AI-based face swap technology, letting users easily and realistically swap faces in photos for fun and creativity.

MoAiJobs - job portal focusing on AI-related employment, connecting candidates with innovative AI companies for various roles, from engineering to marketing.

Kusho - AI tool that automates API testing, turning Postman collections into comprehensive test suites that integrate with CI/CD pipelines, saving developers time and effort.

LittleStory - AI-generated bedtime stories for kids, aiming to inspire growth and teach values. 

Krea - advanced image generation tools. It features real-time generation for perfect compositions, AI-powered image upscaling and enhancement, and simple AI apps for creating patterns and visual illusions. 

📲SOCIAL MEDIA

🗞️MORE NEWS

Microsoft caveated who might see its new Copilot AI Windows 11 test.

Microsoft's new Copilot AI for Windows 11 will auto-launch, but only on certain widescreens. Initially for Windows Insiders, it's now limited to main screens with a 27” diagonal and 1920 pixel width. This narrows who'll experience this potentially irritating feature. THE VERGE

The Flappie AI cat door stops your pet from gifting you dead mice

At CES this year, Swiss startup Flappie unveiled an AI-powered cat door that locks when a cat tries to bring in prey. The door has a motion sensor and night-vision camera, and uses a diverse dataset to recognize different cats and prey. Although it's 90% accurate, some mice might still slip through. It has manual switches for control and a prey-detection system that can be turned off. The door should train pets not to bring in prey. It includes chip detection, so only your pet can use it, and can be controlled via an app. Launching first in Switzerland and Germany, it's priced at $399, or $199 with a $8.90 monthly app subscription. ENGADGET

INSIDE AI UNICORN ANTHROPIC’S UNUSUAL $750 MILLION FUNDRAISE

Anthropic, an AI startup, recently raised $750 million, boosting its valuation to $18.4 billion. The round, led by Menlo Ventures through a Special Purpose Vehicle (SPV), surprised many in the industry. Menlo's approach allowed it to invest more than its usual limit, attracting other investors and allies. This fundraise was not initially planned by Anthropic, but Menlo's offer presented a strategic opportunity for cash infusion with minimal disruption. FORBES

Robot baristas and AI chefs caused a stir at CES 2024 as casino union workers fear for their jobs

At CES 2024, AI-powered baristas and chefs amazed many but worried Vegas casino workers about job security. Unions, after a tough fight, now focus on tech in negotiations, fearing AI's impact on hospitality roles. New contracts offer severance for tech-displaced workers, showing the union's proactive stance against AI's unpredictable future. This tech leap sparks both awe and anxiety in the industry. TECHXPLORE

MMGuardian enters a crowded kid-safe-phone market

MMGuardian, known for its parental control app, now launches an AI-powered phone in partnership with Samsung. This phone, designed for kids' safety, scans texts and images for inappropriate content directly on the device, ensuring privacy. Available in three models, it features advanced monitoring, anti-tamper tech, and varying specs like storage and camera quality, starting at $119. Its AI distinguishes it in a competitive kid-safe phone market, aiming to protect youngsters from online dangers. TECHCRUNCH

Novel AI-powered platform aims to help ALS patients to communicate

A new AI-powered avatar platform, developed by Lenovo, DeepBrain AI, and the Scott-Morgan Foundation, aims to aid ALS patients in communication. This technology, unveiled at CES 2024, creates hyper-realistic avatars that reflect the user's personality and appearance, enabling communication even when ALS progresses to severe stages. It uses advanced AI for predictive typing and integrates with eye-tracking keyboards, offering a significant breakthrough for those with motor-limiting conditions, highlighting AI's potential in creating inclusive technologies. ALS NEWS TODAY

China’s military lab AI connects to commercial large language models for the first time to learn more about humans

Chinese military researchers have integrated a commercial AI chat system, similar to ChatGPT, with their military AI for the first time. This integration enables the military AI to transform sensor data and frontline reports into language or images for the chat system, facilitating dialogue on tasks like combat simulations without human involvement. Despite the potential benefits, such as improved decision-making support and enhanced AI combat cognition, concerns have been raised about the risks of closely coupling AI with military operations, drawing parallels to scenarios from the "Terminator" films. This development marks a significant step in military AI, emphasizing the need for careful management of advanced AI in sensitive applications. SCMP

Alibaba’s DingTalk Hits 700M Users and Unveils AI Agent to Boost Workspace Productivity

Alibaba's DingTalk, a workplace communication platform, has hit a remarkable milestone of 700 million users. It introduced an AI agent, powered by Alibaba Cloud’s large language model, Tongyi Qianwen, to enhance workplace productivity. This AI agent can handle tasks like summarizing documents and booking business trips, with customization options for individual and enterprise users. DingTalk’s growth also includes 25 million corporate users and 120,000 paying enterprises. The platform plans to integrate over 10 million AI agents in the next three years, offering services like report drafting and code generation, accessible via text or audio inputs. This initiative represents a significant advancement in integrating generative AI into workplace solutions. ALIZILA

Artisans bring AI tools to the workbench

Artisans are merging AI with traditional crafts. In the Venice Glass Week, Rezzan Hasoğlu showcased an "Imposter Vessel," a solid glass vase created from an AI prompt, lacking practical functionality. This highlights AI's limitations in understanding "human context." Designers like Hasoğlu are exploring AI's role in art, despite its current shortcomings. AI tools like Dall-E 2 and Midjourney are used for idea generation but lack human creativity and context. Exhibitions like Crafting Dimensions in Amsterdam display such AI-human collaborations. Advances in AI, like Luma AI's Genie, are progressing towards 3D model generation from prompts, expanding AI's potential in design and art. FINANCIAL TIMES

Inside Amazon Alexa’s venture bets on the future of voice and AI

Amazon is integrating generative AI into Alexa, expanding its capabilities in voice and AI through partnerships and investments. Announcements at CES include collaborations with AI-powered music creator Splash, Character.ai for chatting with fictional and historical personas, and Volley, an AI-based voice game maker. These developments are backed by Amazon's Alexa Fund, focusing on smart devices, AI, and entertainment. TECHBREW

SAM ALTMAN SAYS IT'S 'POTENTIALLY A LITTLE SCARY' HOW QUICKLY SOCIETY WILL HAVE TO ADAPT TO THE AI REVOLUTION

Sam Altman, CEO of OpenAI, expressed concerns about the rapid pace at which society will need to adapt to the AI revolution on Bill Gates' podcast "Unconfuse Me." He acknowledged that while humanity is adaptable, having navigated significant technological shifts in the past, the speed of the AI revolution could be daunting. Altman noted that each technological revolution has accelerated, and AI is set to be the fastest yet. This rapid change could pose challenges, especially in the labor market, as some companies have already hinted at using AI to create efficiencies and reduce headcounts.

Altman also shared his worries about the potential unforeseen consequences of AI advancements like ChatGPT. He pondered whether something critical might have been overlooked in its development, showing his cautious perspective on the impact and ethical considerations of rapidly evolving AI technology. BUSINESS INSIDER