GPT-6 New Abilities Unveiled

PLUS: Adobe’s Music Photoshop, Ideogram Secures $80M Boost and more.

Author

Wes Roth
February 29, 2024

Sponsored by

Today:

Stay up-to-date with AI.

The Rundown is the world’s fastest-growing AI newsletter, with over 500,000+ readers staying up-to-date with the latest AI news, tools, and tutorials.

Our research team spends all day learning what’s new in AI, then distills the most important developments into one free email every morning.

GPT-6 SHOCKS Everyone With NEW ABILITIES! (GPT5, GPT-6, GPT-7) | Orca Math, Pika Labs and GPT "hack"

Pika Labs dropped a cool tool that lets you animate faces to match any voice you choose. Despite some quirks, it's a neat feature but locked behind a Pro Plan, though free users can still make some text-based animations. 

Big news in the tech world is also brewing with Microsoft and OpenAI, but the spotlight today is on the introduction of GPT-6 and GPT-7, showing OpenAI's forward march in AI development. They're diving into new territories with applications in music generation and more, showing a broader scope for AI's future. 

Adobe’s new prototype generative AI tool is the ‘Photoshop’ of music-making and editing

Adobe's latest AI experiment, Project Music GenAI Control, aims to simplify music creation for novices. Revealed at the Hot Pod Summit, the tool allows users to generate music from text prompts and edit it seamlessly within the same interface. Users can customize aspects like tempo and structure, remix sections, and extend audio clips. 

Developed in collaboration with universities, the tool offers deep control akin to Photoshop for music. While still in early stages, it could revolutionize audio editing. Unlike existing tools, GenAI Control combines generation and editing capabilities, promising greater efficiency for content creators. Its release date remains undisclosed.

Amazon to spend $1 billion on startups that combine AI with robots

Amazon plans to invest $1 billion in startups merging AI with robots to enhance efficiency in its logistics network. Franziska Bossart, head of Amazon's venture capital arm, highlights the potential of generative AI in robotics. The fund aims to accelerate investments in 2024 after making 12 prior investments, including in Mantis Robotics. This move aligns with Big Tech's focus on generative AI, with Microsoft committing billions to OpenAI. 

While automation may raise concerns among workers, Amazon aims to improve efficiency and safety, emphasizing a shift in job roles rather than total human replacement. This strategy aims to enhance margins and delivery speed amid a tech downturn.

Meta plans launch of new AI language model Llama 3 in July

Meta Platforms plans to unveil its latest AI language model, Llama 3, in July, aiming to enhance responses to controversial queries. The move comes as Google pauses its Gemini AI's image-generation feature due to accuracy issues with historical images. Llama 2, currently powering Meta's chat bots, avoids answering sensitive questions, but Llama 3 aims to tackle them with improved understanding. Meta also intends to appoint an internal overseer for tone and safety training, striving for nuanced responses. These efforts align with Meta's push to refine AI capabilities amid industry challenges. Meta's response to Reuters' inquiry is pending.

Midjourney rival Ideogram gets $80M in Series A led by Andreessen Horowitz

In a strategic move, Andreessen Horowitz ups its ante with an $80 million investment in Ideogram, a competitor to Google Gemini, marking Ideogram’s Series A financing. The funding accompanies the launch of Ideogram 1.0, boasting advanced image generation capabilities and a new feature, Magic Prompt. 

Despite growing competition, Ideogram distinguishes itself with curated image styles and Magic Prompt, aiming to enhance user experience. However, similarities with OpenAI's DALL-E 3 raise questions about originality. While promising for AI art enthusiasts and businesses, the rise of Ideogram 1.0 may also exacerbate the issue of spammy AI-generated content online.

Morph Studio lets you make flims using Stability AI-generated clips

Morph Studio rolls out an AI filmmaking platform, allowing users to craft movies using Stability AI-generated clips. The tool, resembling a storyboard, lets users input text prompts for different scenes, merging them into a cohesive narrative. With a focus on community building, users can share and modify production workflows within Morph's creator community. 

Founded by computer vision Ph.D. dropouts, Morph aims to distinguish itself from competitors like CapCut by fostering a vibrant user base. Initially focused on AI-generated memes, Morph now prioritizes enhancing its film production platform to meet user demands for seamless content creation.

Klarna says its AI assistant does the work of 700 people after it laid off 700 people

Klarna, the Swedish fintech company, lauds its AI assistant, claiming it does the workload equivalent to 700 people, following a layoff of the same number of employees. The AI-powered virtual assistant handles customer service chats adeptly, boasting high satisfaction ratings. Klarna estimates the chatbot could boost profits by $40 million in 2024. 

Despite past layoffs and CEO Sebastian Siemiatkowski's controversial handling of the situation, Klarna emphasizes that the AI's productivity isn't connected to the job cuts. The company underscores the importance of transparency in addressing the transformative impact of AI on the workforce and society.

🧠RESEARCH

Recent research, like BitNet, introduces 1-bit Large Language Models (LLMs). BitNet b1.58, a variant, employs ternary {-1, 0, 1} parameters, matching full-precision models in performance but with enhanced cost-effectiveness in latency, memory, throughput, and energy. This 1.58-bit LLM sets a new standard for efficient and high-performance language modeling, with implications for hardware design.

"EMO: Emote Portrait Alive" tackles the challenge of lifelike talking head video generation by focusing on the intricate relationship between audio cues and facial expressions. The EMO framework directly synthesizes audio to video, ensuring seamless transitions and consistent identity preservation, resulting in expressive and realistic animations, surpassing existing methods.

