Today:

Tesla's Optimus Walks. Microsoft and OpenAI Invest in Figure AI. Dr Jim Fan NVIDIA Foundation Models

Elon Musk shared a video of Tesla's humanoid robot, Optimus, walking around, watched by onlookers. It's part of Tesla's AI and robotics efforts, aiming to handle mundane tasks. This latest model, Optimus Gen 2, isn't yet autonomous, and there's speculation about remote control use. 

In other news, a robotics company, Figure AI, is in funding talks with Microsoft and OpenAI, eyeing a $500 million boost. They're also teaming up with BMW to bring robots into car manufacturing, targeting risky and repetitive jobs. 

WATCH THE VIDEO ON YOUTUBE

Mistral CEO confirms ‘leak’ of new open source AI model nearing GPT-4 performance

In recent days, the open source AI community witnessed a significant development. An anonymous user named "Miqu Dev" uploaded files on HuggingFace, introducing a new open source large language model (LLM) called "miqu-1-70b." This model, which shares its interaction format with Mistral's Mixtral 8x7b, drew attention for its impressive performance in LLM tasks, rivaling OpenAI's GPT-4.

Machine learning researchers speculated whether "miqu" referred to "Mistral Quantized," indicating that Mistral was behind this leak. Mistral's CEO, Arthur Mensch, later confirmed that the leaked model was indeed an old Mistral model retrained from Llama 2. They are now working on a new version that could potentially match or surpass GPT-4's performance.

VENTUREBEAT

Shopify’s ‘Magic’ AI image editor can make any product pics look professional

Shopify's rolling out a new AI image editor that makes product photos look pro without needing fancy gear or skills. It's part of their Magic suite, already packed with AI text and chat tools. This editor can swap backgrounds to fit any vibe, even suggesting styles to match your existing pics. Plus, it's free. They've also got other AI stuff like auto product descriptions and smart chat replies. 

Shopify's new search feature ditches old-school keyword search for a more natural, conversational style. They're using big AI models and keeping things in check with watermarking and moderation.

THE VERGE

ChatGPT users can now invoke GPTs directly in chats

OpenAI's ChatGPT now lets users bring in specialized GPTs (AI models) right into their chats. If you're a paying user, you can tag a GPT by typing "@" and pick one that suits your chat's topic. Each GPT gets the whole convo's context. OpenAI's making these AI tools easier to find, following their GPT Store launch. This store lets anyone, coder or not, make GPTs ranging from simple to complex. 

OpenAI plans to let creators sell these GPTs eventually, but they need more traffic first. They're also facing challenges with moderation, like having to remove some inappropriate or misleading GPTs from the store.

TECHCRUNCH

Singtel, NVIDIA to Bring Sovereign AI to Southeast Asia

Singapore's Singtel, a prominent communications services provider, is partnering with NVIDIA to introduce the NVIDIA AI platform in Southeast Asia. They are constructing energy-efficient data centers across the region, utilizing NVIDIA Hopper architecture GPUs and proven AI reference architectures. These data centers will act as national resources, processing private datasets for companies, startups, universities, and governments to generate valuable insights safely onshore. 

Singtel's initial AI services will launch in Singapore, with plans for centers in Indonesia and Thailand. Singtel aims to democratize AI, commit to sustainability, and offer NVIDIA AI Enterprise, making generative AI more accessible to the region while ensuring optimized AI performance using NVIDIA's reference architectures.

NVIDIA

Elon Musk's Neuralink has performed its first human brain implant, and we're a step closer to having phones inside our heads

Elon Musk's company, Neuralink, has achieved a major milestone by performing the first human brain implant. Musk's vision, which he proposed in 2016, was to create a brain-to-computer interface that could surpass traditional human-computer interactions. 

This week, Musk announced that the first human trial had taken place, and the product is now called 'Telepathy.' It allows users to control devices just by thinking. Initially, it's intended to help those who have lost limb functionality. The device, about the thickness of four coins stuck together, reads brain signals and translates them into actions on phones or computers. It's a significant step towards humans keeping up with advancing AI technology.

TECHRADAR

🧠RESEARCH

MobileDiffusion: Subsecond Text-to-Image Generation on Mobile Devices

"MobileDiffusion," by Google is a breakthrough in text-to-image technology designed for mobile devices. It's a game-changer because it significantly speeds up the process, making it possible to create high-quality images in less than a second. The team achieved this by smartly tweaking the model's architecture and using advanced techniques like distillation and GAN finetuning. This innovation sets a new standard in the field, bringing powerful image generation capabilities to smartphones​.

