- Superhuman AI
- Posts
- Android's AI upgrade
Read time: under 4 minutes
Welcome back, Superhuman
A new model from Sakana Labs is holding up a mirror and asking: “Who am I, really?” The Tokyo-based startup’s AI Scientist appears to be the first that can create AI-related experiments, then carry them out entirely on its own, uncovering new insights about LLMs along the way.
Today’s Insights
All the AI-related announcements from Tuesday’s Pixel event
Tutorial: Generating mind maps with AI
The first-ever autonomous AI scientist
Everything else you should know today
5 new AI tools to boost your productivity
AI-Generated Images: Dark spaces
NEXT IN AI
Inside this week’s AI-focused Pixel event
Source: Engadget
The world’s go-to search giant had a tough few months. It scaled back its AI Overviews feature after users complained it was generating inaccurate results. And it faced criticism over a Gemini ad that some viewers found insensitive.
Its first act of redemption: At Tuesday’s Pixel event, it announced a suite of new AI features for Androids, including one that could give OpenAI’s lauded voice mode a run for its money.
What can it do? It’ll be able to reference and analyze information across different apps, including email, Maps, and Calendar. It can automatically organize your screenshots to help you easily resurface information like lockbox codes or passwords. You can call up Gemini while watching a food-centric YouTube video, then ask it to create a shopping list based on the ingredients shown in the clip.
Gemini Live, meanwhile, appears to be very similar to OpenAI’s voice mode. It’s supposedly best for prepping for tricky conversations and brainstorming new ideas. With 10 voices to choose from, the tool will be rolling out to Gemini Advanced subscribers over the next few weeks.
Here’s a look at all of the other major AI announcements:
The latest Tensor G4 chip promises more efficiency for AI-related tasks across photo, video, and streaming
Gemini Nano can run AI features — like text and phone call summaries — on device, ensuring a deeper level of security
A new Add Me feature lets you add yourself to group photos with the help of augmented reality
Pixel Studio lets you generate images, change their style, and incorporate your own photos — helping you do things like create a party invitation in seconds
Taking a cue from Apple’s playbook, the latest Gemini tools are all about practicality and accessibility — letting anyone create new things without needing any special technical knowledge. Looks like Apple has some catching up to do when it finally unveils more Apple Intelligence features in early September.
PRESENTED BY ASSEMBLY AI
Industry-leading accuracy for voice-driven AI products
Build your AI product with the best–AssemblyAI’s industry‑leading Speech-to-Text models. This powerful API offers:
Industry-leading accuracy (92.5%) and low latency transcriptions
Speech AI capabilities for transcribing and understanding 99+ languages, plus accents and background noise
Seamless API integration for rapid deployment and scalability
Start building today. Get a $50 credit (no cc required).
THE AI ACADEMY
How to turn anything into mind maps using AI
Go to Mapify’s website and sign up to get credits.
Insert your data source (PDF, YouTube Video, Image, Audio, etc) and click on the Mapify button.
Wait for a few seconds and that's it!
You can even prompt it and use given templates to visualize data as per your need.
Example prompt: SWOT analysis for [insert your business idea]
You can use it to brainstorm ideas, make outlines, plan projects, develop strategies, explain concepts, analyze topics, create timelines, and more.
AI & SCIENCE
Sakana builds world’s first AI for automated scientific research
Source: Sakana AI
Four new AI-related research papers just dropped. But what’s just as notable as the research itself is the author behind each of them: Not a distinguished professor of computer science but an LLM designed by Tokyo’s Sakana AI.
Sakana, sometimes described as Japan’s answer to OpenAI, claims to have created the first AI Scientist — a system that can design experiments, carry them out, and write them up totally on its own.
How it works:
One LLM explores various computer science topics, generates interesting ideas, tests them, and finally describes the results with visualizations and text
Another evaluates each experiment and gives feedback “with near-human accuracy” — which, in turn, helps the AI Scientist make better decisions in the future
The model has already generated studies related to diffusion models, algorithms, and the architecture of neural networks
Each study costs about $15 of compute, offering a far more affordable alternative to traditional experiments
Why it matters: Several LLMs are already being used to assist real-life scientists with research — from automating lengthy studies to analyzing datasets to discovering new drugs. But Sakana’s new system appears to be the first that can perform each step without human intervention.
PRESENTED BY RESPONSE WHISPERER
Make selling easy with real-time AI coaching
Response Whisperer is your AI-powered sales coach, offering real-time, data-driven advice during calls to help you win more clients. Ideal for sales teams, small business owners, and trainers, Response Whisperer makes selling easy, reduces training time, and boosts your effectiveness.
AI & TECH NEWS
Everything else you need to know today
Source: Nima Owji/X
Out of the Bag: Supposed screenshots of xAI’s next model, Grok 2.0, have begun circulating on X. It’ll allegedly feature Black Forest Labs’ new text-to-image model Flux as well as new writing and coding capabilities.
Opera’s Next Act: The minimalist search engine Opera One is launching on iOS with several AI features, including a built-in image generator and voice search.
Silicon Showdown: Despite US sanctions, Chinese tech giant Huawei will soon unveil a new AI chip that could challenge Nvidia’s market dominance.
Small but Mighty: Abu Dhabi’s Technology Innovation Institute has introduced Falcon Mamba 7B, a model built with a novel architecture that outperforms similar-sized peers.
💼 AI IRL: A Financial Times investigation found that about half of job applicants are using LLMs to help craft their cover letters and resumes. Recruiters say their pool of candidates is larger than ever now that anyone can generate an application in a matter of seconds. But it also means that those who use AI only sparingly may have a leg up.
🧠 Brain Food: A new computer model from researchers in Iraq and Australia can reportedly detect diseases like diabetes and asthma with 98% accuracy — using nothing more than a photo of someone’s tongue.
PRODUCTIVITY
5 AI Tools to Supercharge Your Productivity
✅ Upmetrics: An all-in-one business planning software that combines actionable planning, real-time forecasting, and insightful reporting.
✅ Jupitrr: Create engaging background footage for your marketing videos with AI.
✅ Profundo: An AI-powered research and reporting tool that helps you gather, analyze, and report data more efficiently.
✅ Recap: A personal newsletter of everything you’ve saved, created, or captured recently, sent directly to you.
✅ Midship: An AI that reads and understands your complex documents, extracting key information and organizing it into your preferred spreadsheet format.
PS: Want more? Check out our Top 100 AI Tools.
* indicates a promoted tool, if any
PROMPT OF THE DAY
UX Help
Prompt: As a designer, what are some gamification techniques I can add to a [insert project/product you are working on here, like language learning app]You can adapt the prompt to your specific needs.
Source: WriteSonic
AI-GENERATED IMAGES
Dark Spaces
Source: @sweet_11765 on Midjourney
Midjourney Prompt: A dark gray [insert area here, like museum hall or hotel lobby] with [sharp/curved] angles and angular shapes, a black wall on the right side of the room is covered in triangular slabs that form an interesting structure, a door leading to another space stands at its center. The interior design reflects modern minimalism, with soft lighting creating a dramatic effect.
--ar 3:4 --v 6.1 --stylize 250Acquire new customers and drive revenue by partnering with us
Superhuman is the world’s biggest AI newsletter for businesses and professionals with 600,000+ readers working at the world’s leading startups and enterprises. Companies like Amazon, Hubspot, and Salesforce feature their products in Superhuman. You can learn more about partnering with us here.