AlphaSignal

Hey ,

Welcome to this week's edition of AlphaSignal the newsletter for AI experts. Whether you are a researcher, engineer, developer, or data scientist, our summaries ensure you're always up-to-date with the latest breakthroughs in AI.

Let's get into it!

Lior

📑 ANNOUNCEMENTS

1. Anthropic announces 100k context window (75,000 words)
Anthropic's Claude model now boasts a massive 100k context window, surpassing MosaicML's 65k window models. It can process entire documents, making it a powerful tool for information retrieval and analysis.

2. OpenAI launches ChatGPT app for iOS
OpenAI has launched the ChatGPT app for iOS, enabling users to access ChatGPT on the go. The app syncs chat history, integrates speech recognition, and offers exclusive features for ChatGPT Plus subscribers. It allows instant answers, tailored advice, creative inspiration, professional input, and learning opportunities. The rollout will expand globally, with an upcoming release for Android users.

3. Hugging Face releases StarCoder, a State-of-the-Art LLM for Code
AI startup Hugging Face and ServiceNow Research have launched StarCoder, a free code-generating AI system. Similar to GitHub's Copilot, code-generating systems have the potential to significantly reduce development costs and allow programmers to focus on creative tasks. StarCoder, trained on over 80 programming languages and GitHub repositories, offers royalty-free use and integrates with Microsoft's Visual Studio Code.

4. Google releases MusicLM, a new experimental AI tool that can turn text descriptions into music.
Users can specify instruments like “electronic” or “classical,” as well as the “vibe, mood, or emotion” they’re aiming for, as they refine their MusicLM-generated creations.

5. Stability AI releases Stable Animation SDK, a powerful text-to-animation tool for developers
The SDK offers three methods: text-to-animation, text plus image input, and video plus text input. Users can customize parameters and prompts, and the SDK provides various settings for frame rate, resolution, and length. Installation is simple via the Stability AI website.

^NEWS
Google Announces PaLM 2: A Foundational Language Model

At its I/O developer conference, Google announced the launch of PaLM 2, its newest large language model (LLM). PaLM 2 will power Google’s updated Bard chat tool, the company’s competitor to OpenAI’s ChatGPT, and function as the foundation model for most of the new AI features.

PaLM 2 standout enhancement lies in its augmented logical reasoning capabilities. By undergoing rigorous training on mathematical and scientific texts, including complex equations, PaLM 2 showcases improved proficiency in common-sense reasoning and problem-solving tasks. This achievement is particularly noteworthy considering the historical difficulties language models have faced when grappling with mathematical inquiries. PaLM 2 also exhibits enhanced support for coding tasks, having been trained on an assortment of programming languages. This renders it a valuable asset for software development endeavors.

Trained on a corpus spanning over 100 languages, it excels in multilingual tasks and offers nuanced phrasing, surpassing previous iterations. PaLM 2 is now available to developers through Google’s PaLM API, Firebase and on Colab.

Our Take
Just one week following the leak of an internal document at Google, which acknowledged the absence of a distinct advantage, the company swiftly responded by unveiling PaLM 2. This assertive move indicates Google's ability to effectively contend with major competitors like Microsoft and OpenAI. The introduction of PaLM 2 is a momentous development and excellent news for businesses, as it holds the promise of expanding market opportunities in the AI industry. Google has also alluded to their upcoming model, Gemini, designed to exhibit exceptional efficiency and accommodate various modes of input.

From my perspective, this further reinforces the ongoing trend of creating multimodal solutions capable of processing visual cues. Additionally, Google's launch of Gecko, a compact model tailored for mobile devices, clearly demonstrates their focus on the mobile market. I perceive this strategic maneuver as a means for Google to gain a competitive advantage over OpenAI, infusing a fresh wave of excitement into the unfolding AI race.

Want to promote a product, job, or event to 100,000+ AI researchers and engineers? You can reach us to us here.

^NEWS
OpenAI CEO, Sam Altman, Calls for Government Regulation of AI

During his testimony before a US Senate Committee, Sam Altman urged lawmakers to establish regulations for the rapidly advancing field of AI. While acknowledging the transformative potential of AI, Altman also recognized its possible negative impacts, such as job displacement and the spread of misinformation during elections. He proposed the creation of a dedicated agency to oversee and license AI companies and advocated for a comprehensive regulatory framework that includes licensing, stringent testing requirements, and independent audits.

Altman's testimony was well received by several senators, leading to bipartisan support for regulatory measures. However, some legislators expressed concern about the feasibility of regulating AI given its swift advancements. Despite these challenges, Altman expressed optimism about the emergence of new, rewarding jobs in the future. Overall, Altman's testimony highlighted the need for effective and timely regulation to ensure the responsible and beneficial development of AI.

Our Take
In my opinion, the recent advancements in AI could have significant implications for our society. I believe this field is progressing rapidly and we have yet to fully grasp how it may impact various aspects of our lives. Large language models that behave like humans could prove dangerous if they reflect and amplify the biases we have.

Yet, one cannot ignore the ongoing power dynamics within the industry. OpenAI's call for regulation may be seen as a strategic move to maintain its dominance.

🤖 TOP OF REDDIT

Drag Your GAN: Interactive Point-based Manipulation on the Generative Image Manifold
↑ 1557

AI is going to put a lot of call centers in India and elsewhere out of business.
↑ 1210

Does anybody else despise OpenAI?
↑ 1190

“Yeah so the Singularity is gonna happen soon. We’ll probably all end up on UBI. That’s if we survive though.”
↑ 1181

OpenAI CEO asking for government's license for building AI
↑ 1120

^NEWS
Meta open-sources multisensory AI model that combines six types of data

Meta's recent open-source release of ImageBind, an AI model that combines multiple data modalities, has piqued the interest of researchers and practitioners in computer vision and AI. ImageBind is designed to generate immersive and multisensory experiences by linking text, audio, visual, depth, temperature, and motion data. What sets ImageBind apart is its ability to create a shared embedding space, eliminating the need for extensive training on all possible combinations of modalities. Leveraging large-scale vision-language models, ImageBind utilizes the binding property of images to establish connections between various modalities. This breakthrough approach enables cross-modal retrieval, semantic composition, and audio-to-image generation.

Our Take
In light of Google and OpenAI's decision to hold back on sharing details of their most recent research breakthroughs, it's important to see that Meta continues to foster open-source. This transparency not only invites the community to build upon these advancements but also stimulates a culture of collaboration within AI research.

ImageBind's introduction represents a significant milestone in multimodal learning, with far-reaching implications for the AI landscape. By bridging the gap between human perception and machine learning, ImageBind showcases the potential for creating fully realized environments from limited input. The impact on fields such as virtual reality, mixed reality, and accessibility solutions is evident. As Meta continues to pursue its metaverse vision, ImageBind lays the foundation for generating immersive content and empowering creators. With its ability to bind diverse modalities, ImageBind augments the understanding of data, fostering richer experiences and advancing the frontiers of AI.

Thank You