🖌️ Figma AI-Powered Design Tools

PLUS: Sohu Replaces 160 Nvidia GPUs

Welcome back!

Figma released a suite of AI tools aimed at boosting UX design productivity. These tools include AI-generated design drafts and visual search. This could revolutionize how UX designers work, making tasks more efficient and accessible. Let's explore...

Today’s Summary:

  • Figma new AI design tools

  • Sohu replaces 160 Nvidia GPUs

  • AI model simulates 500 M years of evolution

  • YouTube AI music deal talks

  • Mitigating “Skeleton Key” AI jailbreak

  • 3 new tools


Figma Unveils AI Tools to Boost Design Productivity

The Summary: Figma announced a suite of new AI-powered features at its Config conference. These tools aim to automate tedious tasks and speed up UX design workflows.

Key additions include AI-generated design drafts, visual search, automatic layer naming, and AI-assisted prototyping. Figma emphasized its careful approach to data privacy and AI training.

Source: Figma

Key details:

  • AI can generate initial app designs from text prompts

  • Visual search finds designs across files using images or text

  • Automatic contextual layer naming with one click

  • AI builds prototypes by connecting design pages automatically

  • Realistic placeholder text and images generated by AI

  • Opt-in required for using customer data to train AI models

  • Features free during beta, pricing TBA for full launch.

  • To get access, click the “?” button at the bottom of the screen in your Figma account, then select “Join UI3 + AI waitlist

Source: Figma

Why it matters: These Figma AI tools could significantly boost designer productivity by automating routine tasks. The features may lower barriers for non-designers to create basic mockups. Figma's approach to data privacy and consent for AI training sets an example for responsible AI development in creative tools.


Sohu Replaces 160 Nvidia GPUs, Delivers 500,000 Tokens per Second

The Summary: Etched is working on a specialized AI chip called Sohu. Unlike general-purpose GPUs, Sohu is designed to run only transformer models, the architecture behind LLMs like ChatGPT.

The company claims Sohu will offer dramatically better performance than traditional GPUs while using less energy. This approach could improve AI infrastructure as the industry grapples with increasing power consumption and costs.

Source: Etched

Key details:

  • Etched raised $120 million Series A funding to work on Sohu

  • Sohu will be manufactured using TSMC 4nm process

  • Can deliver 500,000 tokens per second for Llama 70B

  • One Sohu server allegedly replaces 160 H100 GPUs

  • Etched claims Sohu is 10x faster and cheaper than Nvidia's next-gen Blackwell GPUs

Why it matters: The AI industry faces growing concerns about the costs of running large models. If Etched's claims hold true, Sohu could offer an efficient alternative to current general purpose GPU hardware, potentially accelerating AI development while reducing energy requirements. However, the company’s narrow focus on transformers is a risky bet, as future breakthroughs might favor different architectures.


AI Model Generates Novel Proteins, Simulating 500 Million Years of Evolution

The Summary: Researchers at EvolutionaryScale, an AI lab founded by former Meta engineers, have developed ESM3, a language model that can generate functional proteins. ESM3 created a novel green fluorescent protein (GFP) named esmGFP, which is highly different from existing GFPs.

The innovation achieved by ESM3 is estimated to be equivalent to over 500 million years of natural evolution. This breakthrough shows the potential of AI to accelerate protein engineering and explore new areas of protein design.

Key details:

  • ESM3 is a multimodal AI model that reasons over protein sequence, structure, and function

  • Trained on 2.78 billion natural proteins using 1 trillion teraflops of computing power

  • Generated esmGFP by following prompts

  • esmGFP has 96 mutations compared to its closest known relative

  • Researchers estimate esmGFP represents over 500 million years of evolutionary distance from known proteins

  • The smaller model has been open-sourced on GitHub

Why it matters: This breakthrough shows how AI can explore protein design in ways traditional methods cannot. By creating a functional protein with such novelty, ESM3 demonstrates the potential to accelerate discoveries in biotechnology, medicine, and materials science. The ability to design proteins with specific functions could lead to new therapeutics, industrial enzymes, and biomaterials. This also provides new insights into protein evolution and the vast space of possible proteins that nature hasn't yet explored.


Quick news


🥇 New tools

  • Figma AI - AI powered UX design, free during beta

  • June - Free and simple customer analytics with AI

  • The Fittest - Gamify your fitness journey with AI

That’s all for today!

If you liked the newsletter, share it with your friends and colleagues by sending them this link: https://thesummary.ai/