"Sora: A Review" examines OpenAI's text-to-video AI model, Sora, released in Feb 2024. It synthesizes realistic or imaginative videos from text inputs, showcasing potential in simulating the physical world. The paper discusses Sora's development, underlying technologies, applications across industries, challenges like safe and unbiased video generation, and future directions for enhancing human-AI interaction and creativity in video generation.

"DiffuseKronA" presents a novel fine-tuning method for personalized diffusion models in text-to-image generation. Addressing limitations of previous methods like DreamBooth, it introduces a Kronecker product-based adaptation module, significantly reducing parameters while improving synthesis quality. It mitigates hyperparameter sensitivity, enhancing image diversity and fidelity without extensive fine-tuning. The approach marks a substantial advancement in T2I generative modeling efficiency and performance.

"OmniACT" introduces a pioneering dataset and benchmark for assessing virtual agents' capacity to autonomously execute computer tasks across desktop and web applications. It aims to streamline human-computer interaction by generating executable programs from visual and natural language inputs. Despite strong baseline performance, the task remains challenging, highlighting the need for future multimodal model development bridging language and visual understanding.

"When Scaling Meets LLM Finetuning" investigates the impact of scaling factors on finetuning large language models (LLMs) for downstream tasks. Through systematic experiments, the study examines how LLM model size, pretraining data size, finetuning parameter size, and finetuning data size influence performance. Results reveal scaling behaviors and highlight task- and data-dependent optimal finetuning methods.

🛠️TOP TOOLS

Aspen - free API-testing tool designed specifically for macOS, offering a native app experience without the need for user login.

PDFtoChat - Chat with your PDFs in seconds

Talently - AI-based interviewing platform that automates live, conversational interviews and real-time evaluations, streamlining the recruitment process by providing objective candidate scoring and role-specific interviewing, including live coding assessments.

TextReader - free text-to-speech tool with realistic AI voices, transforming written text into lifelike speech quickly. 

Grantable -  AI-powered grant writing assistant that simplifies and streamlines the grant writing process, helping users craft proposals efficiently and effectively. 

Fireflies AI - AI-based tool that records, transcribes, and summarizes meetings. It integrates with various meeting platforms, offers AI-powered search, facilitates collaboration, and provides analysis on conversation intelligence.

📲SOCIAL MEDIA

🗞️MORE NEWS

Apple will ‘break new ground’ in generative AI this year, Tim Cook teases

Apple CEO Tim Cook hints at groundbreaking developments in generative AI this year, spurring anticipation among investors. With iOS 18's focus on AI features, Cook's remarks signal Apple's commitment to advancing in the AI realm, potentially rivaling competitors like OpenAI and Google. Exciting prospects lie ahead for the tech giant. 9TO5MAC

Google CEO calls AI tool’s controversial responses ‘completely unacceptable’

Google CEO Sundar Pichai condemned the controversial responses from the AI tool Gemini as "completely unacceptable." The tool, suspended last week, produced offensive and biased results, prompting Pichai to pledge structural changes. Criticisms arose over its depiction of race and equating Elon Musk with Adolf Hitler. Pichai acknowledged the errors and vowed to rectify them. SEMAFLOR

Samsung’s new microSD cards could make on-device AI easier

Samsung introduces faster microSD cards, enabling on-device AI for data-heavy apps. Sampling its 256GB SD Express card and mass-producing the 1TB version, Samsung boasts speeds surpassing traditional cards and SSDs. With demand for on-device AI rising, these cards could expand phone memory for intensive tasks without cloud reliance. THE VERGE

WordPress says VIP customer data won’t be shared to AI firms without consent

WordPress assures VIP customers that their data won't be shared with AI firms like OpenAI without explicit consent. Amid concerns over data sharing, CEO Nick Gernert emphasizes privacy protection. However, questions remain about whether this policy extends to all WordPress users, given its extensive reach across the internet. VENTUREBEAT

Amazon’s Road House reboot is accused of copyright infringement — and AI voice cloning

A lawsuit accuses Amazon Studios and MGM of copyright infringement over the Road House reboot. Screenwriter R. Lance Hill alleges that Amazon used generative AI to clone actors' voices during Hollywood strikes. Amazon denies using AI, stating filmmakers controlled editing. The lawsuit seeks to block the film's release. THE VERGE

SambaNova debuts 1 trillion parameter Composition of Experts model for enterprise gen AI

SambaNova, an AI startup with $1.1B funding, challenges OpenAI with Samba-1, a bundle of 56 generative AI models for enterprise tasks like text rewriting. CEO Rodrigo Liang touts its modularity and cost-effectiveness, claiming it offers tailored solutions. Its multi-model approach allows for better control and potentially more reliable responses, appealing to enterprises seeking full-stack AI solutions. VENTUREBEAT

BrainChip showcases AI-enabled human behavioral analysis with Akida neuromorphic computing

BrainChip, collaborating with NVISO Group Ltd., introduces AI-driven human behavioral analysis using Akida neuromorphic computing at CES 2024. This real-time system merges BrainChip's Akida IP and processors with NVISO's behavior analysis software, promising applications in consumer electronics, surveillance, and automotive sectors. BrainChip also relaunches its Akida Edge AI Box for pre-orders, aiming to deliver intelligent, secure devices for various industries. BIOMETRIC UPDATE

New AI model could streamline operations in a robotic warehouse

MIT researchers use AI inspired by traffic congestion solutions to optimize robotic warehouse operations. Facing challenges akin to navigating a bustling city center, they devised a deep-learning model dividing robots into groups for efficient coordination, resulting in up to four times faster decongestion. The approach has broader applications beyond warehouses. MIT

What'd you think of today's edition?

Login or Subscribe to participate in polls.

What are MOST interested in learning about AI?

What stories or resources will be most interesting for you to hear about?

Login or Subscribe to participate in polls.

Join the conversation

or to participate.