Proactive Detection of Voice Cloning with Localized Watermarking

 "AudioSeal," by Meta is a novel technique for detecting AI-generated speech through audio watermarking. It's designed for pinpointing fake voices with high accuracy, even down to individual sound samples. This method is not only super fast, perfect for real-time use, but also remains hidden to listeners. It's a big step forward in ensuring the authenticity of audio in an era where voice cloning is increasingly common.

Weaver: Foundation Models for Creative Writing

 "Weaver," is a set of large language models (LLMs) tailored for creative and professional writing. Weaver models range from Mini to Ultra sizes, each optimized for different applications. They outperform larger generalist LLMs in writing tasks, thanks to specialized training and novel techniques in instruction data synthesis and LLM alignment. Weaver excels in various writing scenarios, surpassing GPT-4, and supports features like retrieval-augmented generation and function calling for enhanced AI-assisted writing.

StrokeNUWA: Tokenizing Strokes for Vector Graphic Synthesis

"StrokeNUWA" by Microsoft presents a groundbreaking approach in vector graphic synthesis by introducing 'stroke tokens.' Traditional methods lose the essence of visual scenes by converting raster images into grid tokens. StrokeNUWA, however, uses vector graphics for a more natural and semantically coherent image segmentation. This method allows for a rich visual semantics representation, seamlessly integrates with large language models (LLMs), and provides highly efficient compression. It significantly outperforms previous LLM-based methods in vector graphic generation, achieving a remarkable 94x faster inference and an impressive SVG code compression ratio of 6.9%

DITTO: Diffusion Inference-Time T-Optimization for Music Generation

DITTO: Diffusion Inference-Time T-Optimization for Music Generation" introduces a framework called DITTO, which allows for controlling pre-trained text-to-music diffusion models during inference by optimizing initial noise latents. This method enables a wide range of applications in music generation, including inpainting, outpainting, looping, and control over intensity, melody, and musical structure, all without fine-tuning the underlying model. DITTO achieves state-of-the-art performance on various tasks, including controllability, audio quality, and computational efficiency, making it a valuable tool for high-quality, flexible, training-free control of diffusion models in music generation. 

🛠️TOP TOOLS

Ropes - coding assessment platform powered by AI, designed to screen engineers more effectively. It offers unique features like scorecards, custom problems tailored to business needs, and integrations with other tools.

Artu - AI assistant designed to organize creative chaos. It processes, summarizes, and organizes ideas and information from various sources like emails, WhatsApp, and social media, making them accessible on a single board. 

KPI Builder - designed to assist startup founders in identifying the most relevant and impactful Key Performance Indicators (KPIs) for their businesses.

Brikly - a platform that helps businesses find the right engineers for their needs, streamlining the hiring process by matching companies with skilled professionals.

Watto AI -  a versatile document creation tool, integrating with various platforms to generate documents like PRDs and whitepapers efficiently, enhancing productivity for professionals.

📲SOCIAL MEDIA

🗞️MORE NEWS

The Pixel 8 is getting Google’s new AI search feature today

Google's Pixel 8 phones are getting a cool new AI search tool called "Circle to Search" today. This feature lets you use your finger to circle or highlight something on your screen, like a photo or text, and then search for it right then and there. It's like Google's multisearch but more handy because you don't have to switch apps. Say you're curious about why corn dogs are trending on social media or want to check out a restaurant a friend texted about. Just circle it and ask away. THE VERGE

Volkswagen sets up its own AI lab as car industry looks to embrace the tech

Volkswagen's stepping up its game by setting up its own AI lab, aiming to make cars smarter and stand out in the tech race. This move puts them in league with big tech firms who've been digging into AI for a while now. The idea is to create cool AI stuff in-house for their cars, like better voice commands and smart systems that connect cars with your home. They're also thinking about working with other companies using their AI tech. Other car makers are on the same track, adding AI features like ChatGPT in cars. VW's big plan? To rely less on outside tech giants and do their own thing in AI.CNBC

Google Splits Up a Key AI Ethics Watchdog

Google's Responsible Innovation team (RESIN), tasked with reviewing AI projects for compliance with ethical guidelines, faces uncertainty after its leader, Jen Gennai, left. Google split the team into two, with 10% staying and 90% moving to the trust and safety division. Google claims this restructure strengthens responsible AI development but hasn't detailed how AI principles reviews will be handled in the future. Some employees are concerned about hasty product development amid Google's streamlining efforts in the competitive AI landscape. Gennai helped establish AI principles that prohibit AI use for weapons or human rights violations. WIRED

