👣 Carbon footprint

🗞️ Your Weekly Round-Up

1️⃣ Google Gemini

Google is looking to get on level terms with OpenAI by updating Gemini, a natively multimodal AI that simultaneously tackles text, images, and video. 

Here are the details: 

• It is being rolled out in Bard.

• Gemini comes in three versions—Ultra, Nano, and Pro. The latter powers Bard, enhancing its reasoning capabilities.

• The model outperforms ChatGPT's earlier version on critical benchmarks, with plans for integration into search, ads, and Chrome.

Plus, their demo video was one of the craziest things I’ve ever seen! Check it out here. 

Read more: Wired

2️⃣ Google’s latest Cloud TPU

Along with the launch of Gemini, Google also released its newest lightning-fast Cloud TPU v5p. 

For you tech junkies out there, the v5p has 8,960 chips per v5p pod, backed by Google's speediest interconnect at 4,800 Gbps per chip. These chips also promise 2x FLOPS improvement and 3x high-bandwidth memory boost over their predecessor, the v4 TPUs.

Google claims the TPU v5p can train models like the 175B parameter GPT3 2.8X faster than the v4, delivering cost-effective efficiency. 

Read more: TechCrunch

3️⃣ Meta’s speech translation tool

Meta has updated its AI translation model SeamlessM4T v2, taking conversational AI to new heights. 

Some of the new features include: 

• “SeamlessExpressive,” a tool that captures users’ unique expressions - pitch, volume, emotional tones - in translated speech, breaking free from the robotic monotony of traditional translations. 

• “SeamlessStreaming,” a tool that reduces translation lag to under two seconds.

This “Seamless Communication” suite outshines competitors like Google and Samsung's mobile interpreters, with potential integration opportunities into Meta's smart glasses.

Read more: Engadget

4️⃣ Sequoia’s latest AI tool

VC giant Sequoia Capital just launched Visual Electric, an AI-based image generation tool for designers. 

How does it work? Designers can type in simple prompts to get started, letting them “spread out, make a mess, and work spatially.” They can also iterate freely, selecting color palettes, choosing formats, and setting moods from Airbrush to Neon. 

Visual Electric's features include a “Touch Up” option for modification, upscaling for premium users, background removal, “creativity” sliders for variations, and a “remix” feature for a fresh take on existing images.

The free-to-use tool offers 40 daily image generations, while the premium plan at $16 per month provides unlimited creations, faster generation speeds, and commercial usage licenses.

Read more: TechCrunch

5️⃣ MS Copilot gets an upgrade

Microsoft is ready to release a host of updates to its Copilot service. Here’s the TL;DR:

• Copilot will soon harness GPT-4 and its 128K context window for query understanding and supercharged responses.

• Copilot users will also get the upgraded DALL-E 3 model, promising higher quality and more accurate image creation. 

• Its latest code interpreter will get the biggest upgrade, allowing users to perform accurate calculations, data analysis, and write code in response to their natural-language requests.

Read more: The Verge

6️⃣ New AI chips for faster AI training

AMD recently launched a bunch of new chips and accelerators to power AI training:

• MI300X Accelerator: With 1.5X more memory than its predecessor, the MI300X is hailed as the world's highest-performing accelerator, rivaling Nvidia's H100 in training LLMs while outshining in inference tasks. AMD plans to place MI300X in Azure virtual machines, which can only mean one thing - an incoming move to cloud computing. 

• MI300A APU: The MI300A APU, a powerhouse in data centers, comes with higher-performance computing, faster model training, and a 30X energy efficiency improvement. Meta is onboard and set to deploy MI300 processors in its data centers.

• Ryzen 8040: AMD is also launching Ryzen 8040, the latest in its processor family. With 1.6X more AI processing power, integrated neural processing units (NPUs), and video editing and gaming performance, the Ryzen 8040 aims to dramatically improve mobile devices.

Read more: The Verge

7️⃣ Zillow’s new AI-powered search

Zillow is leveling up the house-hunting game with its AI-powered natural-language search. 

Soon, users can search for their dream home just like chatting with a friend, with Jenny Arden, Zillow's chief design officer, hailing the feature as a “game changer” in the home-shopping journey. 

How does it work? The AI-powered search dives into the intricacies of users' queries, scanning millions of listing details and training machine learning models to understand natural, human-like sentences.

Read more: Zillow Group Blog

8️⃣ AI’s fight against cancer

AstraZeneca has sealed a deal worth up to $247 million with U.S. AI biologics powerhouse Absci. 

Their mission? To unleash the power of generative AI in the fight against cancer, helping craft enhanced antibody therapeutics and revolutionizing oncology treatments. Their approach will leverage AI to design optimal drug candidates focusing on target affinity, safety, manufacturability, and more.

The Financial Times reported that the deal encompasses an upfront fee, substantial R&D funding, milestone payments, and royalties from product sales.

Read more: Yahoo Finance

9️⃣ AI’s carbon footprint

In the race for sustainability, AI is facing a carbon conundrum. 

A recent study by Hugging Face and Carnegie Mellon University revealed that generating AI images comes with a hefty carbon footprint—equivalent to charging a smartphone.

On the flip side, AI-generated text sprints ahead with remarkable energy efficiency, requiring only 16% of the energy needed to charge a smartphone to full capacity.

The research delves into 13 diverse tasks, measuring carbon emissions per 1000 grams for tasks ranging from image generation to text classification. 

The result? Tasks demanding AI to create new content, be it text or images, top the energy-intensive charts. While text classification emerges as the more sustainable option, image generation takes the lead in emissions.

Read more: Engadget

🔟 The AI Alliance

Meta and IBM have teamed up to create the AI Alliance—a powerhouse of 50 companies, startups, universities, and government bodies with a shared mission to drive open-sourced, responsible innovation in AI. 

Taking on giants like OpenAI and Google, the Alliance, including Hugging Face, Intel, and AMD, forms a network of independent partners rather than a closed-source nonprofit.

Their aim? To shape the future of generative AI through collaboration and open ecosystems, addressing concerns about stability and closed-source models. 

Their focus? Advancing open foundation models, developing evaluation tools, boosting AI hardware accelerators, and educating the public about AI risks and benefits.

Read more: Gizmodo

🎙️ New Episode ft. Fergal Reid, VP of AI at Intercom

Hey team! ICYMI, our latest episode featuring Fergal Reid, VP of AI at Intercom, is out now!

Catch this episode on YouTube, Spotify, or Apple Podcasts today!

🤑 AI Fundraising News

• Rohirrim raises $15M in Series A funding to help teams unlock value from proprietary data by securely ingesting and organizing information in presentations, spreadsheets, proposals, whitepapers, and emails.

• Seismic Therapeutic raises $121M in Series B funding to create a platform that uses machine learning to discover and develop immunology therapies.

• ArmorCode raises $40M in Series B funding to help companies ship secure software.

• AssemblyAI raises $50M in Series C funding to offer AI models optimized for audio file processing and analysis, such as transcribing speech, generating automated summaries, and detecting sensitive data.

• AutogenAI raises $39.5M in Series B funding to develop its bid-writing product suite, which uses AI technology to help organizations write bids, tenders, and proposals more efficiently and effectively.

• Vast Data raises $118M in Series E funding to provide a platform that unifies storage, database, and containerized compute engine services into a single, scalable software platform architected from the ground up to power AI and GPU-accelerated tools in modern data centers and clouds.

• Opal Security raises $22M in Series B funding to help companies manage access and identities.

• Sarvam raises $41M in Series A funding to build large language models targeted at unique uses in Indian languages, making apps affordable for the country's 1.4 billion people.

🛠️ AI Tools to Check Out

• Animaker AI: An AI tool that generates lightning-fast animations with unparalleled accuracy and precision. Check them out!

• Fathom: Discover podcasts at the speed of thought with mind-blowing AI-powered search, transcripts, chapters, clipping, and highlights. Check them out!

🐦 Tweet Post of the Day

AI is getting better and better at research every day 🔍️