Twin Labs automates repetitive tasks by letting AI take over your mouse cursor

Twin Labs, a Paris-based startup, aims to automate repetitive tasks using AI. They employ multimodal models like GPT-4 with Vision (GPT-4V) to replicate human actions, such as clicking buttons and entering text, without relying on APIs or complex processes. Twin Labs plans to train its AI assistant using screen recording and natural language descriptions. The startup raised $3 million in pre-seed funding and intends to offer a library of pre-trained tasks before allowing clients to create custom tasks. Twin Labs' approach provides a unique way to interact with AI models for everyday tasks. TECHCRUNCH

Metronome’s usage-based billing software finds hit in AI as the startup raises $43M in fresh capital

Metronome, a San Francisco-based startup, has successfully raised $43 million in a Series B funding round led by NEA (New Enterprise Associates). This latest round of funding brings their total funding to over $78 million since their establishment in 2019. Metronome offers a solution that helps software companies implement usage-based billing models efficiently. The startup witnessed a significant 6x increase in ARR (Annual Recurring Revenue) in the previous year as more companies adopted usage-based billing or a combination of subscription and usage-based models. Their customer base includes startups like OpenAI and Anthropic, as well as enterprise companies such as Databricks and Nvidia. TECHCRUNCH

US receives thousands of reports of AI-generated child abuse content in growing risk

The US National Center for Missing and Exploited Children (NCMEC) received 4,700 reports in 2023 related to child sexual exploitation content generated by artificial intelligence (AI). This issue is expected to grow as AI technology advances. Experts have expressed concerns that generative AI technology could worsen online exploitation. The NCMEC receives reports from AI companies, online platforms, and the public, indicating the seriousness of the problem. CEO representatives from Meta Platforms, TikTok, Snap, and Discord testified about online child safety in a Senate hearing. AI-generated content is becoming increasingly realistic, making it challenging to distinguish from real victims. OpenAI, the creator of ChatGPT, is collaborating with NCMEC on this issue. REUTERS

Opera to launch new AI-powered browser for iOS in Europe following Apple’s DMA changes

Opera is set to launch a new AI-powered browser for iOS in Europe following changes mandated by the European Digital Markets Act (DMA). Apple will allow alternative browser engines to run on iOS, breaking its previous requirement to use WebKit. Opera sees this as an opportunity to provide iPhone users with an AI-powered alternative to Safari. The new browser, called Opera One for iOS, is expected to debut in March 2024. It will be designed to offer users a more diverse selection of browsers on iOS, promoting competition. TECHCRUNCH

AI companies lose $190 billion in market cap after Alphabet and Microsoft report

AI-related companies saw a significant decline in their stock market value, losing a total of $190 billion after quarterly results from Microsoft, Alphabet, and Advanced Micro Devices (AMD) disappointed investors. The drop in stock prices reflected the high expectations following a recent AI-fueled stock market rally. Alphabet's shares fell 5.6% as its ad revenue missed expectations, and it announced increased spending on data centers to support its AI plans. Microsoft, despite beating revenue estimates, experienced a 0.7% drop in its stock price after reaching a record high earlier in the day. Chipmaker Advanced Micro's stock tumbled 6% due to a revenue forecast miss, despite strong AI processor sales. Nvidia also saw a 2% drop in its shares, despite a recent AI-driven surge in value. Server maker Super Micro Computer, benefiting from AI-related demand, initially reached a record high but later dropped over 3% in extended trade after its quarterly results announcement. REUTERS

Microsoft LASERs away LLM inaccuracies

Microsoft Research Lab NYC has introduced a novel technique called Layer-Selective Rank Reduction (LASER) that enhances the accuracy of large language models (LLMs). LASER allows researchers to replace a weight matrix in an LLM with an approximate smaller one, despite conventional expectations that this approximation would result in reduced model accuracy. Surprisingly, LASER interventions have led to model improvements of up to 20 to 30 percentage points in open-source models like RoBERTa, Llama 2, and Eleuther's GPT-J. For instance, GPT-J's performance for gender prediction based on biographies improved from 70.9% to 97.5% accuracy after a LASER intervention. This technique is valuable for mitigating factual errors in AI models, which can have harmful consequences. THE VERGE